diff --git a/common/CMakeLists.txt b/common/CMakeLists.txt index 789154e83..c5c450e24 100644 --- a/common/CMakeLists.txt +++ b/common/CMakeLists.txt @@ -52,6 +52,7 @@ set(TARGET common) add_library(${TARGET} STATIC base64.hpp + chat-template.hpp common.h common.cpp chat.h @@ -72,6 +73,7 @@ add_library(${TARGET} STATIC json-schema-to-grammar.cpp train.h train.cpp + minja.hpp ngram-cache.h ngram-cache.cpp ) diff --git a/common/chat-template.hpp b/common/chat-template.hpp new file mode 100644 index 000000000..b4a90145c --- /dev/null +++ b/common/chat-template.hpp @@ -0,0 +1,249 @@ +/* + Copyright 2024 Google LLC + + Use of this source code is governed by an MIT-style + license that can be found in the LICENSE file or at + https://opensource.org/licenses/MIT. +*/ +// SPDX-License-Identifier: MIT +#pragma once + +#include "minja.hpp" +#include +#include +#include + +using json = nlohmann::ordered_json; + +namespace minja { + +class chat_template { + public: + + private: + bool supports_tools_ = true; + // Meta-Llama-3.1-8B-Instruct's template expects arguments to be an object. + // Most other templates (and OpenAI's API) expect the arguments object to be stringified. + bool requires_object_arguments_ = false; + bool supports_system_role_ = true; + bool supports_parallel_tool_calls_ = false; + std::string source_; + std::string bos_token_; + std::string eos_token_; + std::shared_ptr template_root_; + + std::string try_render( + const nlohmann::ordered_json & messages, + const nlohmann::ordered_json & tools, + bool add_generation_prompt, + const nlohmann::ordered_json & extra_context = nlohmann::ordered_json()) const + { + try { + auto prompt = apply(messages, tools, add_generation_prompt, extra_context); + // fprintf(stderr, "Prompt: %s\n", prompt.c_str()); + return prompt; + } catch (const std::exception & e) { + // fprintf(stderr, "Error: %s\n", e.what()); + return ""; + } + } + + public: + chat_template(const std::string & source, const std::string & bos_token, const std::string & eos_token) + : source_(source), bos_token_(bos_token), eos_token_(eos_token) + { + template_root_ = minja::Parser::parse(source_, { + /* .trim_blocks = */ true, + /* .lstrip_blocks = */ true, + /* .keep_trailing_newline = */ false, + }); + supports_tools_ = source.find("tools") != std::string::npos; + + auto renders_string_arguments = + try_render({ + { + {"role", "user"}, + {"content", "Hey"} + }, + { + {"role", "assistant"}, + {"tool_calls", json::array({ + { + {"id", "call_1___"}, + {"type", "function"}, + {"function", { + {"arguments", "{\"code\": \"print('Hello, World!')\"}"}, + {"name", "ipython"}, + }}, + }, + })}, + } + }, {}, false).find("{\"code\": \"print") != std::string::npos; + if (!renders_string_arguments) { + auto renders_object_arguments = + try_render({ + { + {"role", "user"}, + {"content", "Hey"} + }, + { + {"role", "assistant"}, + {"tool_calls", json::array({ + { + {"id", "call_1___"}, + {"type", "function"}, + {"function", { + {"arguments", { + {"code", "print('Hello, World!')"}, + }}, + {"name", "ipython"}, + }}, + }, + })}, + } + }, {}, false).find("{\"code\": \"print") != std::string::npos; + requires_object_arguments_ = renders_object_arguments; + } + supports_parallel_tool_calls_ = source.find("tool_call_id") != std::string::npos; + + supports_system_role_ = try_render({ + {{"role", "system"}, {"content", ""}}, + {{"role", "user"}, {"content", "Hey"}} + }, {}, false).find("") != std::string::npos; + } + + const std::string & source() const { return source_; } + const std::string & bos_token() const { return bos_token_; } + const std::string & eos_token() const { return eos_token_; } + bool supports_tools() const { return supports_tools_; } + bool supports_parallel_tool_calls() const { return supports_parallel_tool_calls_; } + + std::string apply( + const nlohmann::ordered_json & messages, + const nlohmann::ordered_json & tools, + bool add_generation_prompt, + const nlohmann::ordered_json & extra_context = nlohmann::ordered_json()) const + { + json actual_messages; + + // First, "fix" messages so they have a chance to be rendered correctly by the template + + if (requires_object_arguments_ || !supports_system_role_ || !supports_tools_) { + actual_messages = json::array(); + + std::string pending_system; + auto flush_sys = [&]() { + if (!pending_system.empty()) { + actual_messages.push_back({ + {"role", "user"}, + {"content", pending_system}, + }); + pending_system.clear(); + } + }; + for (const auto & message_ : messages) { + auto message = message_; + if (!message.contains("role") || !message.contains("content")) { + throw std::runtime_error("message must have 'role' and 'content' fields: " + message.dump()); + } + std::string role = message.at("role"); + + if (message.contains("tool_calls")) { + if (requires_object_arguments_ || !supports_tools_) { + for (auto & tool_call : message.at("tool_calls")) { + if (tool_call["type"] == "function") { + auto & function = tool_call.at("function"); + std::string arguments = function.at("arguments"); + function["arguments"] = json::parse(arguments); + } + } + } + if (!supports_tools_) { + auto content = message.at("content"); + auto tool_calls = json::array(); + for (const auto & tool_call : message.at("tool_calls")) { + if (tool_call.at("type") != "function") { + continue; + } + const auto & function = tool_call.at("function"); + auto tc = json { + {"name", function.at("name")}, + {"arguments", function.at("arguments")}, + }; + if (tool_call.contains("id")) { + tc["id"] = tool_call["id"]; + } + tool_calls.push_back(tc); + } + auto obj = json { + {"tool_calls", tool_calls}, + }; + if (!content.is_null() && content != "") { + obj["content"] = content; + } + message["content"] = obj.dump(2); + message.erase("tool_calls"); + } + } + if (!supports_tools_ && role == "tool") { + message["role"] = "user"; + auto obj = json { + {"tool_response", { + {"tool", message.at("name")}, + {"content", message.at("content")}, + }}, + }; + if (message.contains("tool_call_id")) { + obj["tool_response"]["tool_call_id"] = message.at("tool_call_id"); + } + message["content"] = obj.dump(2); + message.erase("name"); + } + + if (!message["content"].is_null() && !supports_system_role_) { + std::string content = message.at("content"); + if (role == "system") { + if (!pending_system.empty()) pending_system += "\n"; + pending_system += content; + continue; + } else { + if (role == "user") { + if (!pending_system.empty()) { + message["content"] = pending_system + (content.empty() ? "" : "\n" + content); + pending_system.clear(); + } + } else { + flush_sys(); + } + } + } + actual_messages.push_back(message); + } + flush_sys(); + } else { + actual_messages = messages; + } + + auto context = minja::Context::make(json({ + {"messages", actual_messages}, + {"add_generation_prompt", add_generation_prompt}, + {"bos_token", bos_token_}, + {"eos_token", eos_token_}, + })); + + if (!tools.is_null()) { + auto tools_val = minja::Value(tools); + context->set("tools", tools_val); + } + if (!extra_context.is_null()) { + for (auto & kv : extra_context.items()) { + minja::Value val(kv.value()); + context->set(kv.key(), val); + } + } + + return template_root_->render(context); + } +}; + +} // namespace minja diff --git a/common/common.cpp b/common/common.cpp index da702368d..ed50a0098 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -15,9 +15,11 @@ #include "json.hpp" #include "json-schema-to-grammar.h" #include "llama.h" +#include "chat-template.hpp" #include #include +#include #include #include #include @@ -199,6 +201,16 @@ int32_t cpu_get_num_math() { return cpu_get_num_physical_cores(); } + +static std::string read_file(const std::string& fname) { + std::ifstream file(fname); + if (!file) { + throw std::runtime_error(string_format("error: failed to open file '%s'\n", fname.c_str())); + } + std::string content((std::istreambuf_iterator(file)), std::istreambuf_iterator()); + file.close(); + return content; +} // // CLI argument parsing // @@ -278,6 +290,13 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) { params.tensor_buft_overrides.push_back({nullptr, nullptr}); } + if (!params.chat_template.empty() && !llama_chat_verify_template(nullptr, params.chat_template, params.use_jinja)) { + throw std::runtime_error(string_format( + "error: the supplied chat template is not supported: %s%s\n", + params.chat_template.c_str(), + params.use_jinja ? "" : "\nnote: llama.cpp was started without --jinja, we only support commonly used templates" + )); + } return true; } @@ -1425,7 +1444,7 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa } if (arg == "--chat-template") { CHECK_ARG - if (!llama_chat_verify_template(argv[i])) { + if (!llama_chat_verify_template(nullptr, argv[i], false)) { fprintf(stderr, "error: the supplied chat template is not supported: %s\n", argv[i]); fprintf(stderr, "note: llama.cpp does not use jinja parser, we only support commonly used templates\n"); invalid_param = true; @@ -1434,6 +1453,22 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa params.chat_template = argv[i]; return true; } + if (arg == "--chat-template-file") { + CHECK_ARG + std::string chat_template = read_file(std::string(argv[i])); + if (!llama_chat_verify_template(nullptr, chat_template, false)) { + fprintf(stderr, "error: the supplied chat template is not supported: %s\n", argv[i]); + fprintf(stderr, "note: llama.cpp does not use jinja parser, we only support commonly used templates\n"); + invalid_param = true; + return true; + } + params.chat_template = chat_template; + return true; + } + if (arg == "--jinja") { + params.use_jinja = true; + return true; + } if (arg == "--slot-prompt-similarity" || arg == "-sps") { CHECK_ARG params.slot_prompt_similarity = std::stof(argv[i]); @@ -1984,6 +2019,22 @@ std::string gpt_params_get_system_info(const gpt_params & params) { // String utils // +std::string string_format(const char* fmt, ...) { + va_list ap; + va_list ap2; + va_start(ap, fmt); + va_copy(ap2, ap); + int size = vsnprintf(NULL, 0, fmt, ap); + GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT + std::vector buf(size + 1); + int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2); + GGML_ASSERT(size2 == size); + va_end(ap2); + va_end(ap); + return std::string(buf.data(), size); +} + + std::vector string_split(std::string input, char separator) { std::vector parts; size_t separator_pos = input.find(separator); @@ -2985,6 +3036,22 @@ std::string llama_token_to_piece(const struct llama_context * ctx, llama_token t return piece; } +std::string llama_token_to_piece(const struct llama_model* model, llama_token token, bool special) { + std::string piece; + piece.resize(piece.capacity()); // using string internal cache, 15 bytes + '\n' + const int n_chars = llama_token_to_piece(model, token, &piece[0], piece.size(), 0, special); + if (n_chars < 0) { + piece.resize(-n_chars); + int check = llama_token_to_piece(model, token, &piece[0], piece.size(), 0, special); + GGML_ASSERT(check == -n_chars); + } + else { + piece.resize(n_chars); + } + + return piece; +} + std::string llama_detokenize(llama_context * ctx, const std::vector & tokens, bool special) { std::string text; text.resize(std::max(text.capacity(), tokens.size())); @@ -3011,50 +3078,60 @@ bool llama_should_add_bos_token(const llama_model * model) { // Chat template utils // -bool llama_chat_verify_template(const std::string & tmpl) { +bool llama_chat_verify_template(const struct llama_model* model, const std::string& tmpl, bool use_jinja) { + if (use_jinja) { + try { + auto chat_template = minja::chat_template(tmpl, "", ""); + chat_template.apply({ { + {"role", "user"}, + {"content", "test"}, + } }, json(), true); + return true; + } + catch (const std::exception& e) { + fprintf(stdout,"%s: failed to apply template: %s\n", __func__, e.what()); + return false; + } + } llama_chat_message chat[] = {{"user", "test"}}; - int res = llama_chat_apply_template(nullptr, tmpl.c_str(), chat, 1, true, nullptr, 0); + const int res = llama_chat_apply_template(model, tmpl.c_str(), chat, 1, true, nullptr, 0); return res >= 0; } std::string llama_chat_apply_template(const struct llama_model * model, - const std::string & tmpl, + const common_chat_template& tmpl, const std::vector & msgs, - bool add_ass) { + bool add_ass, + bool use_jinja) { + if (use_jinja) { + auto messages = json::array(); + for (const auto& msg : msgs) { + messages.push_back({ {"role", msg.role}, {"content", msg.content} }); + } + return tmpl.apply(messages, /* tools= */ json(), add_ass); + } int alloc_size = 0; - bool fallback = false; // indicate if we must fallback to default chatml std::vector chat; for (auto & msg : msgs) { chat.push_back({msg.role.c_str(), msg.content.c_str()}); alloc_size += (msg.role.size() + msg.content.size()) * 1.25; } - const char * ptr_tmpl = tmpl.empty() ? nullptr : tmpl.c_str(); std::vector buf(alloc_size); // run the first time to get the total output length - int32_t res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size()); - + int32_t res = llama_chat_apply_template(model, tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size()); // error: chat template is not supported if (res < 0) { - if (ptr_tmpl != nullptr) { // if the custom "tmpl" is not supported, we throw an error // this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template() throw std::runtime_error("this custom template is not supported"); - } else { - // If the built-in template is not supported, we default to chatml - res = llama_chat_apply_template(nullptr, "chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size()); - fallback = true; } - } // if it turns out that our buffer is too small, we resize it if ((size_t) res > buf.size()) { buf.resize(res); - res = llama_chat_apply_template( - fallback ? nullptr : model, - fallback ? "chatml" : ptr_tmpl, - chat.data(), chat.size(), add_ass, buf.data(), buf.size()); + res = llama_chat_apply_template(model, tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size()); } std::string formatted_chat(buf.data(), res); @@ -3062,12 +3139,13 @@ std::string llama_chat_apply_template(const struct llama_model * model, } std::string llama_chat_format_single(const struct llama_model * model, - const std::string & tmpl, + const common_chat_template& tmpl, const std::vector & past_msg, const llama_chat_msg & new_msg, - bool add_ass) { + bool add_ass, + bool use_jinja) { std::ostringstream ss; - auto fmt_past_msg = past_msg.empty() ? "" : llama_chat_apply_template(model, tmpl, past_msg, false); + auto fmt_past_msg = past_msg.empty() ? "" : llama_chat_apply_template(model, tmpl, past_msg, false, use_jinja); std::vector chat_new(past_msg); // if the past_msg ends with a newline, we must preserve it in the formatted version if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') { @@ -3075,21 +3153,77 @@ std::string llama_chat_format_single(const struct llama_model * model, }; // format chat with new_msg chat_new.push_back(new_msg); - auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass); + auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass, use_jinja); // get the diff part ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size()); return ss.str(); } -std::string llama_chat_format_example(const struct llama_model * model, - const std::string & tmpl) { +std::string llama_chat_format_example(const struct llama_model * model, const common_chat_template& tmpl, bool use_jinja) { std::vector msgs = { {"system", "You are a helpful assistant"}, {"user", "Hello"}, {"assistant", "Hi there"}, {"user", "How are you?"}, }; - return llama_chat_apply_template(model, tmpl, msgs, true); + return llama_chat_apply_template(model, tmpl, msgs, true, use_jinja); +} + + +common_chat_templates llama_chat_templates_from_model(const struct llama_model* model, const std::string& chat_template_override) +{ + auto vocab = llama_model_get_vocab(model); + std::string default_template_src = chat_template_override; + std::string template_tool_use_src = chat_template_override; + bool has_explicit_template = !chat_template_override.empty(); + if (chat_template_override.empty()) { + auto str = llama_model_chat_template(model, /* name */ nullptr); + if (str) { + default_template_src = str; + has_explicit_template = true; + } + str = llama_model_chat_template(model, /* name */ "tool_use"); + if (str) { + template_tool_use_src = str; + has_explicit_template = true; + } + } + if (default_template_src.empty() || default_template_src == "chatml") { + if (!template_tool_use_src.empty()) { + default_template_src = template_tool_use_src; + } + else { + default_template_src = R"( + {%- for message in messages -%} + {{- "<|im_start|>" + message.role + "\n" + message.content + "<|im_end|>\n" -}} + {%- endfor -%} + {%- if add_generation_prompt -%} + {{- "<|im_start|>assistant\n" -}} + {%- endif -%} + )"; + } + } + const auto get_token = [&](llama_token token, const char* name, const char* jinja_variable_name) { + if (token == LLAMA_TOKEN_NULL) { + if (default_template_src.find(jinja_variable_name) != std::string::npos + || template_tool_use_src.find(jinja_variable_name) != std::string::npos) { + fprintf(stdout, "%s: warning: vocab does not have a %s token, jinja template won't work as intended.\n", __func__, name); + } + return std::string(); + } + else { + return llama_token_to_piece(model, token, true); + } + }; + auto token_bos = get_token(llama_token_bos(model), "BOS", "bos_token"); + auto token_eos = get_token(llama_token_eos(model), "EOS", "eos_token"); + return { + has_explicit_template, + std::make_unique(default_template_src, token_bos, token_eos), + template_tool_use_src.empty() + ? nullptr + : std::make_unique(template_tool_use_src, token_bos, token_eos) + }; } // diff --git a/common/common.h b/common/common.h index 99048cd2a..4c62a53e2 100644 --- a/common/common.h +++ b/common/common.h @@ -228,6 +228,7 @@ struct gpt_params { std::string hostname = "127.0.0.1"; std::string public_path = ""; std::string chat_template = ""; + bool use_jinja = false; // NOLINT std::string system_prompt = ""; bool enable_chat_template = true; @@ -400,6 +401,11 @@ std::string llama_token_to_piece( llama_token token, bool special = true); +std::string llama_token_to_piece( + const struct llama_model* model, + llama_token token, + bool special = true); + // detokenizes a vector of tokens into a string // should work similar to Python's `tokenizer.decode` // optionally renders special/control tokens @@ -423,26 +429,45 @@ struct llama_chat_msg { }; // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid -bool llama_chat_verify_template(const std::string & tmpl); +bool llama_chat_verify_template(const struct llama_model* , const std::string& tmpl, bool use_jinja); + +namespace minja { + class chat_template; +} + +typedef minja::chat_template common_chat_template; + +struct common_chat_templates { + bool has_explicit_template; // Model had builtin template or template overridde was specified. + std::unique_ptr template_default; // always set (defaults to chatml) + std::unique_ptr template_tool_use; +}; + // CPP wrapper for llama_chat_apply_template // If the built-in template is not supported, we default to chatml // If the custom "tmpl" is not supported, we throw an error -std::string llama_chat_apply_template(const struct llama_model * model, - const std::string & tmpl, - const std::vector & chat, - bool add_ass); +std::string llama_chat_apply_template( + const struct llama_model* model, + const common_chat_template& tmpl, + const std::vector< llama_chat_msg>& chat, + bool add_ass, + bool use_jinja); // Format single message, while taking into account the position of that message in chat history -std::string llama_chat_format_single(const struct llama_model * model, - const std::string & tmpl, - const std::vector & past_msg, - const llama_chat_msg & new_msg, - bool add_ass); +std::string llama_chat_format_single(const struct llama_model* model, + const common_chat_template& tmpl, + const std::vector< llama_chat_msg>& past_msg, + const llama_chat_msg& new_msg, + bool add_ass, + bool use_jinja); // Returns an example of formatted chat -std::string llama_chat_format_example(const struct llama_model * model, - const std::string & tmpl); +std::string llama_chat_format_example(const struct llama_model* model, + const common_chat_template& tmpl, bool use_jinja); + +common_chat_templates llama_chat_templates_from_model(const struct llama_model* model, const std::string& chat_template_override); + // // KV cache utils @@ -502,3 +527,5 @@ void yaml_dump_string_multiline(FILE * stream, const char * prop_name, const cha void yaml_dump_non_result_info( FILE * stream, const gpt_params & params, const llama_context * lctx, const std::string & timestamp, const std::vector & prompt_tokens, const char * model_desc); + +std::string string_format(const char* fmt, ...); diff --git a/common/minja.hpp b/common/minja.hpp new file mode 100644 index 000000000..3176280e4 --- /dev/null +++ b/common/minja.hpp @@ -0,0 +1,3029 @@ +/* + Copyright 2024 Google LLC + + Use of this source code is governed by an MIT-style + license that can be found in the LICENSE file or at + https://opensource.org/licenses/MIT. +*/ +// SPDX-License-Identifier: MIT +#pragma once + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +using json = nlohmann::ordered_json; + +namespace minja { + +class Context; + +struct Options { + bool trim_blocks; // removes the first newline after a block + bool lstrip_blocks; // removes leading whitespace on the line of the block + bool keep_trailing_newline; // don't remove last newline +}; + +struct ArgumentsValue; + +inline std::string normalize_newlines(const std::string & s) { +#ifdef _WIN32 + static const std::regex nl_regex("\r\n"); + return std::regex_replace(s, nl_regex, "\n"); +#else + return s; +#endif +} + +/* Values that behave roughly like in Python. */ +class Value : public std::enable_shared_from_this { +public: + using CallableType = std::function &, ArgumentsValue &)>; + using FilterType = std::function &, ArgumentsValue &)>; + +private: + using ObjectType = nlohmann::ordered_map; // Only contains primitive keys + using ArrayType = std::vector; + + std::shared_ptr array_; + std::shared_ptr object_; + std::shared_ptr callable_; + json primitive_; + + Value(const std::shared_ptr & array) : array_(array) {} + Value(const std::shared_ptr & object) : object_(object) {} + Value(const std::shared_ptr & callable) : object_(std::make_shared()), callable_(callable) {} + + /* Python-style string repr */ + static void dump_string(const json & primitive, std::ostringstream & out, char string_quote = '\'') { + if (!primitive.is_string()) throw std::runtime_error("Value is not a string: " + primitive.dump()); + auto s = primitive.dump(); + if (string_quote == '"' || s.find('\'') != std::string::npos) { + out << s; + return; + } + // Reuse json dump, just changing string quotes + out << string_quote; + for (size_t i = 1, n = s.size() - 1; i < n; ++i) { + if (s[i] == '\\' && s[i + 1] == '"') { + out << '"'; + i++; + } else if (s[i] == string_quote) { + out << '\\' << string_quote; + } else { + out << s[i]; + } + } + out << string_quote; + } + void dump(std::ostringstream & out, int indent = -1, int level = 0, bool to_json = false) const { + auto print_indent = [&](int level) { + if (indent > 0) { + out << "\n"; + for (int i = 0, n = level * indent; i < n; ++i) out << ' '; + } + }; + auto print_sub_sep = [&]() { + out << ','; + if (indent < 0) out << ' '; + else print_indent(level + 1); + }; + + auto string_quote = to_json ? '"' : '\''; + + if (is_null()) out << "null"; + else if (array_) { + out << "["; + print_indent(level + 1); + for (size_t i = 0; i < array_->size(); ++i) { + if (i) print_sub_sep(); + (*array_)[i].dump(out, indent, level + 1, to_json); + } + print_indent(level); + out << "]"; + } else if (object_) { + out << "{"; + print_indent(level + 1); + for (auto begin = object_->begin(), it = begin; it != object_->end(); ++it) { + if (it != begin) print_sub_sep(); + if (it->first.is_string()) { + dump_string(it->first, out, string_quote); + } else { + out << string_quote << it->first.dump() << string_quote; + } + out << ": "; + it->second.dump(out, indent, level + 1, to_json); + } + print_indent(level); + out << "}"; + } else if (callable_) { + throw std::runtime_error("Cannot dump callable to JSON"); + } else if (is_boolean() && !to_json) { + out << (this->to_bool() ? "True" : "False"); + } else if (is_string() && !to_json) { + dump_string(primitive_, out, string_quote); + } else { + out << primitive_.dump(); + } + } + +public: + Value() {} + Value(const bool& v) : primitive_(v) {} + Value(const int64_t & v) : primitive_(v) {} + Value(const double& v) : primitive_(v) {} + Value(const std::nullptr_t &) {} + Value(const std::string & v) : primitive_(v) {} + Value(const char * v) : primitive_(std::string(v)) {} + + Value(const json & v) { + if (v.is_object()) { + auto object = std::make_shared(); + for (auto it = v.begin(); it != v.end(); ++it) { + (*object)[it.key()] = it.value(); + } + object_ = std::move(object); + } else if (v.is_array()) { + auto array = std::make_shared(); + for (const auto& item : v) { + array->push_back(Value(item)); + } + array_ = array; + } else { + primitive_ = v; + } + } + + std::vector keys() { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + std::vector res; + for (const auto& item : *object_) { + res.push_back(item.first); + } + return res; + } + + size_t size() const { + if (is_object()) return object_->size(); + if (is_array()) return array_->size(); + if (is_string()) return primitive_.get().length(); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + + static Value array(const std::vector values = {}) { + auto array = std::make_shared(); + for (const auto& item : values) { + array->push_back(item); + } + return Value(array); + } + static Value object(const std::shared_ptr object = std::make_shared()) { + return Value(object); + } + static Value callable(const CallableType & callable) { + return Value(std::make_shared(callable)); + } + + void insert(size_t index, const Value& v) { + if (!array_) + throw std::runtime_error("Value is not an array: " + dump()); + array_->insert(array_->begin() + index, v); + } + void push_back(const Value& v) { + if (!array_) + throw std::runtime_error("Value is not an array: " + dump()); + array_->push_back(v); + } + Value pop(const Value& index) { + if (is_array()) { + if (array_->empty()) + throw std::runtime_error("pop from empty list"); + if (index.is_null()) { + auto ret = array_->back(); + array_->pop_back(); + return ret; + } else if (!index.is_number_integer()) { + throw std::runtime_error("pop index must be an integer: " + index.dump()); + } else { + auto i = index.get(); + if (i < 0 || i >= static_cast(array_->size())) + throw std::runtime_error("pop index out of range: " + index.dump()); + auto it = array_->begin() + (i < 0 ? array_->size() + i : i); + auto ret = *it; + array_->erase(it); + return ret; + } + } else if (is_object()) { + if (!index.is_hashable()) + throw std::runtime_error("Unhashable type: " + index.dump()); + auto it = object_->find(index.primitive_); + if (it == object_->end()) + throw std::runtime_error("Key not found: " + index.dump()); + auto ret = it->second; + object_->erase(it); + return ret; + } else { + throw std::runtime_error("Value is not an array or object: " + dump()); + } + } + Value get(const Value& key) { + if (array_) { + if (!key.is_number_integer()) { + return Value(); + } + auto index = key.get(); + return array_->at(index < 0 ? array_->size() + index : index); + } else if (object_) { + if (!key.is_hashable()) throw std::runtime_error("Unhashable type: " + dump()); + auto it = object_->find(key.primitive_); + if (it == object_->end()) return Value(); + return it->second; + } + return Value(); + } + void set(const Value& key, const Value& value) { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + if (!key.is_hashable()) throw std::runtime_error("Unhashable type: " + dump()); + (*object_)[key.primitive_] = value; + } + Value call(const std::shared_ptr & context, ArgumentsValue & args) const { + if (!callable_) throw std::runtime_error("Value is not callable: " + dump()); + return (*callable_)(context, args); + } + + bool is_object() const { return !!object_; } + bool is_array() const { return !!array_; } + bool is_callable() const { return !!callable_; } + bool is_null() const { return !object_ && !array_ && primitive_.is_null() && !callable_; } + bool is_boolean() const { return primitive_.is_boolean(); } + bool is_number_integer() const { return primitive_.is_number_integer(); } + bool is_number_float() const { return primitive_.is_number_float(); } + bool is_number() const { return primitive_.is_number(); } + bool is_string() const { return primitive_.is_string(); } + bool is_iterable() const { return is_array() || is_object() || is_string(); } + + bool is_primitive() const { return !array_ && !object_ && !callable_; } + bool is_hashable() const { return is_primitive(); } + + bool empty() const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_string()) return primitive_.empty(); + if (is_array()) return array_->empty(); + if (is_object()) return object_->empty(); + return false; + } + + void for_each(const std::function & callback) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (array_) { + for (auto& item : *array_) { + callback(item); + } + } else if (object_) { + for (auto & item : *object_) { + Value key(item.first); + callback(key); + } + } else if (is_string()) { + for (char c : primitive_.get()) { + auto val = Value(std::string(1, c)); + callback(val); + } + } else { + throw std::runtime_error("Value is not iterable: " + dump()); + } + } + + bool to_bool() const { + if (is_null()) return false; + if (is_boolean()) return get(); + if (is_number()) return get() != 0; + if (is_string()) return !get().empty(); + if (is_array()) return !empty(); + return true; + } + + int64_t to_int() const { + if (is_null()) return 0; + if (is_boolean()) return get() ? 1 : 0; + if (is_number()) return static_cast(get()); + if (is_string()) { + try { + return std::stol(get()); + } catch (const std::exception &) { + return 0; + } + } + return 0; + } + + bool operator<(const Value & other) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_number() && other.is_number()) return get() < other.get(); + if (is_string() && other.is_string()) return get() < other.get(); + throw std::runtime_error("Cannot compare values: " + dump() + " < " + other.dump()); + } + bool operator>=(const Value & other) const { return !(*this < other); } + + bool operator>(const Value & other) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_number() && other.is_number()) return get() > other.get(); + if (is_string() && other.is_string()) return get() > other.get(); + throw std::runtime_error("Cannot compare values: " + dump() + " > " + other.dump()); + } + bool operator<=(const Value & other) const { return !(*this > other); } + + bool operator==(const Value & other) const { + if (callable_ || other.callable_) { + if (callable_.get() != other.callable_.get()) return false; + } + if (array_) { + if (!other.array_) return false; + if (array_->size() != other.array_->size()) return false; + for (size_t i = 0; i < array_->size(); ++i) { + if (!(*array_)[i].to_bool() || !(*other.array_)[i].to_bool() || (*array_)[i] != (*other.array_)[i]) return false; + } + return true; + } else if (object_) { + if (!other.object_) return false; + if (object_->size() != other.object_->size()) return false; + for (const auto& item : *object_) { + if (!item.second.to_bool() || !other.object_->count(item.first) || item.second != other.object_->at(item.first)) return false; + } + return true; + } else { + return primitive_ == other.primitive_; + } + } + bool operator!=(const Value & other) const { return !(*this == other); } + + bool contains(const char * key) const { return contains(std::string(key)); } + bool contains(const std::string & key) const { + if (array_) { + return false; + } else if (object_) { + return object_->find(key) != object_->end(); + } else { + throw std::runtime_error("contains can only be called on arrays and objects: " + dump()); + } + } + bool contains(const Value & value) const { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (array_) { + for (const auto& item : *array_) { + if (item.to_bool() && item == value) return true; + } + return false; + } else if (object_) { + if (!value.is_hashable()) throw std::runtime_error("Unhashable type: " + value.dump()); + return object_->find(value.primitive_) != object_->end(); + } else { + throw std::runtime_error("contains can only be called on arrays and objects: " + dump()); + } + } + void erase(size_t index) { + if (!array_) throw std::runtime_error("Value is not an array: " + dump()); + array_->erase(array_->begin() + index); + } + void erase(const std::string & key) { + if (!object_) throw std::runtime_error("Value is not an object: " + dump()); + object_->erase(key); + } + const Value& at(const Value & index) const { + return const_cast(this)->at(index); + } + Value& at(const Value & index) { + if (!index.is_hashable()) throw std::runtime_error("Unhashable type: " + dump()); + if (is_array()) return array_->at(index.get()); + if (is_object()) return object_->at(index.primitive_); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + const Value& at(size_t index) const { + return const_cast(this)->at(index); + } + Value& at(size_t index) { + if (is_null()) + throw std::runtime_error("Undefined value or reference"); + if (is_array()) return array_->at(index); + if (is_object()) return object_->at(index); + throw std::runtime_error("Value is not an array or object: " + dump()); + } + + template + T get(const std::string & key, T default_value) const { + if (!contains(key)) return default_value; + return at(key).get(); + } + + template + T get() const { + if (is_primitive()) return primitive_.get(); + throw std::runtime_error("get not defined for this value type: " + dump()); + } + + std::string dump(int indent=-1, bool to_json=false) const { + std::ostringstream out; + dump(out, indent, 0, to_json); + return out.str(); + } + + Value operator-() const { + if (is_number_integer()) + return -get(); + else + return -get(); + } + std::string to_str() const { + if (is_string()) return get(); + if (is_number_integer()) return std::to_string(get()); + if (is_number_float()) return std::to_string(get()); + if (is_boolean()) return get() ? "True" : "False"; + if (is_null()) return "None"; + return dump(); + } + Value operator+(const Value& rhs) const { + if (is_string() || rhs.is_string()) { + return to_str() + rhs.to_str(); + } else if (is_number_integer() && rhs.is_number_integer()) { + return get() + rhs.get(); + } else if (is_array() && rhs.is_array()) { + auto res = Value::array(); + for (const auto& item : *array_) res.push_back(item); + for (const auto& item : *rhs.array_) res.push_back(item); + return res; + } else { + return get() + rhs.get(); + } + } + Value operator-(const Value& rhs) const { + if (is_number_integer() && rhs.is_number_integer()) + return get() - rhs.get(); + else + return get() - rhs.get(); + } + Value operator*(const Value& rhs) const { + if (is_string() && rhs.is_number_integer()) { + std::ostringstream out; + for (int64_t i = 0, n = rhs.get(); i < n; ++i) { + out << to_str(); + } + return out.str(); + } + else if (is_number_integer() && rhs.is_number_integer()) + return get() * rhs.get(); + else + return get() * rhs.get(); + } + Value operator/(const Value& rhs) const { + if (is_number_integer() && rhs.is_number_integer()) + return get() / rhs.get(); + else + return get() / rhs.get(); + } + Value operator%(const Value& rhs) const { + return get() % rhs.get(); + } +}; + +struct ArgumentsValue { + std::vector args; + std::vector> kwargs; + + bool has_named(const std::string & name) { + for (const auto & p : kwargs) { + if (p.first == name) return true; + } + return false; + } + + Value get_named(const std::string & name) { + for (const auto & [key, value] : kwargs) { + if (key == name) return value; + } + return Value(); + } + + bool empty() { + return args.empty() && kwargs.empty(); + } + + void expectArgs(const std::string & method_name, const std::pair & pos_count, const std::pair & kw_count) { + if (args.size() < pos_count.first || args.size() > pos_count.second || kwargs.size() < kw_count.first || kwargs.size() > kw_count.second) { + std::ostringstream out; + out << method_name << " must have between " << pos_count.first << " and " << pos_count.second << " positional arguments and between " << kw_count.first << " and " << kw_count.second << " keyword arguments"; + throw std::runtime_error(out.str()); + } + } +}; + +template <> +inline json Value::get() const { + if (is_primitive()) return primitive_; + if (is_null()) return json(); + if (array_) { + std::vector res; + for (const auto& item : *array_) { + res.push_back(item.get()); + } + return res; + } + if (object_) { + json res = json::object(); + for (const auto& [key, value] : *object_) { + if (key.is_string()) { + res[key.get()] = value.get(); + } else if (key.is_primitive()) { + res[key.dump()] = value.get(); + } else { + throw std::runtime_error("Invalid key type for conversion to JSON: " + key.dump()); + } + } + if (is_callable()) { + res["__callable__"] = true; + } + return res; + } + throw std::runtime_error("get not defined for this value type: " + dump()); +} + +} // namespace minja + +namespace std { + template <> + struct hash { + size_t operator()(const minja::Value & v) const { + if (!v.is_hashable()) + throw std::runtime_error("Unsupported type for hashing: " + v.dump()); + return std::hash()(v.get()); + } + }; +} // namespace std + +namespace minja { + +static std::string error_location_suffix(const std::string & source, size_t pos) { + auto get_line = [&](size_t line) { + auto start = source.begin(); + for (size_t i = 1; i < line; ++i) { + start = std::find(start, source.end(), '\n') + 1; + } + auto end = std::find(start, source.end(), '\n'); + return std::string(start, end); + }; + auto start = source.begin(); + auto end = source.end(); + auto it = start + pos; + auto line = std::count(start, it, '\n') + 1; + auto max_line = std::count(start, end, '\n') + 1; + auto col = pos - std::string(start, it).rfind('\n'); + std::ostringstream out; + out << " at row " << line << ", column " << col << ":\n"; + if (line > 1) out << get_line(line - 1) << "\n"; + out << get_line(line) << "\n"; + out << std::string(col - 1, ' ') << "^\n"; + if (line < max_line) out << get_line(line + 1) << "\n"; + + return out.str(); +} + +class Context : public std::enable_shared_from_this { + protected: + Value values_; + std::shared_ptr parent_; + public: + Context(Value && values, const std::shared_ptr & parent = nullptr) : values_(std::move(values)), parent_(parent) { + if (!values_.is_object()) throw std::runtime_error("Context values must be an object: " + values_.dump()); + } + virtual ~Context() {} + + static std::shared_ptr builtins(); + static std::shared_ptr make(Value && values, const std::shared_ptr & parent = builtins()); + + std::vector keys() { + return values_.keys(); + } + virtual Value get(const Value & key) { + if (values_.contains(key)) return values_.at(key); + if (parent_) return parent_->get(key); + return Value(); + } + virtual Value & at(const Value & key) { + if (values_.contains(key)) return values_.at(key); + if (parent_) return parent_->at(key); + throw std::runtime_error("Undefined variable: " + key.dump()); + } + virtual bool contains(const Value & key) { + if (values_.contains(key)) return true; + if (parent_) return parent_->contains(key); + return false; + } + virtual void set(const Value & key, const Value & value) { + values_.set(key, value); + } +}; + +struct Location { + std::shared_ptr source; + size_t pos; +}; + +class Expression { +protected: + virtual Value do_evaluate(const std::shared_ptr & context) const = 0; +public: + using Parameters = std::vector>>; + + Location location; + + Expression(const Location & location) : location(location) {} + virtual ~Expression() = default; + + Value evaluate(const std::shared_ptr & context) const { + try { + return do_evaluate(context); + } catch (const std::exception & e) { + std::ostringstream out; + out << e.what(); + if (location.source) out << error_location_suffix(*location.source, location.pos); + throw std::runtime_error(out.str()); + } + } +}; + +class VariableExpr : public Expression { + std::string name; +public: + VariableExpr(const Location & loc, const std::string& n) + : Expression(loc), name(n) {} + std::string get_name() const { return name; } + Value do_evaluate(const std::shared_ptr & context) const override { + if (!context->contains(name)) { + return Value(); + } + return context->at(name); + } +}; + +static void destructuring_assign(const std::vector & var_names, const std::shared_ptr & context, Value& item) { + if (var_names.size() == 1) { + Value name(var_names[0]); + context->set(name, item); + } else { + if (!item.is_array() || item.size() != var_names.size()) { + throw std::runtime_error("Mismatched number of variables and items in destructuring assignment"); + } + for (size_t i = 0; i < var_names.size(); ++i) { + context->set(var_names[i], item.at(i)); + } + } +} + +enum SpaceHandling { Keep, Strip, StripSpaces, StripNewline }; + +class TemplateToken { +public: + enum class Type { Text, Expression, If, Else, Elif, EndIf, For, EndFor, Generation, EndGeneration, Set, EndSet, Comment, Macro, EndMacro, Filter, EndFilter, Break, Continue }; + + static std::string typeToString(Type t) { + switch (t) { + case Type::Text: return "text"; + case Type::Expression: return "expression"; + case Type::If: return "if"; + case Type::Else: return "else"; + case Type::Elif: return "elif"; + case Type::EndIf: return "endif"; + case Type::For: return "for"; + case Type::EndFor: return "endfor"; + case Type::Set: return "set"; + case Type::EndSet: return "endset"; + case Type::Comment: return "comment"; + case Type::Macro: return "macro"; + case Type::EndMacro: return "endmacro"; + case Type::Filter: return "filter"; + case Type::EndFilter: return "endfilter"; + case Type::Generation: return "generation"; + case Type::EndGeneration: return "endgeneration"; + case Type::Break: return "break"; + case Type::Continue: return "continue"; + } + return "Unknown"; + } + + TemplateToken(Type type, const Location & location, SpaceHandling pre, SpaceHandling post) : type(type), location(location), pre_space(pre), post_space(post) {} + virtual ~TemplateToken() = default; + + Type type; + Location location; + SpaceHandling pre_space = SpaceHandling::Keep; + SpaceHandling post_space = SpaceHandling::Keep; +}; + +struct TextTemplateToken : public TemplateToken { + std::string text; + TextTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, const std::string& t) : TemplateToken(Type::Text, loc, pre, post), text(t) {} +}; + +struct ExpressionTemplateToken : public TemplateToken { + std::shared_ptr expr; + ExpressionTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, std::shared_ptr && e) : TemplateToken(Type::Expression, loc, pre, post), expr(std::move(e)) {} +}; + +struct IfTemplateToken : public TemplateToken { + std::shared_ptr condition; + IfTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, std::shared_ptr && c) : TemplateToken(Type::If, loc, pre, post), condition(std::move(c)) {} +}; + +struct ElifTemplateToken : public TemplateToken { + std::shared_ptr condition; + ElifTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, std::shared_ptr && c) : TemplateToken(Type::Elif, loc, pre, post), condition(std::move(c)) {} +}; + +struct ElseTemplateToken : public TemplateToken { + ElseTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::Else, loc, pre, post) {} +}; + +struct EndIfTemplateToken : public TemplateToken { + EndIfTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndIf, loc, pre, post) {} +}; + +struct MacroTemplateToken : public TemplateToken { + std::shared_ptr name; + Expression::Parameters params; + MacroTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, std::shared_ptr && n, Expression::Parameters && p) + : TemplateToken(Type::Macro, loc, pre, post), name(std::move(n)), params(std::move(p)) {} +}; + +struct EndMacroTemplateToken : public TemplateToken { + EndMacroTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndMacro, loc, pre, post) {} +}; + +struct FilterTemplateToken : public TemplateToken { + std::shared_ptr filter; + FilterTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, std::shared_ptr && filter) + : TemplateToken(Type::Filter, loc, pre, post), filter(std::move(filter)) {} +}; + +struct EndFilterTemplateToken : public TemplateToken { + EndFilterTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndFilter, loc, pre, post) {} +}; + +struct ForTemplateToken : public TemplateToken { + std::vector var_names; + std::shared_ptr iterable; + std::shared_ptr condition; + bool recursive; + ForTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, const std::vector & vns, std::shared_ptr && iter, + std::shared_ptr && c, bool r) + : TemplateToken(Type::For, loc, pre, post), var_names(vns), iterable(std::move(iter)), condition(std::move(c)), recursive(r) {} +}; + +struct EndForTemplateToken : public TemplateToken { + EndForTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndFor, loc, pre, post) {} +}; + +struct GenerationTemplateToken : public TemplateToken { + GenerationTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::Generation, loc, pre, post) {} +}; + +struct EndGenerationTemplateToken : public TemplateToken { + EndGenerationTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndGeneration, loc, pre, post) {} +}; + +struct SetTemplateToken : public TemplateToken { + std::string ns; + std::vector var_names; + std::shared_ptr value; + SetTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, const std::string & ns, const std::vector & vns, std::shared_ptr && v) + : TemplateToken(Type::Set, loc, pre, post), ns(ns), var_names(vns), value(std::move(v)) {} +}; + +struct EndSetTemplateToken : public TemplateToken { + EndSetTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post) : TemplateToken(Type::EndSet, loc, pre, post) {} +}; + +struct CommentTemplateToken : public TemplateToken { + std::string text; + CommentTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, const std::string& t) : TemplateToken(Type::Comment, loc, pre, post), text(t) {} +}; + +enum class LoopControlType { Break, Continue }; + +class LoopControlException : public std::runtime_error { +public: + LoopControlType control_type; + LoopControlException(const std::string & message, LoopControlType control_type) : std::runtime_error(message), control_type(control_type) {} + LoopControlException(LoopControlType control_type) + : std::runtime_error((control_type == LoopControlType::Continue ? "continue" : "break") + std::string(" outside of a loop")), + control_type(control_type) {} +}; + +struct LoopControlTemplateToken : public TemplateToken { + LoopControlType control_type; + LoopControlTemplateToken(const Location & loc, SpaceHandling pre, SpaceHandling post, LoopControlType control_type) : TemplateToken(Type::Break, loc, pre, post), control_type(control_type) {} +}; + +class TemplateNode { + Location location_; +protected: + virtual void do_render(std::ostringstream & out, const std::shared_ptr & context) const = 0; + +public: + TemplateNode(const Location & location) : location_(location) {} + void render(std::ostringstream & out, const std::shared_ptr & context) const { + try { + do_render(out, context); + } catch (const LoopControlException & e) { + // TODO: make stack creation lazy. Only needed if it was thrown outside of a loop. + std::ostringstream err; + err << e.what(); + if (location_.source) err << error_location_suffix(*location_.source, location_.pos); + throw LoopControlException(err.str(), e.control_type); + } catch (const std::exception & e) { + std::ostringstream err; + err << e.what(); + if (location_.source) err << error_location_suffix(*location_.source, location_.pos); + throw std::runtime_error(err.str()); + } + } + const Location & location() const { return location_; } + virtual ~TemplateNode() = default; + std::string render(const std::shared_ptr & context) const { + std::ostringstream out; + render(out, context); + return out.str(); + } +}; + +class SequenceNode : public TemplateNode { + std::vector> children; +public: + SequenceNode(const Location & loc, std::vector> && c) + : TemplateNode(loc), children(std::move(c)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + for (const auto& child : children) child->render(out, context); + } +}; + +class TextNode : public TemplateNode { + std::string text; +public: + TextNode(const Location & loc, const std::string& t) : TemplateNode(loc), text(t) {} + void do_render(std::ostringstream & out, const std::shared_ptr &) const override { + out << text; + } +}; + +class ExpressionNode : public TemplateNode { + std::shared_ptr expr; +public: + ExpressionNode(const Location & loc, std::shared_ptr && e) : TemplateNode(loc), expr(std::move(e)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + if (!expr) throw std::runtime_error("ExpressionNode.expr is null"); + auto result = expr->evaluate(context); + if (result.is_string()) { + out << result.get(); + } else if (result.is_boolean()) { + out << (result.get() ? "True" : "False"); + } else if (!result.is_null()) { + out << result.dump(); + } + } +}; + +class IfNode : public TemplateNode { + std::vector, std::shared_ptr>> cascade; +public: + IfNode(const Location & loc, std::vector, std::shared_ptr>> && c) + : TemplateNode(loc), cascade(std::move(c)) {} + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + for (const auto& branch : cascade) { + auto enter_branch = true; + if (branch.first) { + enter_branch = branch.first->evaluate(context).to_bool(); + } + if (enter_branch) { + if (!branch.second) throw std::runtime_error("IfNode.cascade.second is null"); + branch.second->render(out, context); + return; + } + } + } +}; + +class LoopControlNode : public TemplateNode { + LoopControlType control_type_; + public: + LoopControlNode(const Location & loc, LoopControlType control_type) : TemplateNode(loc), control_type_(control_type) {} + void do_render(std::ostringstream &, const std::shared_ptr &) const override { + throw LoopControlException(control_type_); + } +}; + +class ForNode : public TemplateNode { + std::vector var_names; + std::shared_ptr iterable; + std::shared_ptr condition; + std::shared_ptr body; + bool recursive; + std::shared_ptr else_body; +public: + ForNode(const Location & loc, std::vector && var_names, std::shared_ptr && iterable, + std::shared_ptr && condition, std::shared_ptr && body, bool recursive, std::shared_ptr && else_body) + : TemplateNode(loc), var_names(var_names), iterable(std::move(iterable)), condition(std::move(condition)), body(std::move(body)), recursive(recursive), else_body(std::move(else_body)) {} + + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + // https://jinja.palletsprojects.com/en/3.0.x/templates/#for + if (!iterable) throw std::runtime_error("ForNode.iterable is null"); + if (!body) throw std::runtime_error("ForNode.body is null"); + + auto iterable_value = iterable->evaluate(context); + Value::CallableType loop_function; + + std::function visit = [&](Value& iter) { + auto filtered_items = Value::array(); + if (!iter.is_null()) { + if (!iterable_value.is_iterable()) { + throw std::runtime_error("For loop iterable must be iterable: " + iterable_value.dump()); + } + iterable_value.for_each([&](Value & item) { + destructuring_assign(var_names, context, item); + if (!condition || condition->evaluate(context).to_bool()) { + filtered_items.push_back(item); + } + }); + } + if (filtered_items.empty()) { + if (else_body) { + else_body->render(out, context); + } + } else { + auto loop = recursive ? Value::callable(loop_function) : Value::object(); + loop.set("length", (int64_t) filtered_items.size()); + + size_t cycle_index = 0; + loop.set("cycle", Value::callable([&](const std::shared_ptr &, ArgumentsValue & args) { + if (args.args.empty() || !args.kwargs.empty()) { + throw std::runtime_error("cycle() expects at least 1 positional argument and no named arg"); + } + auto item = args.args[cycle_index]; + cycle_index = (cycle_index + 1) % args.args.size(); + return item; + })); + auto loop_context = Context::make(Value::object(), context); + loop_context->set("loop", loop); + for (size_t i = 0, n = filtered_items.size(); i < n; ++i) { + auto & item = filtered_items.at(i); + destructuring_assign(var_names, loop_context, item); + loop.set("index", (int64_t) i + 1); + loop.set("index0", (int64_t) i); + loop.set("revindex", (int64_t) (n - i)); + loop.set("revindex0", (int64_t) (n - i - 1)); + loop.set("length", (int64_t) n); + loop.set("first", i == 0); + loop.set("last", i == (n - 1)); + loop.set("previtem", i > 0 ? filtered_items.at(i - 1) : Value()); + loop.set("nextitem", i < n - 1 ? filtered_items.at(i + 1) : Value()); + try { + body->render(out, loop_context); + } catch (const LoopControlException & e) { + if (e.control_type == LoopControlType::Break) break; + if (e.control_type == LoopControlType::Continue) continue; + } + } + } + }; + + if (recursive) { + loop_function = [&](const std::shared_ptr &, ArgumentsValue & args) { + if (args.args.size() != 1 || !args.kwargs.empty() || !args.args[0].is_array()) { + throw std::runtime_error("loop() expects exactly 1 positional iterable argument"); + } + auto & items = args.args[0]; + visit(items); + return Value(); + }; + } + + visit(iterable_value); + } +}; + +class MacroNode : public TemplateNode { + std::shared_ptr name; + Expression::Parameters params; + std::shared_ptr body; + std::unordered_map named_param_positions; +public: + MacroNode(const Location & loc, std::shared_ptr && n, Expression::Parameters && p, std::shared_ptr && b) + : TemplateNode(loc), name(std::move(n)), params(std::move(p)), body(std::move(b)) { + for (size_t i = 0; i < params.size(); ++i) { + const auto & name = params[i].first; + if (!name.empty()) { + named_param_positions[name] = i; + } + } + } + void do_render(std::ostringstream &, const std::shared_ptr & macro_context) const override { + if (!name) throw std::runtime_error("MacroNode.name is null"); + if (!body) throw std::runtime_error("MacroNode.body is null"); + auto callable = Value::callable([&](const std::shared_ptr & context, ArgumentsValue & args) { + auto call_context = macro_context; + std::vector param_set(params.size(), false); + for (size_t i = 0, n = args.args.size(); i < n; i++) { + auto & arg = args.args[i]; + if (i >= params.size()) throw std::runtime_error("Too many positional arguments for macro " + name->get_name()); + param_set[i] = true; + auto & param_name = params[i].first; + call_context->set(param_name, arg); + } + for (auto & [arg_name, value] : args.kwargs) { + auto it = named_param_positions.find(arg_name); + if (it == named_param_positions.end()) throw std::runtime_error("Unknown parameter name for macro " + name->get_name() + ": " + arg_name); + + call_context->set(arg_name, value); + param_set[it->second] = true; + } + // Set default values for parameters that were not passed + for (size_t i = 0, n = params.size(); i < n; i++) { + if (!param_set[i] && params[i].second != nullptr) { + auto val = params[i].second->evaluate(context); + call_context->set(params[i].first, val); + } + } + return body->render(call_context); + }); + macro_context->set(name->get_name(), callable); + } +}; + +class FilterNode : public TemplateNode { + std::shared_ptr filter; + std::shared_ptr body; + +public: + FilterNode(const Location & loc, std::shared_ptr && f, std::shared_ptr && b) + : TemplateNode(loc), filter(std::move(f)), body(std::move(b)) {} + + void do_render(std::ostringstream & out, const std::shared_ptr & context) const override { + if (!filter) throw std::runtime_error("FilterNode.filter is null"); + if (!body) throw std::runtime_error("FilterNode.body is null"); + auto filter_value = filter->evaluate(context); + if (!filter_value.is_callable()) { + throw std::runtime_error("Filter must be a callable: " + filter_value.dump()); + } + std::string rendered_body = body->render(context); + + ArgumentsValue filter_args = {{Value(rendered_body)}, {}}; + auto result = filter_value.call(context, filter_args); + out << result.to_str(); + } +}; + +class SetNode : public TemplateNode { + std::string ns; + std::vector var_names; + std::shared_ptr value; +public: + SetNode(const Location & loc, const std::string & ns, const std::vector & vns, std::shared_ptr && v) + : TemplateNode(loc), ns(ns), var_names(vns), value(std::move(v)) {} + void do_render(std::ostringstream &, const std::shared_ptr & context) const override { + if (!value) throw std::runtime_error("SetNode.value is null"); + if (!ns.empty()) { + if (var_names.size() != 1) { + throw std::runtime_error("Namespaced set only supports a single variable name"); + } + auto & name = var_names[0]; + auto ns_value = context->get(ns); + if (!ns_value.is_object()) throw std::runtime_error("Namespace '" + ns + "' is not an object"); + ns_value.set(name, this->value->evaluate(context)); + } else { + auto val = value->evaluate(context); + destructuring_assign(var_names, context, val); + } + } +}; + +class SetTemplateNode : public TemplateNode { + std::string name; + std::shared_ptr template_value; +public: + SetTemplateNode(const Location & loc, const std::string & name, std::shared_ptr && tv) + : TemplateNode(loc), name(name), template_value(std::move(tv)) {} + void do_render(std::ostringstream &, const std::shared_ptr & context) const override { + if (!template_value) throw std::runtime_error("SetTemplateNode.template_value is null"); + Value value { template_value->render(context) }; + context->set(name, value); + } +}; + +class IfExpr : public Expression { + std::shared_ptr condition; + std::shared_ptr then_expr; + std::shared_ptr else_expr; +public: + IfExpr(const Location & loc, std::shared_ptr && c, std::shared_ptr && t, std::shared_ptr && e) + : Expression(loc), condition(std::move(c)), then_expr(std::move(t)), else_expr(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!condition) throw std::runtime_error("IfExpr.condition is null"); + if (!then_expr) throw std::runtime_error("IfExpr.then_expr is null"); + if (condition->evaluate(context).to_bool()) { + return then_expr->evaluate(context); + } + if (else_expr) { + return else_expr->evaluate(context); + } + return nullptr; + } +}; + +class LiteralExpr : public Expression { + Value value; +public: + LiteralExpr(const Location & loc, const Value& v) + : Expression(loc), value(v) {} + Value do_evaluate(const std::shared_ptr &) const override { return value; } +}; + +class ArrayExpr : public Expression { + std::vector> elements; +public: + ArrayExpr(const Location & loc, std::vector> && e) + : Expression(loc), elements(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + auto result = Value::array(); + for (const auto& e : elements) { + if (!e) throw std::runtime_error("Array element is null"); + result.push_back(e->evaluate(context)); + } + return result; + } +}; + +class DictExpr : public Expression { + std::vector, std::shared_ptr>> elements; +public: + DictExpr(const Location & loc, std::vector, std::shared_ptr>> && e) + : Expression(loc), elements(std::move(e)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + auto result = Value::object(); + for (const auto& [key, value] : elements) { + if (!key) throw std::runtime_error("Dict key is null"); + if (!value) throw std::runtime_error("Dict value is null"); + result.set(key->evaluate(context), value->evaluate(context)); + } + return result; + } +}; + +class SliceExpr : public Expression { +public: + std::shared_ptr start, end, step; + SliceExpr(const Location & loc, std::shared_ptr && s, std::shared_ptr && e, std::shared_ptr && st = nullptr) + : Expression(loc), start(std::move(s)), end(std::move(e)), step(std::move(st)) {} + Value do_evaluate(const std::shared_ptr &) const override { + throw std::runtime_error("SliceExpr not implemented"); + } +}; + +class SubscriptExpr : public Expression { + std::shared_ptr base; + std::shared_ptr index; +public: + SubscriptExpr(const Location & loc, std::shared_ptr && b, std::shared_ptr && i) + : Expression(loc), base(std::move(b)), index(std::move(i)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!base) throw std::runtime_error("SubscriptExpr.base is null"); + if (!index) throw std::runtime_error("SubscriptExpr.index is null"); + auto target_value = base->evaluate(context); + if (auto slice = dynamic_cast(index.get())) { + auto len = target_value.size(); + auto wrap = [len](int64_t i) -> int64_t { + if (i < 0) { + return i + len; + } + return i; + }; + int64_t step = slice->step ? slice->step->evaluate(context).get() : 1; + if (!step) { + throw std::runtime_error("slice step cannot be zero"); + } + int64_t start = slice->start ? wrap(slice->start->evaluate(context).get()) : (step < 0 ? len - 1 : 0); + int64_t end = slice->end ? wrap(slice->end->evaluate(context).get()) : (step < 0 ? -1 : len); + if (target_value.is_string()) { + std::string s = target_value.get(); + + std::string result; + if (start < end && step == 1) { + result = s.substr(start, end - start); + } else { + for (int64_t i = start; step > 0 ? i < end : i > end; i += step) { + result += s[i]; + } + } + return result; + + } else if (target_value.is_array()) { + auto result = Value::array(); + for (int64_t i = start; step > 0 ? i < end : i > end; i += step) { + result.push_back(target_value.at(i)); + } + return result; + } else { + throw std::runtime_error(target_value.is_null() ? "Cannot subscript null" : "Subscripting only supported on arrays and strings"); + } + } else { + auto index_value = index->evaluate(context); + if (target_value.is_null()) { + if (auto t = dynamic_cast(base.get())) { + throw std::runtime_error("'" + t->get_name() + "' is " + (context->contains(t->get_name()) ? "null" : "not defined")); + } + throw std::runtime_error("Trying to access property '" + index_value.dump() + "' on null!"); + } + return target_value.get(index_value); + } + } +}; + +class UnaryOpExpr : public Expression { +public: + enum class Op { Plus, Minus, LogicalNot, Expansion, ExpansionDict }; + std::shared_ptr expr; + Op op; + UnaryOpExpr(const Location & loc, std::shared_ptr && e, Op o) + : Expression(loc), expr(std::move(e)), op(o) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!expr) throw std::runtime_error("UnaryOpExpr.expr is null"); + auto e = expr->evaluate(context); + switch (op) { + case Op::Plus: return e; + case Op::Minus: return -e; + case Op::LogicalNot: return !e.to_bool(); + case Op::Expansion: + case Op::ExpansionDict: + throw std::runtime_error("Expansion operator is only supported in function calls and collections"); + + } + throw std::runtime_error("Unknown unary operator"); + } +}; + +class BinaryOpExpr : public Expression { +public: + enum class Op { StrConcat, Add, Sub, Mul, MulMul, Div, DivDiv, Mod, Eq, Ne, Lt, Gt, Le, Ge, And, Or, In, NotIn, Is, IsNot }; +private: + std::shared_ptr left; + std::shared_ptr right; + Op op; +public: + BinaryOpExpr(const Location & loc, std::shared_ptr && l, std::shared_ptr && r, Op o) + : Expression(loc), left(std::move(l)), right(std::move(r)), op(o) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!left) throw std::runtime_error("BinaryOpExpr.left is null"); + if (!right) throw std::runtime_error("BinaryOpExpr.right is null"); + auto l = left->evaluate(context); + + auto do_eval = [&](const Value & l) -> Value { + if (op == Op::Is || op == Op::IsNot) { + auto t = dynamic_cast(right.get()); + if (!t) throw std::runtime_error("Right side of 'is' operator must be a variable"); + + auto eval = [&]() { + const auto & name = t->get_name(); + if (name == "none") return l.is_null(); + if (name == "boolean") return l.is_boolean(); + if (name == "integer") return l.is_number_integer(); + if (name == "float") return l.is_number_float(); + if (name == "number") return l.is_number(); + if (name == "string") return l.is_string(); + if (name == "mapping") return l.is_object(); + if (name == "iterable") return l.is_iterable(); + if (name == "sequence") return l.is_array(); + if (name == "defined") return !l.is_null(); + if (name == "true") return l.to_bool(); + if (name == "false") return !l.to_bool(); + throw std::runtime_error("Unknown type for 'is' operator: " + name); + }; + auto value = eval(); + return Value(op == Op::Is ? value : !value); + } + + if (op == Op::And) { + if (!l.to_bool()) return Value(false); + return right->evaluate(context).to_bool(); + } else if (op == Op::Or) { + if (l.to_bool()) return l; + return right->evaluate(context); + } + + auto r = right->evaluate(context); + switch (op) { + case Op::StrConcat: return l.to_str() + r.to_str(); + case Op::Add: return l + r; + case Op::Sub: return l - r; + case Op::Mul: return l * r; + case Op::Div: return l / r; + case Op::MulMul: return std::pow(l.get(), r.get()); + case Op::DivDiv: return l.get() / r.get(); + case Op::Mod: return l.get() % r.get(); + case Op::Eq: return l == r; + case Op::Ne: return l != r; + case Op::Lt: return l < r; + case Op::Gt: return l > r; + case Op::Le: return l <= r; + case Op::Ge: return l >= r; + case Op::In: return (((r.is_array() || r.is_object()) && r.contains(l)) || + (l.is_string() && r.is_string() && + r.to_str().find(l.to_str()) != std::string::npos)); + case Op::NotIn: + return !(((r.is_array() || r.is_object()) && r.contains(l)) || + (l.is_string() && r.is_string() && + r.to_str().find(l.to_str()) != std::string::npos)); + default: break; + } + throw std::runtime_error("Unknown binary operator"); + }; + + if (l.is_callable()) { + return Value::callable([l, do_eval](const std::shared_ptr & context, ArgumentsValue & args) { + auto ll = l.call(context, args); + return do_eval(ll); //args[0].second); + }); + } else { + return do_eval(l); + } + } +}; + +struct ArgumentsExpression { + std::vector> args; + std::vector>> kwargs; + + ArgumentsValue evaluate(const std::shared_ptr & context) const { + ArgumentsValue vargs; + for (const auto& arg : this->args) { + if (auto un_expr = std::dynamic_pointer_cast(arg)) { + if (un_expr->op == UnaryOpExpr::Op::Expansion) { + auto array = un_expr->expr->evaluate(context); + if (!array.is_array()) { + throw std::runtime_error("Expansion operator only supported on arrays"); + } + array.for_each([&](Value & value) { + vargs.args.push_back(value); + }); + continue; + } else if (un_expr->op == UnaryOpExpr::Op::ExpansionDict) { + auto dict = un_expr->expr->evaluate(context); + if (!dict.is_object()) { + throw std::runtime_error("ExpansionDict operator only supported on objects"); + } + dict.for_each([&](const Value & key) { + vargs.kwargs.push_back({key.get(), dict.at(key)}); + }); + continue; + } + } + vargs.args.push_back(arg->evaluate(context)); + } + for (const auto& [name, value] : this->kwargs) { + vargs.kwargs.push_back({name, value->evaluate(context)}); + } + return vargs; + } +}; + +static std::string strip(const std::string & s, const std::string & chars = "", bool left = true, bool right = true) { + auto charset = chars.empty() ? " \t\n\r" : chars; + auto start = left ? s.find_first_not_of(charset) : 0; + if (start == std::string::npos) return ""; + auto end = right ? s.find_last_not_of(charset) : s.size() - 1; + return s.substr(start, end - start + 1); +} + +static std::vector split(const std::string & s, const std::string & sep) { + std::vector result; + size_t start = 0; + size_t end = s.find(sep); + while (end != std::string::npos) { + result.push_back(s.substr(start, end - start)); + start = end + sep.length(); + end = s.find(sep, start); + } + result.push_back(s.substr(start)); + return result; +} + +static std::string capitalize(const std::string & s) { + if (s.empty()) return s; + auto result = s; + result[0] = std::toupper(result[0]); + return result; +} + +static std::string html_escape(const std::string & s) { + std::string result; + result.reserve(s.size()); + for (const auto & c : s) { + switch (c) { + case '&': result += "&"; break; + case '<': result += "<"; break; + case '>': result += ">"; break; + case '"': result += """; break; + case '\'': result += "'"; break; + default: result += c; break; + } + } + return result; +} + +class MethodCallExpr : public Expression { + std::shared_ptr object; + std::shared_ptr method; + ArgumentsExpression args; +public: + MethodCallExpr(const Location & loc, std::shared_ptr && obj, std::shared_ptr && m, ArgumentsExpression && a) + : Expression(loc), object(std::move(obj)), method(std::move(m)), args(std::move(a)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!object) throw std::runtime_error("MethodCallExpr.object is null"); + if (!method) throw std::runtime_error("MethodCallExpr.method is null"); + auto obj = object->evaluate(context); + auto vargs = args.evaluate(context); + if (obj.is_null()) { + throw std::runtime_error("Trying to call method '" + method->get_name() + "' on null"); + } + if (obj.is_array()) { + if (method->get_name() == "append") { + vargs.expectArgs("append method", {1, 1}, {0, 0}); + obj.push_back(vargs.args[0]); + return Value(); + } else if (method->get_name() == "pop") { + vargs.expectArgs("pop method", {0, 1}, {0, 0}); + return obj.pop(vargs.args.empty() ? Value() : vargs.args[0]); + } else if (method->get_name() == "insert") { + vargs.expectArgs("insert method", {2, 2}, {0, 0}); + auto index = vargs.args[0].get(); + if (index < 0 || index > (int64_t) obj.size()) throw std::runtime_error("Index out of range for insert method"); + obj.insert(index, vargs.args[1]); + return Value(); + } + } else if (obj.is_object()) { + if (method->get_name() == "items") { + vargs.expectArgs("items method", {0, 0}, {0, 0}); + auto result = Value::array(); + for (const auto& key : obj.keys()) { + result.push_back(Value::array({key, obj.at(key)})); + } + return result; + } else if (method->get_name() == "pop") { + vargs.expectArgs("pop method", {1, 1}, {0, 0}); + return obj.pop(vargs.args[0]); + } else if (method->get_name() == "get") { + vargs.expectArgs("get method", {1, 2}, {0, 0}); + auto key = vargs.args[0]; + if (vargs.args.size() == 1) { + return obj.contains(key) ? obj.at(key) : Value(); + } else { + return obj.contains(key) ? obj.at(key) : vargs.args[1]; + } + } else if (obj.contains(method->get_name())) { + auto callable = obj.at(method->get_name()); + if (!callable.is_callable()) { + throw std::runtime_error("Property '" + method->get_name() + "' is not callable"); + } + return callable.call(context, vargs); + } + } else if (obj.is_string()) { + auto str = obj.get(); + if (method->get_name() == "strip") { + vargs.expectArgs("strip method", {0, 1}, {0, 0}); + auto chars = vargs.args.empty() ? "" : vargs.args[0].get(); + return Value(strip(str, chars)); + } else if (method->get_name() == "lstrip") { + vargs.expectArgs("lstrip method", {0, 1}, {0, 0}); + auto chars = vargs.args.empty() ? "" : vargs.args[0].get(); + return Value(strip(str, chars, /* left= */ true, /* right= */ false)); + } else if (method->get_name() == "rstrip") { + vargs.expectArgs("rstrip method", {0, 1}, {0, 0}); + auto chars = vargs.args.empty() ? "" : vargs.args[0].get(); + return Value(strip(str, chars, /* left= */ false, /* right= */ true)); + } else if (method->get_name() == "split") { + vargs.expectArgs("split method", {1, 1}, {0, 0}); + auto sep = vargs.args[0].get(); + auto parts = split(str, sep); + Value result = Value::array(); + for (const auto& part : parts) { + result.push_back(Value(part)); + } + return result; + } else if (method->get_name() == "capitalize") { + vargs.expectArgs("capitalize method", {0, 0}, {0, 0}); + return Value(capitalize(str)); + } else if (method->get_name() == "endswith") { + vargs.expectArgs("endswith method", {1, 1}, {0, 0}); + auto suffix = vargs.args[0].get(); + return suffix.length() <= str.length() && std::equal(suffix.rbegin(), suffix.rend(), str.rbegin()); + } else if (method->get_name() == "startswith") { + vargs.expectArgs("startswith method", {1, 1}, {0, 0}); + auto prefix = vargs.args[0].get(); + return prefix.length() <= str.length() && std::equal(prefix.begin(), prefix.end(), str.begin()); + } else if (method->get_name() == "title") { + vargs.expectArgs("title method", {0, 0}, {0, 0}); + auto res = str; + for (size_t i = 0, n = res.size(); i < n; ++i) { + if (i == 0 || std::isspace(res[i - 1])) res[i] = std::toupper(res[i]); + else res[i] = std::tolower(res[i]); + } + return res; + + } else if (method->get_name() == "replace") { + vargs.expectArgs("replace method", {2, 3}, {0, 0}); + auto before = vargs.args[0].get(); + auto after = vargs.args[1].get(); + auto count = vargs.args.size() == 3 ? vargs.args[2].get() + : str.length(); + size_t start_pos = 0; + while ((start_pos = str.find(before, start_pos)) != std::string::npos && + count-- > 0) { + str.replace(start_pos, before.length(), after); + start_pos += after.length(); + } + return str; + } + } + throw std::runtime_error("Unknown method: " + method->get_name()); + } +}; + +class CallExpr : public Expression { +public: + std::shared_ptr object; + ArgumentsExpression args; + CallExpr(const Location & loc, std::shared_ptr && obj, ArgumentsExpression && a) + : Expression(loc), object(std::move(obj)), args(std::move(a)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + if (!object) throw std::runtime_error("CallExpr.object is null"); + auto obj = object->evaluate(context); + if (!obj.is_callable()) { + throw std::runtime_error("Object is not callable: " + obj.dump(2)); + } + auto vargs = args.evaluate(context); + return obj.call(context, vargs); + } +}; + +class FilterExpr : public Expression { + std::vector> parts; +public: + FilterExpr(const Location & loc, std::vector> && p) + : Expression(loc), parts(std::move(p)) {} + Value do_evaluate(const std::shared_ptr & context) const override { + Value result; + bool first = true; + for (const auto& part : parts) { + if (!part) throw std::runtime_error("FilterExpr.part is null"); + if (first) { + first = false; + result = part->evaluate(context); + } else { + if (auto ce = dynamic_cast(part.get())) { + auto target = ce->object->evaluate(context); + ArgumentsValue args = ce->args.evaluate(context); + args.args.insert(args.args.begin(), result); + result = target.call(context, args); + } else { + auto callable = part->evaluate(context); + ArgumentsValue args; + args.args.insert(args.args.begin(), result); + result = callable.call(context, args); + } + } + } + return result; + } + + void prepend(std::shared_ptr && e) { + parts.insert(parts.begin(), std::move(e)); + } +}; + +class Parser { +private: + using CharIterator = std::string::const_iterator; + + std::shared_ptr template_str; + CharIterator start, end, it; + Options options; + + Parser(const std::shared_ptr& template_str, const Options & options) : template_str(template_str), options(options) { + if (!template_str) throw std::runtime_error("Template string is null"); + start = it = this->template_str->begin(); + end = this->template_str->end(); + } + + bool consumeSpaces(SpaceHandling space_handling = SpaceHandling::Strip) { + if (space_handling == SpaceHandling::Strip) { + while (it != end && std::isspace(*it)) ++it; + } + return true; + } + + std::unique_ptr parseString() { + auto doParse = [&](char quote) -> std::unique_ptr { + if (it == end || *it != quote) return nullptr; + std::string result; + bool escape = false; + for (++it; it != end; ++it) { + if (escape) { + escape = false; + switch (*it) { + case 'n': result += '\n'; break; + case 'r': result += '\r'; break; + case 't': result += '\t'; break; + case 'b': result += '\b'; break; + case 'f': result += '\f'; break; + case '\\': result += '\\'; break; + default: + if (*it == quote) { + result += quote; + } else { + result += *it; + } + break; + } + } else if (*it == '\\') { + escape = true; + } else if (*it == quote) { + ++it; + return std::make_unique(std::move(result)); + } else { + result += *it; + } + } + return nullptr; + }; + + consumeSpaces(); + if (it == end) return nullptr; + if (*it == '"') return doParse('"'); + if (*it == '\'') return doParse('\''); + return nullptr; + } + + json parseNumber(CharIterator& it, const CharIterator& end) { + auto before = it; + consumeSpaces(); + auto start = it; + bool hasDecimal = false; + bool hasExponent = false; + + if (it != end && (*it == '-' || *it == '+')) ++it; + + while (it != end) { + if (std::isdigit(*it)) { + ++it; + } else if (*it == '.') { + if (hasDecimal) throw std::runtime_error("Multiple decimal points"); + hasDecimal = true; + ++it; + } else if (it != start && (*it == 'e' || *it == 'E')) { + if (hasExponent) throw std::runtime_error("Multiple exponents"); + hasExponent = true; + ++it; + } else { + break; + } + } + if (start == it) { + it = before; + return json(); // No valid characters found + } + + std::string str(start, it); + try { + return json::parse(str); + } catch (json::parse_error& e) { + throw std::runtime_error("Failed to parse number: '" + str + "' (" + std::string(e.what()) + ")"); + return json(); + } + } + + /** integer, float, bool, string */ + std::shared_ptr parseConstant() { + auto start = it; + consumeSpaces(); + if (it == end) return nullptr; + if (*it == '"' || *it == '\'') { + auto str = parseString(); + if (str) return std::make_shared(*str); + } + static std::regex prim_tok(R"(true\b|True\b|false\b|False\b|None\b)"); + auto token = consumeToken(prim_tok); + if (!token.empty()) { + if (token == "true" || token == "True") return std::make_shared(true); + if (token == "false" || token == "False") return std::make_shared(false); + if (token == "None") return std::make_shared(nullptr); + throw std::runtime_error("Unknown constant token: " + token); + } + + auto number = parseNumber(it, end); + if (!number.is_null()) return std::make_shared(number); + + it = start; + return nullptr; + } + + class expression_parsing_error : public std::runtime_error { + const CharIterator it; + public: + expression_parsing_error(const std::string & message, const CharIterator & it) + : std::runtime_error(message), it(it) {} + size_t get_pos(const CharIterator & begin) const { + return std::distance(begin, it); + } + }; + + bool peekSymbols(const std::vector & symbols) const { + for (const auto & symbol : symbols) { + if (std::distance(it, end) >= (int64_t) symbol.size() && std::string(it, it + symbol.size()) == symbol) { + return true; + } + } + return false; + } + + std::vector consumeTokenGroups(const std::regex & regex, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + std::smatch match; + if (std::regex_search(it, end, match, regex) && match.position() == 0) { + it += match[0].length(); + std::vector ret; + for (size_t i = 0, n = match.size(); i < n; ++i) { + ret.push_back(match[i].str()); + } + return ret; + } + it = start; + return {}; + } + std::string consumeToken(const std::regex & regex, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + std::smatch match; + if (std::regex_search(it, end, match, regex) && match.position() == 0) { + it += match[0].length(); + return match[0].str(); + } + it = start; + return ""; + } + + std::string consumeToken(const std::string & token, SpaceHandling space_handling = SpaceHandling::Strip) { + auto start = it; + consumeSpaces(space_handling); + if (std::distance(it, end) >= (int64_t) token.size() && std::string(it, it + token.size()) == token) { + it += token.size(); + return token; + } + it = start; + return ""; + } + + std::shared_ptr parseExpression(bool allow_if_expr = true) { + auto left = parseLogicalOr(); + if (it == end) return left; + + if (!allow_if_expr) return left; + + static std::regex if_tok(R"(if\b)"); + if (consumeToken(if_tok).empty()) { + return left; + } + + auto location = get_location(); + auto [condition, else_expr] = parseIfExpression(); + return std::make_shared(location, std::move(condition), std::move(left), std::move(else_expr)); + } + + Location get_location() const { + return {template_str, (size_t) std::distance(start, it)}; + } + + std::pair, std::shared_ptr> parseIfExpression() { + auto condition = parseLogicalOr(); + if (!condition) throw std::runtime_error("Expected condition expression"); + + static std::regex else_tok(R"(else\b)"); + std::shared_ptr else_expr; + if (!consumeToken(else_tok).empty()) { + else_expr = parseExpression(); + if (!else_expr) throw std::runtime_error("Expected 'else' expression"); + } + return std::pair(std::move(condition), std::move(else_expr)); + } + + std::shared_ptr parseLogicalOr() { + auto left = parseLogicalAnd(); + if (!left) throw std::runtime_error("Expected left side of 'logical or' expression"); + + static std::regex or_tok(R"(or\b)"); + auto location = get_location(); + while (!consumeToken(or_tok).empty()) { + auto right = parseLogicalAnd(); + if (!right) throw std::runtime_error("Expected right side of 'or' expression"); + left = std::make_shared(location, std::move(left), std::move(right), BinaryOpExpr::Op::Or); + } + return left; + } + + std::shared_ptr parseLogicalNot() { + static std::regex not_tok(R"(not\b)"); + auto location = get_location(); + + if (!consumeToken(not_tok).empty()) { + auto sub = parseLogicalNot(); + if (!sub) throw std::runtime_error("Expected expression after 'not' keyword"); + return std::make_shared(location, std::move(sub), UnaryOpExpr::Op::LogicalNot); + } + return parseLogicalCompare(); + } + + std::shared_ptr parseLogicalAnd() { + auto left = parseLogicalNot(); + if (!left) throw std::runtime_error("Expected left side of 'logical and' expression"); + + static std::regex and_tok(R"(and\b)"); + auto location = get_location(); + while (!consumeToken(and_tok).empty()) { + auto right = parseLogicalNot(); + if (!right) throw std::runtime_error("Expected right side of 'and' expression"); + left = std::make_shared(location, std::move(left), std::move(right), BinaryOpExpr::Op::And); + } + return left; + } + + std::shared_ptr parseLogicalCompare() { + auto left = parseStringConcat(); + if (!left) throw std::runtime_error("Expected left side of 'logical compare' expression"); + + static std::regex compare_tok(R"(==|!=|<=?|>=?|in\b|is\b|not\s+in\b)"); + static std::regex not_tok(R"(not\b)"); + std::string op_str; + while (!(op_str = consumeToken(compare_tok)).empty()) { + auto location = get_location(); + if (op_str == "is") { + auto negated = !consumeToken(not_tok).empty(); + + auto identifier = parseIdentifier(); + if (!identifier) throw std::runtime_error("Expected identifier after 'is' keyword"); + + return std::make_shared( + left->location, + std::move(left), std::move(identifier), + negated ? BinaryOpExpr::Op::IsNot : BinaryOpExpr::Op::Is); + } + auto right = parseStringConcat(); + if (!right) throw std::runtime_error("Expected right side of 'logical compare' expression"); + BinaryOpExpr::Op op; + if (op_str == "==") op = BinaryOpExpr::Op::Eq; + else if (op_str == "!=") op = BinaryOpExpr::Op::Ne; + else if (op_str == "<") op = BinaryOpExpr::Op::Lt; + else if (op_str == ">") op = BinaryOpExpr::Op::Gt; + else if (op_str == "<=") op = BinaryOpExpr::Op::Le; + else if (op_str == ">=") op = BinaryOpExpr::Op::Ge; + else if (op_str == "in") op = BinaryOpExpr::Op::In; + else if (op_str.substr(0, 3) == "not") op = BinaryOpExpr::Op::NotIn; + else throw std::runtime_error("Unknown comparison operator: " + op_str); + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + return left; + } + + Expression::Parameters parseParameters() { + consumeSpaces(); + if (consumeToken("(").empty()) throw std::runtime_error("Expected opening parenthesis in param list"); + + Expression::Parameters result; + + while (it != end) { + if (!consumeToken(")").empty()) { + return result; + } + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in call args"); + + if (auto ident = dynamic_cast(expr.get())) { + if (!consumeToken("=").empty()) { + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected expression in for named arg"); + result.emplace_back(ident->get_name(), std::move(value)); + } else { + result.emplace_back(ident->get_name(), nullptr); + } + } else { + result.emplace_back(std::string(), std::move(expr)); + } + if (consumeToken(",").empty()) { + if (consumeToken(")").empty()) { + throw std::runtime_error("Expected closing parenthesis in call args"); + } + return result; + } + } + throw std::runtime_error("Expected closing parenthesis in call args"); + } + + ArgumentsExpression parseCallArgs() { + consumeSpaces(); + if (consumeToken("(").empty()) throw std::runtime_error("Expected opening parenthesis in call args"); + + ArgumentsExpression result; + + while (it != end) { + if (!consumeToken(")").empty()) { + return result; + } + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in call args"); + + if (auto ident = dynamic_cast(expr.get())) { + if (!consumeToken("=").empty()) { + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected expression in for named arg"); + result.kwargs.emplace_back(ident->get_name(), std::move(value)); + } else { + result.args.emplace_back(std::move(expr)); + } + } else { + result.args.emplace_back(std::move(expr)); + } + if (consumeToken(",").empty()) { + if (consumeToken(")").empty()) { + throw std::runtime_error("Expected closing parenthesis in call args"); + } + return result; + } + } + throw std::runtime_error("Expected closing parenthesis in call args"); + } + + std::shared_ptr parseIdentifier() { + static std::regex ident_regex(R"((?!(?:not|is|and|or|del)\b)[a-zA-Z_]\w*)"); + auto location = get_location(); + auto ident = consumeToken(ident_regex); + if (ident.empty()) + return nullptr; + return std::make_shared(location, ident); + } + + std::shared_ptr parseStringConcat() { + auto left = parseMathPow(); + if (!left) throw std::runtime_error("Expected left side of 'string concat' expression"); + + static std::regex concat_tok(R"(~(?!\}))"); + if (!consumeToken(concat_tok).empty()) { + auto right = parseLogicalAnd(); + if (!right) throw std::runtime_error("Expected right side of 'string concat' expression"); + left = std::make_shared(get_location(), std::move(left), std::move(right), BinaryOpExpr::Op::StrConcat); + } + return left; + } + + std::shared_ptr parseMathPow() { + auto left = parseMathPlusMinus(); + if (!left) throw std::runtime_error("Expected left side of 'math pow' expression"); + + while (!consumeToken("**").empty()) { + auto right = parseMathPlusMinus(); + if (!right) throw std::runtime_error("Expected right side of 'math pow' expression"); + left = std::make_shared(get_location(), std::move(left), std::move(right), BinaryOpExpr::Op::MulMul); + } + return left; + } + + std::shared_ptr parseMathPlusMinus() { + static std::regex plus_minus_tok(R"(\+|-(?![}%#]\}))"); + + auto left = parseMathMulDiv(); + if (!left) throw std::runtime_error("Expected left side of 'math plus/minus' expression"); + std::string op_str; + while (!(op_str = consumeToken(plus_minus_tok)).empty()) { + auto right = parseMathMulDiv(); + if (!right) throw std::runtime_error("Expected right side of 'math plus/minus' expression"); + auto op = op_str == "+" ? BinaryOpExpr::Op::Add : BinaryOpExpr::Op::Sub; + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + return left; + } + + std::shared_ptr parseMathMulDiv() { + auto left = parseMathUnaryPlusMinus(); + if (!left) throw std::runtime_error("Expected left side of 'math mul/div' expression"); + + static std::regex mul_div_tok(R"(\*\*?|//?|%(?!\}))"); + std::string op_str; + while (!(op_str = consumeToken(mul_div_tok)).empty()) { + auto right = parseMathUnaryPlusMinus(); + if (!right) throw std::runtime_error("Expected right side of 'math mul/div' expression"); + auto op = op_str == "*" ? BinaryOpExpr::Op::Mul + : op_str == "**" ? BinaryOpExpr::Op::MulMul + : op_str == "/" ? BinaryOpExpr::Op::Div + : op_str == "//" ? BinaryOpExpr::Op::DivDiv + : BinaryOpExpr::Op::Mod; + left = std::make_shared(get_location(), std::move(left), std::move(right), op); + } + + if (!consumeToken("|").empty()) { + auto expr = parseMathMulDiv(); + if (auto filter = dynamic_cast(expr.get())) { + filter->prepend(std::move(left)); + return expr; + } else { + std::vector> parts; + parts.emplace_back(std::move(left)); + parts.emplace_back(std::move(expr)); + return std::make_shared(get_location(), std::move(parts)); + } + } + return left; + } + + std::shared_ptr call_func(const std::string & name, ArgumentsExpression && args) const { + return std::make_shared(get_location(), std::make_shared(get_location(), name), std::move(args)); + } + + std::shared_ptr parseMathUnaryPlusMinus() { + static std::regex unary_plus_minus_tok(R"(\+|-(?![}%#]\}))"); + auto op_str = consumeToken(unary_plus_minus_tok); + auto expr = parseExpansion(); + if (!expr) throw std::runtime_error("Expected expr of 'unary plus/minus/expansion' expression"); + + if (!op_str.empty()) { + auto op = op_str == "+" ? UnaryOpExpr::Op::Plus : UnaryOpExpr::Op::Minus; + return std::make_shared(get_location(), std::move(expr), op); + } + return expr; + } + + std::shared_ptr parseExpansion() { + static std::regex expansion_tok(R"(\*\*?)"); + auto op_str = consumeToken(expansion_tok); + auto expr = parseValueExpression(); + if (op_str.empty()) return expr; + if (!expr) throw std::runtime_error("Expected expr of 'expansion' expression"); + return std::make_shared(get_location(), std::move(expr), op_str == "*" ? UnaryOpExpr::Op::Expansion : UnaryOpExpr::Op::ExpansionDict); + } + + std::shared_ptr parseValueExpression() { + auto parseValue = [&]() -> std::shared_ptr { + auto location = get_location(); + auto constant = parseConstant(); + if (constant) return std::make_shared(location, *constant); + + static std::regex null_regex(R"(null\b)"); + if (!consumeToken(null_regex).empty()) return std::make_shared(location, Value()); + + auto identifier = parseIdentifier(); + if (identifier) return identifier; + + auto braced = parseBracedExpressionOrArray(); + if (braced) return braced; + + auto array = parseArray(); + if (array) return array; + + auto dictionary = parseDictionary(); + if (dictionary) return dictionary; + + throw std::runtime_error("Expected value expression"); + }; + + auto value = parseValue(); + + while (it != end && consumeSpaces() && peekSymbols({ "[", "." })) { + if (!consumeToken("[").empty()) { + std::shared_ptr index; + auto slice_loc = get_location(); + std::shared_ptr start, end, step; + bool has_first_colon = false, has_second_colon = false; + + + + + + + + + + + + + + + + + + + + + + + + + + + + if (!peekSymbols({ ":" })) { + start = parseExpression(); + } + + + + + if (!consumeToken(":").empty()) { + has_first_colon = true; + if (!peekSymbols({ ":", "]" })) { + end = parseExpression(); + } + if (!consumeToken(":").empty()) { + has_second_colon = true; + if (!peekSymbols({ "]" })) { + step = parseExpression(); + } + } + } + + if ((has_first_colon || has_second_colon)) { + index = std::make_shared(slice_loc, std::move(start), std::move(end), std::move(step)); + } else { + index = std::move(start); + } + if (!index) throw std::runtime_error("Empty index in subscript"); + if (consumeToken("]").empty()) throw std::runtime_error("Expected closing bracket in subscript"); + + value = std::make_shared(value->location, std::move(value), std::move(index)); + } else if (!consumeToken(".").empty()) { + auto identifier = parseIdentifier(); + if (!identifier) throw std::runtime_error("Expected identifier in subscript"); + + consumeSpaces(); + if (peekSymbols({ "(" })) { + auto callParams = parseCallArgs(); + value = std::make_shared(identifier->location, std::move(value), std::move(identifier), std::move(callParams)); + } else { + auto key = std::make_shared(identifier->location, Value(identifier->get_name())); + value = std::make_shared(identifier->location, std::move(value), std::move(key)); + } + } + consumeSpaces(); + } + + if (peekSymbols({ "(" })) { + auto location = get_location(); + auto callParams = parseCallArgs(); + value = std::make_shared(location, std::move(value), std::move(callParams)); + } + return value; + } + + std::shared_ptr parseBracedExpressionOrArray() { + if (consumeToken("(").empty()) return nullptr; + + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in braced expression"); + + if (!consumeToken(")").empty()) { + return expr; // Drop the parentheses + } + + std::vector> tuple; + tuple.emplace_back(std::move(expr)); + + while (it != end) { + if (consumeToken(",").empty()) throw std::runtime_error("Expected comma in tuple"); + auto next = parseExpression(); + if (!next) throw std::runtime_error("Expected expression in tuple"); + tuple.push_back(std::move(next)); + + if (!consumeToken(")").empty()) { + return std::make_shared(get_location(), std::move(tuple)); + } + } + throw std::runtime_error("Expected closing parenthesis"); + } + + std::shared_ptr parseArray() { + if (consumeToken("[").empty()) return nullptr; + + std::vector> elements; + if (!consumeToken("]").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } + auto first_expr = parseExpression(); + if (!first_expr) throw std::runtime_error("Expected first expression in array"); + elements.push_back(std::move(first_expr)); + + while (it != end) { + if (!consumeToken(",").empty()) { + auto expr = parseExpression(); + if (!expr) throw std::runtime_error("Expected expression in array"); + elements.push_back(std::move(expr)); + } else if (!consumeToken("]").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } else { + throw std::runtime_error("Expected comma or closing bracket in array"); + } + } + throw std::runtime_error("Expected closing bracket"); + } + + std::shared_ptr parseDictionary() { + if (consumeToken("{").empty()) return nullptr; + + std::vector, std::shared_ptr>> elements; + if (!consumeToken("}").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } + + auto parseKeyValuePair = [&]() { + auto key = parseExpression(); + if (!key) throw std::runtime_error("Expected key in dictionary"); + if (consumeToken(":").empty()) throw std::runtime_error("Expected colon betweek key & value in dictionary"); + auto value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in dictionary"); + elements.emplace_back(std::pair(std::move(key), std::move(value))); + }; + + parseKeyValuePair(); + + while (it != end) { + if (!consumeToken(",").empty()) { + parseKeyValuePair(); + } else if (!consumeToken("}").empty()) { + return std::make_shared(get_location(), std::move(elements)); + } else { + throw std::runtime_error("Expected comma or closing brace in dictionary"); + } + } + throw std::runtime_error("Expected closing brace"); + } + + SpaceHandling parsePreSpace(const std::string& s) const { + if (s == "-") + return SpaceHandling::Strip; + return SpaceHandling::Keep; + } + + SpaceHandling parsePostSpace(const std::string& s) const { + if (s == "-") return SpaceHandling::Strip; + return SpaceHandling::Keep; + } + + using TemplateTokenVector = std::vector>; + using TemplateTokenIterator = TemplateTokenVector::const_iterator; + + std::vector parseVarNames() { + static std::regex varnames_regex(R"(((?:\w+)(?:\s*,\s*(?:\w+))*)\s*)"); + + std::vector group; + if ((group = consumeTokenGroups(varnames_regex)).empty()) throw std::runtime_error("Expected variable names"); + std::vector varnames; + std::istringstream iss(group[1]); + std::string varname; + while (std::getline(iss, varname, ',')) { + varnames.push_back(strip(varname)); + } + return varnames; + } + + std::runtime_error unexpected(const TemplateToken & token) const { + return std::runtime_error("Unexpected " + TemplateToken::typeToString(token.type) + + error_location_suffix(*template_str, token.location.pos)); + } + std::runtime_error unterminated(const TemplateToken & token) const { + return std::runtime_error("Unterminated " + TemplateToken::typeToString(token.type) + + error_location_suffix(*template_str, token.location.pos)); + } + + TemplateTokenVector tokenize() { + static std::regex comment_tok(R"(\{#([-~]?)([\s\S]*?)([-~]?)#\})"); + static std::regex expr_open_regex(R"(\{\{([-~])?)"); + static std::regex block_open_regex(R"(^\{%([-~])?\s*)"); + static std::regex block_keyword_tok(R"((if|else|elif|endif|for|endfor|generation|endgeneration|set|endset|block|endblock|macro|endmacro|filter|endfilter|break|continue)\b)"); + static std::regex non_text_open_regex(R"(\{\{|\{%|\{#)"); + static std::regex expr_close_regex(R"(\s*([-~])?\}\})"); + static std::regex block_close_regex(R"(\s*([-~])?%\})"); + + TemplateTokenVector tokens; + std::vector group; + std::string text; + std::smatch match; + + try { + while (it != end) { + auto location = get_location(); + + if (!(group = consumeTokenGroups(comment_tok, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + auto content = group[2]; + auto post_space = parsePostSpace(group[3]); + tokens.push_back(std::make_unique(location, pre_space, post_space, content)); + } else if (!(group = consumeTokenGroups(expr_open_regex, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + auto expr = parseExpression(); + + if ((group = consumeTokenGroups(expr_close_regex)).empty()) { + throw std::runtime_error("Expected closing expression tag"); + } + + auto post_space = parsePostSpace(group[1]); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(expr))); + } else if (!(group = consumeTokenGroups(block_open_regex, SpaceHandling::Keep)).empty()) { + auto pre_space = parsePreSpace(group[1]); + + std::string keyword; + + auto parseBlockClose = [&]() -> SpaceHandling { + if ((group = consumeTokenGroups(block_close_regex)).empty()) throw std::runtime_error("Expected closing block tag"); + return parsePostSpace(group[1]); + }; + + if ((keyword = consumeToken(block_keyword_tok)).empty()) throw std::runtime_error("Expected block keyword"); + + if (keyword == "if") { + auto condition = parseExpression(); + if (!condition) throw std::runtime_error("Expected condition in if block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(condition))); + } else if (keyword == "elif") { + auto condition = parseExpression(); + if (!condition) throw std::runtime_error("Expected condition in elif block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(condition))); + } else if (keyword == "else") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "endif") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "for") { + static std::regex recursive_tok(R"(recursive\b)"); + static std::regex if_tok(R"(if\b)"); + + auto varnames = parseVarNames(); + static std::regex in_tok(R"(in\b)"); + if (consumeToken(in_tok).empty()) throw std::runtime_error("Expected 'in' keyword in for block"); + auto iterable = parseExpression(/* allow_if_expr = */ false); + if (!iterable) throw std::runtime_error("Expected iterable in for block"); + + std::shared_ptr condition; + if (!consumeToken(if_tok).empty()) { + condition = parseExpression(); + } + auto recursive = !consumeToken(recursive_tok).empty(); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(varnames), std::move(iterable), std::move(condition), recursive)); + } else if (keyword == "endfor") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "generation") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "endgeneration") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "set") { + static std::regex namespaced_var_regex(R"((\w+)\s*\.\s*(\w+))"); + + std::string ns; + std::vector var_names; + std::shared_ptr value; + if (!(group = consumeTokenGroups(namespaced_var_regex)).empty()) { + ns = group[1]; + var_names.push_back(group[2]); + + if (consumeToken("=").empty()) throw std::runtime_error("Expected equals sign in set block"); + + value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in set block"); + } else { + var_names = parseVarNames(); + + if (!consumeToken("=").empty()) { + value = parseExpression(); + if (!value) throw std::runtime_error("Expected value in set block"); + } + } + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, ns, var_names, std::move(value))); + } else if (keyword == "endset") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "macro") { + auto macroname = parseIdentifier(); + if (!macroname) throw std::runtime_error("Expected macro name in macro block"); + auto params = parseParameters(); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(macroname), std::move(params))); + } else if (keyword == "endmacro") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "filter") { + auto filter = parseExpression(); + if (!filter) throw std::runtime_error("Expected expression in filter block"); + + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, std::move(filter))); + } else if (keyword == "endfilter") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space)); + } else if (keyword == "break" || keyword == "continue") { + auto post_space = parseBlockClose(); + tokens.push_back(std::make_unique(location, pre_space, post_space, keyword == "break" ? LoopControlType::Break : LoopControlType::Continue)); + } else { + throw std::runtime_error("Unexpected block: " + keyword); + } + } else if (std::regex_search(it, end, match, non_text_open_regex)) { + if (!match.position()) { + if (match[0] != "{#") + throw std::runtime_error("Internal error: Expected a comment"); + throw std::runtime_error("Missing end of comment tag"); + } + auto text_end = it + match.position(); + text = std::string(it, text_end); + it = text_end; + tokens.push_back(std::make_unique(location, SpaceHandling::Keep, SpaceHandling::Keep, text)); + } else { + text = std::string(it, end); + it = end; + tokens.push_back(std::make_unique(location, SpaceHandling::Keep, SpaceHandling::Keep, text)); + } + } + return tokens; + } catch (const std::exception & e) { + throw std::runtime_error(e.what() + error_location_suffix(*template_str, std::distance(start, it))); + } + } + + std::shared_ptr parseTemplate( + const TemplateTokenIterator & begin, + TemplateTokenIterator & it, + const TemplateTokenIterator & end, + bool fully = false) const { + std::vector> children; + while (it != end) { + const auto start = it; + const auto & token = *(it++); + if (auto if_token = dynamic_cast(token.get())) { + std::vector, std::shared_ptr>> cascade; + cascade.emplace_back(std::move(if_token->condition), parseTemplate(begin, it, end)); + + while (it != end && (*it)->type == TemplateToken::Type::Elif) { + auto elif_token = dynamic_cast((*(it++)).get()); + cascade.emplace_back(std::move(elif_token->condition), parseTemplate(begin, it, end)); + } + + if (it != end && (*it)->type == TemplateToken::Type::Else) { + cascade.emplace_back(nullptr, parseTemplate(begin, ++it, end)); + } + if (it == end || (*(it++))->type != TemplateToken::Type::EndIf) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(cascade))); + } else if (auto for_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + auto else_body = std::shared_ptr(); + if (it != end && (*it)->type == TemplateToken::Type::Else) { + else_body = parseTemplate(begin, ++it, end); + } + if (it == end || (*(it++))->type != TemplateToken::Type::EndFor) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(for_token->var_names), std::move(for_token->iterable), std::move(for_token->condition), std::move(body), for_token->recursive, std::move(else_body))); + } else if (dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndGeneration) { + throw unterminated(**start); + } + // Treat as a no-op, as our scope is templates for inference, not training (`{% generation %}` wraps generated tokens for masking). + children.emplace_back(std::move(body)); + } else if (auto text_token = dynamic_cast(token.get())) { + SpaceHandling pre_space = (it - 1) != begin ? (*(it - 2))->post_space : SpaceHandling::Keep; + SpaceHandling post_space = it != end ? (*it)->pre_space : SpaceHandling::Keep; + + auto text = text_token->text; + if (post_space == SpaceHandling::Strip) { + static std::regex trailing_space_regex(R"(\s+$)"); + text = std::regex_replace(text, trailing_space_regex, ""); + } else if (options.lstrip_blocks && it != end) { + auto i = text.size(); + while (i > 0 && (text[i - 1] == ' ' || text[i - 1] == '\t')) i--; + if ((i == 0 && (it - 1) == begin) || (i > 0 && text[i - 1] == '\n')) { + text.resize(i); + } + } + if (pre_space == SpaceHandling::Strip) { + static std::regex leading_space_regex(R"(^\s+)"); + text = std::regex_replace(text, leading_space_regex, ""); + } else if (options.trim_blocks && (it - 1) != begin && !dynamic_cast((*(it - 2)).get())) { + if (!text.empty() && text[0] == '\n') { + text.erase(0, 1); + } + } + if (it == end && !options.keep_trailing_newline) { + auto i = text.size(); + if (i > 0 && text[i - 1] == '\n') { + i--; + if (i > 0 && text[i - 1] == '\r') i--; + text.resize(i); + } + } + children.emplace_back(std::make_shared(token->location, text)); + } else if (auto expr_token = dynamic_cast(token.get())) { + children.emplace_back(std::make_shared(token->location, std::move(expr_token->expr))); + } else if (auto set_token = dynamic_cast(token.get())) { + if (set_token->value) { + children.emplace_back(std::make_shared(token->location, set_token->ns, set_token->var_names, std::move(set_token->value))); + } else { + auto value_template = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndSet) { + throw unterminated(**start); + } + if (!set_token->ns.empty()) throw std::runtime_error("Namespaced set not supported in set with template value"); + if (set_token->var_names.size() != 1) throw std::runtime_error("Structural assignment not supported in set with template value"); + auto & name = set_token->var_names[0]; + children.emplace_back(std::make_shared(token->location, name, std::move(value_template))); + } + } else if (auto macro_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndMacro) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(macro_token->name), std::move(macro_token->params), std::move(body))); + } else if (auto filter_token = dynamic_cast(token.get())) { + auto body = parseTemplate(begin, it, end); + if (it == end || (*(it++))->type != TemplateToken::Type::EndFilter) { + throw unterminated(**start); + } + children.emplace_back(std::make_shared(token->location, std::move(filter_token->filter), std::move(body))); + } else if (dynamic_cast(token.get())) { + // Ignore comments + } else if (auto ctrl_token = dynamic_cast(token.get())) { + children.emplace_back(std::make_shared(token->location, ctrl_token->control_type)); + } else if (dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get()) + || dynamic_cast(token.get())) { + it--; // unconsume the token + break; // exit the loop + } else { + throw unexpected(**(it-1)); + } + } + if (fully && it != end) { + throw unexpected(**it); + } + if (children.empty()) { + return std::make_shared(Location { template_str, 0 }, std::string()); + } else if (children.size() == 1) { + return std::move(children[0]); + } else { + return std::make_shared(children[0]->location(), std::move(children)); + } + } + +public: + + static std::shared_ptr parse(const std::string& template_str, const Options & options) { + Parser parser(std::make_shared(normalize_newlines(template_str)), options); + auto tokens = parser.tokenize(); + TemplateTokenIterator begin = tokens.begin(); + auto it = begin; + TemplateTokenIterator end = tokens.end(); + return parser.parseTemplate(begin, it, end, /* fully= */ true); + } +}; + +static Value simple_function(const std::string & fn_name, const std::vector & params, const std::function &, Value & args)> & fn) { + std::map named_positions; + for (size_t i = 0, n = params.size(); i < n; i++) named_positions[params[i]] = i; + + return Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) -> Value { + auto args_obj = Value::object(); + std::vector provided_args(params.size()); + for (size_t i = 0, n = args.args.size(); i < n; i++) { + auto & arg = args.args[i]; + if (i < params.size()) { + args_obj.set(params[i], arg); + provided_args[i] = true; + } else { + throw std::runtime_error("Too many positional params for " + fn_name); + } + } + for (auto & [name, value] : args.kwargs) { + auto named_pos_it = named_positions.find(name); + if (named_pos_it == named_positions.end()) { + throw std::runtime_error("Unknown argument " + name + " for function " + fn_name); + } + provided_args[named_pos_it->second] = true; + args_obj.set(name, value); + } + return fn(context, args_obj); + }); +} + +inline std::shared_ptr Context::builtins() { + auto globals = Value::object(); + + globals.set("raise_exception", simple_function("raise_exception", { "message" }, [](const std::shared_ptr &, Value & args) -> Value { + throw std::runtime_error(args.at("message").get()); + })); + globals.set("tojson", simple_function("tojson", { "value", "indent" }, [](const std::shared_ptr &, Value & args) { + return Value(args.at("value").dump(args.get("indent", -1), /* to_json= */ true)); + })); + globals.set("items", simple_function("items", { "object" }, [](const std::shared_ptr &, Value & args) { + auto items = Value::array(); + if (args.contains("object")) { + auto & obj = args.at("object"); + if (obj.is_string()) { + auto json_obj = json::parse(obj.get()); + for (const auto & kv : json_obj.items()) { + items.push_back(Value::array({kv.key(), kv.value()})); + } + } else if (!obj.is_null()) { + for (auto & key : obj.keys()) { + items.push_back(Value::array({key, obj.at(key)})); + } + } + } + return items; + })); + globals.set("last", simple_function("last", { "items" }, [](const std::shared_ptr &, Value & args) { + auto items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not a list"); + if (items.empty()) return Value(); + return items.at(items.size() - 1); + })); + globals.set("trim", simple_function("trim", { "text" }, [](const std::shared_ptr &, Value & args) { + auto & text = args.at("text"); + return text.is_null() ? text : Value(strip(text.get())); + })); + + + + + + + + + auto char_transform_function = [](const std::string & name, const std::function & fn) { + return simple_function(name, { "text" }, [=](const std::shared_ptr &, Value & args) { + auto text = args.at("text"); + if (text.is_null()) return text; + std::string res; + auto str = text.get(); + std::transform(str.begin(), str.end(), std::back_inserter(res), fn); + return Value(res); + }); + }; + globals.set("lower", char_transform_function("lower", ::tolower)); + globals.set("upper", char_transform_function("upper", ::toupper)); + globals.set("default", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + args.expectArgs("default", {2, 3}, {0, 1}); + auto & value = args.args[0]; + auto & default_value = args.args[1]; + bool boolean = false; + if (args.args.size() == 3) { + boolean = args.args[2].get(); + } else { + Value bv = args.get_named("boolean"); + if (!bv.is_null()) { + boolean = bv.get(); + } + } + return boolean ? (value.to_bool() ? value : default_value) : value.is_null() ? default_value : value; + })); + auto escape = simple_function("escape", { "text" }, [](const std::shared_ptr &, Value & args) { + return Value(html_escape(args.at("text").get())); + }); + globals.set("e", escape); + globals.set("escape", escape); + globals.set("joiner", simple_function("joiner", { "sep" }, [](const std::shared_ptr &, Value & args) { + auto sep = args.get("sep", ""); + auto first = std::make_shared(true); + return simple_function("", {}, [sep, first](const std::shared_ptr &, const Value &) -> Value { + if (*first) { + *first = false; + return ""; + } + return sep; + }); + return Value(html_escape(args.at("text").get())); + })); + globals.set("count", simple_function("count", { "items" }, [](const std::shared_ptr &, Value & args) { + return Value((int64_t) args.at("items").size()); + })); + globals.set("dictsort", simple_function("dictsort", { "value" }, [](const std::shared_ptr &, Value & args) { + if (args.size() != 1) throw std::runtime_error("dictsort expects exactly 1 argument (TODO: fix implementation)"); + auto & value = args.at("value"); + auto keys = value.keys(); + std::sort(keys.begin(), keys.end()); + auto res = Value::array(); + for (auto & key : keys) { + res.push_back(Value::array({key, value.at(key)})); + } + return res; + })); + globals.set("join", simple_function("join", { "items", "d" }, [](const std::shared_ptr &, Value & args) { + auto do_join = [](Value & items, const std::string & sep) { + if (!items.is_array()) throw std::runtime_error("object is not iterable: " + items.dump()); + std::ostringstream oss; + auto first = true; + for (size_t i = 0, n = items.size(); i < n; ++i) { + if (first) first = false; + else oss << sep; + oss << items.at(i).to_str(); + } + return Value(oss.str()); + }; + auto sep = args.get("d", ""); + if (args.contains("items")) { + auto & items = args.at("items"); + return do_join(items, sep); + } else { + return simple_function("", {"items"}, [sep, do_join](const std::shared_ptr &, Value & args) { + auto & items = args.at("items"); + if (!items.to_bool() || !items.is_array()) throw std::runtime_error("join expects an array for items, got: " + items.dump()); + return do_join(items, sep); + }); + } + })); + globals.set("namespace", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + auto ns = Value::object(); + args.expectArgs("namespace", {0, 0}, {0, (std::numeric_limits::max)()}); + for (auto & [name, value] : args.kwargs) { + ns.set(name, value); + } + return ns; + })); + auto equalto = simple_function("equalto", { "expected", "actual" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("actual") == args.at("expected"); + }); + globals.set("equalto", equalto); + globals.set("==", equalto); + globals.set("length", simple_function("length", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + return (int64_t) items.size(); + })); + globals.set("safe", simple_function("safe", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_str(); + })); + globals.set("string", simple_function("string", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_str(); + })); + globals.set("int", simple_function("int", { "value" }, [](const std::shared_ptr &, Value & args) -> Value { + return args.at("value").to_int(); + })); + globals.set("list", simple_function("list", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not iterable"); + return items; + })); + globals.set("unique", simple_function("unique", { "items" }, [](const std::shared_ptr &, Value & args) -> Value { + auto & items = args.at("items"); + if (!items.is_array()) throw std::runtime_error("object is not iterable"); + std::unordered_set seen; + auto result = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto pair = seen.insert(items.at(i)); + if (pair.second) { + result.push_back(items.at(i)); + } + } + return result; + })); + auto make_filter = [](const Value & filter, Value & extra_args) -> Value { + return simple_function("", { "value" }, [=](const std::shared_ptr & context, Value & args) { + auto & value = args.at("value"); + ArgumentsValue actual_args; + actual_args.args.emplace_back(value); + for (size_t i = 0, n = extra_args.size(); i < n; i++) { + actual_args.args.emplace_back(extra_args.at(i)); + } + return filter.call(context, actual_args); + }); + }; + auto select_or_reject = [make_filter](bool is_select) { + return Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + args.expectArgs(is_select ? "select" : "reject", {2, (std::numeric_limits::max)()}, {0, 0}); + auto & items = args.args[0]; + if (items.is_null()) { + return Value::array(); + } + if (!items.is_array()) { + throw std::runtime_error("object is not iterable: " + items.dump()); + } + + auto filter_fn = context->get(args.args[1]); + if (filter_fn.is_null()) { + throw std::runtime_error("Undefined filter: " + args.args[1].dump()); + } + + + auto filter_args = Value::array(); + for (size_t i = 2, n = args.args.size(); i < n; i++) { + + + filter_args.push_back(args.args[i]); + + + + } + auto filter = make_filter(filter_fn, filter_args); + + auto res = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + ArgumentsValue filter_args; + filter_args.args.emplace_back(item); + auto pred_res = filter.call(context, filter_args); + if (pred_res.to_bool() == (is_select ? true : false)) { + res.push_back(item); + } + } + return res; + }); + }; + globals.set("select", select_or_reject(/* is_select= */ true)); + globals.set("reject", select_or_reject(/* is_select= */ false)); + globals.set("map", Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + auto res = Value::array(); + if (args.args.size() == 1 && + ((args.has_named("attribute") && args.kwargs.size() == 1) || (args.has_named("default") && args.kwargs.size() == 2))) { + auto & items = args.args[0]; + auto attr_name = args.get_named("attribute"); + auto default_value = args.get_named("default"); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + auto attr = item.get(attr_name); + res.push_back(attr.is_null() ? default_value : attr); + } + } else if (args.kwargs.empty() && args.args.size() >= 2) { + auto fn = context->get(args.args[1]); + if (fn.is_null()) throw std::runtime_error("Undefined filter: " + args.args[1].dump()); + ArgumentsValue filter_args { {Value()}, {} }; + for (size_t i = 2, n = args.args.size(); i < n; i++) { + filter_args.args.emplace_back(args.args[i]); + } + for (size_t i = 0, n = args.args[0].size(); i < n; i++) { + auto & item = args.args[0].at(i); + filter_args.args[0] = item; + res.push_back(fn.call(context, filter_args)); + } + } else { + throw std::runtime_error("Invalid or unsupported arguments for map"); + } + return res; + })); + globals.set("indent", simple_function("indent", { "text", "indent", "first" }, [](const std::shared_ptr &, Value & args) { + auto text = args.at("text").get(); + auto first = args.get("first", false); + std::string out; + std::string indent(args.get("indent", 0), ' '); + std::istringstream iss(text); + std::string line; + auto is_first = true; + while (std::getline(iss, line, '\n')) { + auto needs_indent = !is_first || first; + if (is_first) is_first = false; + else out += "\n"; + if (needs_indent) out += indent; + out += line; + } + if (!text.empty() && text.back() == '\n') out += "\n"; + return out; + })); + auto select_or_reject_attr = [](bool is_select) { + return Value::callable([=](const std::shared_ptr & context, ArgumentsValue & args) { + args.expectArgs(is_select ? "selectattr" : "rejectattr", {2, (std::numeric_limits::max)()}, {0, 0}); + auto & items = args.args[0]; + if (items.is_null()) + return Value::array(); + if (!items.is_array()) throw std::runtime_error("object is not iterable: " + items.dump()); + auto attr_name = args.args[1].get(); + + bool has_test = false; + Value test_fn; + ArgumentsValue test_args {{Value()}, {}}; + if (args.args.size() >= 3) { + has_test = true; + test_fn = context->get(args.args[2]); + if (test_fn.is_null()) throw std::runtime_error("Undefined test: " + args.args[2].dump()); + for (size_t i = 3, n = args.args.size(); i < n; i++) { + test_args.args.emplace_back(args.args[i]); + } + test_args.kwargs = args.kwargs; + } + + + + auto res = Value::array(); + for (size_t i = 0, n = items.size(); i < n; i++) { + auto & item = items.at(i); + auto attr = item.get(attr_name); + if (has_test) { + test_args.args[0] = attr; + if (test_fn.call(context, test_args).to_bool() == (is_select ? true : false)) { + res.push_back(item); + } + } else { + res.push_back(attr); + } + + + } + + return res; + }); + }; + globals.set("selectattr", select_or_reject_attr(/* is_select= */ true)); + globals.set("rejectattr", select_or_reject_attr(/* is_select= */ false)); + globals.set("range", Value::callable([=](const std::shared_ptr &, ArgumentsValue & args) { + std::vector startEndStep(3); + std::vector param_set(3); + if (args.args.size() == 1) { + startEndStep[1] = args.args[0].get(); + param_set[1] = true; + } else { + for (size_t i = 0; i < args.args.size(); i++) { + auto & arg = args.args[i]; + auto v = arg.get(); + startEndStep[i] = v; + param_set[i] = true; + + } + } + for (auto & [name, value] : args.kwargs) { + size_t i; + if (name == "start") { + i = 0; + } else if (name == "end") { + i = 1; + } else if (name == "step") { + i = 2; + } else { + throw std::runtime_error("Unknown argument " + name + " for function range"); + } + + if (param_set[i]) { + throw std::runtime_error("Duplicate argument " + name + " for function range"); + } + startEndStep[i] = value.get(); + param_set[i] = true; + } + if (!param_set[1]) { + throw std::runtime_error("Missing required argument 'end' for function range"); + } + int64_t start = param_set[0] ? startEndStep[0] : 0; + int64_t end = startEndStep[1]; + int64_t step = param_set[2] ? startEndStep[2] : 1; + + auto res = Value::array(); + if (step > 0) { + for (int64_t i = start; i < end; i += step) { + res.push_back(Value(i)); + } + } else { + for (int64_t i = start; i > end; i += step) { + res.push_back(Value(i)); + } + } + return res; + })); + + return std::make_shared(std::move(globals)); +} + +inline std::shared_ptr Context::make(Value && values, const std::shared_ptr & parent) { + return std::make_shared(values.is_null() ? Value::object() : std::move(values), parent); +} + +} // namespace minja diff --git a/examples/main/main.cpp b/examples/main/main.cpp index de736f082..5bf8fca05 100644 --- a/examples/main/main.cpp +++ b/examples/main/main.cpp @@ -2,7 +2,7 @@ #include "console.h" #include "llama.h" - +#include "chat-template.hpp" #include #include #include @@ -119,10 +119,10 @@ static void llama_log_callback_logTee(ggml_log_level level, const char * text, v LOG_TEE("%s", text); } -static std::string chat_add_and_format(struct llama_model * model, std::vector & chat_msgs, std::string role, std::string content) { +static std::string chat_add_and_format(struct llama_model * model, common_chat_templates &chat_templates, std::vector & chat_msgs, std::string role, std::string content) { llama_chat_msg new_msg{role, content}; - auto formatted = llama_chat_format_single( - model, g_params->chat_template, chat_msgs, new_msg, role == "user"); + auto formatted = llama_chat_format_single(model, + *chat_templates.template_default, chat_msgs, new_msg, role == "user", g_params->use_jinja); chat_msgs.push_back({role, content}); LOG("formatted: %s\n", formatted.c_str()); return formatted; @@ -220,6 +220,7 @@ int main(int argc, char ** argv) { LOG_TEE("%s: error: unable to load model\n", __func__); return 1; } + auto chat_templates = llama_chat_templates_from_model(model, params.chat_template); const int n_ctx_train = llama_n_ctx_train(model); const int n_ctx = llama_n_ctx(ctx); @@ -229,11 +230,10 @@ int main(int argc, char ** argv) { LOG_TEE("%s: warning: model was trained on only %d context tokens (%d specified)\n", __func__, n_ctx_train, n_ctx); } - // print chat template example in conversation mode if (params.conversation) { if (params.enable_chat_template) { - LOG_TEE("%s: chat template example: %s\n", __func__, llama_chat_format_example(model, params.chat_template).c_str()); + LOG_TEE("%s: chat template example: %s\n", __func__, llama_chat_format_example(model, *chat_templates.template_default, params.use_jinja).c_str()); } else { LOG_TEE("%s: in-suffix/prefix is specified, chat template will be disabled\n", __func__); } @@ -274,11 +274,29 @@ int main(int argc, char ** argv) { LOG("add_bos: %d\n", add_bos); std::vector embd_inp; + bool waiting_for_first_input = params.conversation && params.enable_chat_template && params.system_prompt.empty(); { - auto prompt = (params.conversation && params.enable_chat_template && !params.prompt.empty()) - ? chat_add_and_format(model, chat_msgs, "system", params.prompt) // format the system prompt in conversation mode - : params.prompt; + //auto prompt = (params.conversation && params.enable_chat_template && !params.prompt.empty()) + // ? chat_add_and_format(model, chat_templates,chat_msgs, "system", params.prompt) // format the system prompt in conversation mode + // : params.prompt; + std::string prompt; + + if (params.conversation && params.enable_chat_template) { + // format the system prompt in conversation mode (will use template default if empty) + prompt = params.system_prompt; + + if (!prompt.empty()) { + prompt = chat_add_and_format(model, chat_templates,chat_msgs, "system", prompt); + } + } + else { + // otherwise use the prompt as is + prompt = params.prompt; + } + + + if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) { LOG("tokenize the prompt\n"); embd_inp = ::llama_tokenize(ctx, prompt, true, true); @@ -292,7 +310,7 @@ int main(int argc, char ** argv) { } // Should not run without any tokens - if (embd_inp.empty()) { + if (!params.conversation && embd_inp.empty()) { if (add_bos) { embd_inp.push_back(llama_token_bos(model)); LOG("embd_inp was considered empty and bos was added: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, embd_inp).c_str()); @@ -837,7 +855,7 @@ int main(int argc, char ** argv) { } // deal with end of generation tokens in interactive mode - if (llama_token_is_eog(model, llama_sampling_last(ctx_sampling))) { + if (!waiting_for_first_input && llama_token_is_eog(model, llama_sampling_last(ctx_sampling))) { LOG("found an EOG token\n"); if (params.interactive) { @@ -849,7 +867,7 @@ int main(int argc, char ** argv) { } if (params.enable_chat_template) { - chat_add_and_format(model, chat_msgs, "assistant", assistant_ss.str()); + chat_add_and_format(model, chat_templates, chat_msgs, "assistant", assistant_ss.str()); } is_interacting = true; printf("\n"); @@ -857,12 +875,12 @@ int main(int argc, char ** argv) { } // if current token is not EOG, we add it to current assistant message - if (params.conversation) { + if (params.conversation && !waiting_for_first_input) { auto id = llama_sampling_last(ctx_sampling); assistant_ss << llama_token_to_piece(ctx, id, false); } - if (n_past > 0 && is_interacting) { + if ((n_past > 0 || waiting_for_first_input) && is_interacting) { LOG("waiting for user input\n"); if (params.conversation) { @@ -914,7 +932,7 @@ int main(int argc, char ** argv) { bool format_chat = params.conversation && params.enable_chat_template; std::string user_inp = format_chat - ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer)) + ? chat_add_and_format(model, chat_templates, chat_msgs, "user", std::move(buffer)) : std::move(buffer); // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix) const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true); @@ -952,11 +970,12 @@ int main(int argc, char ** argv) { input_echo = false; // do not echo this again } - if (n_past > 0) { + if (n_past > 0 || waiting_for_first_input) { if (is_interacting) { llama_sampling_reset(ctx_sampling); } is_interacting = false; + waiting_for_first_input = false; } } diff --git a/examples/server/server.cpp b/examples/server/server.cpp index d9f427a22..511a1515b 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -814,6 +814,7 @@ struct server_context { server_metrics metrics; + common_chat_templates chat_templates; // Necessary similarity of prompt for slot selection float slot_prompt_similarity = 0.0f; @@ -860,15 +861,47 @@ struct server_context { add_bos_token = llama_should_add_bos_token(model); GGML_ASSERT(llama_add_eos_token(model) != 1); + if (params.chat_template.empty() && !validate_model_chat_template(params.use_jinja)) { + LOG_WARNING("%s: The chat template that comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses\n", __func__); + chat_templates = llama_chat_templates_from_model(model, "chatml"); + } + else { + chat_templates = llama_chat_templates_from_model(model, params.chat_template); + } + GGML_ASSERT(chat_templates.template_default.get() != nullptr); return true; } - bool validate_model_chat_template() const { + bool validate_model_chat_template(bool use_jinja) const { llama_chat_message chat[] = {{"user", "test"}}; - const int res = llama_chat_apply_template(model, nullptr, chat, 1, true, nullptr, 0); - return res > 0; + if (use_jinja) { + auto templates = llama_chat_templates_from_model(model, ""); + GGML_ASSERT(templates.template_default); + try { + templates.template_default->apply({ { + {"role", "user"}, + {"content", "test"}, + } }, json(), true); + if (templates.template_tool_use) { + templates.template_tool_use->apply({ { + {"role", "user"}, + {"content", "test"}, + } }, json(), true); + } + return true; + } + catch (const std::exception& e) { + LOG_ERROR("failed to apply template: %s\n", e.what()); + return false; + } + } + else { + const char* tmpl = llama_model_chat_template(model, /* name */ nullptr); + const int32_t chat_res = llama_chat_apply_template(model, tmpl, chat, 1, true, nullptr, 0); + return chat_res > 0; + } } void init() { @@ -3182,22 +3215,16 @@ int main(int argc, char ** argv) { const auto model_meta = ctx_server.model_meta(); - // if a custom chat template is not supplied, we will use the one that comes with the model (if any) - if (params.chat_template.empty()) { - if (!ctx_server.validate_model_chat_template()) { - LOG_WARNING("The chat template that comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", {}); - params.chat_template = "chatml"; - } - } - // print sample chat example to make it clear which template is used - { + LOG_INFO("chat template", { - {"chat_example", llama_chat_format_example(ctx_server.model, params.chat_template)}, + {"chat_template", ctx_server.chat_templates.template_default->source().c_str()}, + }); + + LOG_INFO("chat template", { + {"chat_example", llama_chat_format_example(ctx_server.model, *ctx_server.chat_templates.template_default, ctx_server.params.use_jinja).c_str()}, {"built_in", params.chat_template.empty()}, }); - } - // // Middlewares // @@ -3560,9 +3587,11 @@ int main(int argc, char ** argv) { { "system_prompt", ctx_server.system_prompt.c_str() }, { "default_generation_settings", ctx_server.default_generation_settings_for_props }, { "total_slots", ctx_server.params.n_parallel }, - { "chat_template", curr_tmpl.c_str() } + { "chat_template", ctx_server.chat_templates.template_default->source() }, }; - + if (ctx_server.params.use_jinja && ctx_server.chat_templates.template_tool_use) { + data["chat_template_tool_use"] = ctx_server.chat_templates.template_tool_use->source(); + } res.set_content(data.dump(), "application/json; charset=utf-8"); }; @@ -3573,8 +3602,9 @@ int main(int argc, char ** argv) { } res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); - - json data = json::parse(req.body); + auto body = json::parse(req.body); + const auto& chat_template = body.contains("tools") && ctx_server.chat_templates.template_tool_use ? *ctx_server.chat_templates.template_tool_use : *ctx_server.chat_templates.template_default; + json data = oaicompat_completion_params_parse(ctx_server.model, body, chat_template, ctx_server.params.use_jinja); const int id_task = ctx_server.queue_tasks.get_new_id(); @@ -3674,7 +3704,11 @@ int main(int argc, char ** argv) { } res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); - json data = oaicompat_completion_params_parse(ctx_server.model, json::parse(req.body), params.chat_template); + + auto body = json::parse(req.body); + const auto& chat_template = body.contains("tools") && ctx_server.chat_templates.template_tool_use ? *ctx_server.chat_templates.template_tool_use : *ctx_server.chat_templates.template_default; + json data = oaicompat_completion_params_parse(ctx_server.model,body, chat_template, params.use_jinja); + const int id_task = ctx_server.queue_tasks.get_new_id(); diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp index 35e887fdb..36d22cc9e 100644 --- a/examples/server/utils.hpp +++ b/examples/server/utils.hpp @@ -6,6 +6,8 @@ // Change JSON_ASSERT from assert() to GGML_ASSERT: #define JSON_ASSERT GGML_ASSERT #include "json.hpp" +#include "minja.hpp" +#include "chat-template.hpp" #include "kimi_k2_tools.hpp" #include "qwen3_tools.hpp" #include "deepseek_r1_tools.hpp" @@ -125,7 +127,7 @@ static inline void server_log(const char * level, const char * function, int lin // // Format given chat. If tmpl is empty, we take the template from model metadata -inline std::string format_chat(const struct llama_model * model, const std::string & tmpl, const std::vector & messages, const json & tools = json::array(), const std::string & model_name = "") { +inline std::string format_chat(const struct llama_model * model, common_chat_template tmpl, const std::vector & messages, const json & tools = json::array(), const std::string & model_name = "") { std::vector chat; // Inject tools into the first system message, or create one if none exists @@ -197,8 +199,8 @@ inline std::string format_chat(const struct llama_model * model, const std::stri chat.push_back({role, content}); } + auto formatted_chat = llama_chat_apply_template(model, tmpl, chat, true, /* use_jinja= */ false); - auto formatted_chat = llama_chat_apply_template(model, tmpl, chat, true); LOG_VERBOSE("formatted_chat", {{"text", formatted_chat.c_str()}}); return formatted_chat; } @@ -425,46 +427,24 @@ static tool_choice_type tool_choice_parse_oaicompat(const std::string & tool_cho static json oaicompat_completion_params_parse( const struct llama_model * model, const json & body, /* openai api json semantics */ - const std::string & chat_template) { + const common_chat_template& tmpl, + bool use_jinja) { json llama_params; llama_params["__oaicompat"] = true; + auto tools = json_value(body, "tools", json()); + auto has_tools = tools.is_array() && !tools.empty(); + if (has_tools) { + if (use_jinja) { + fprintf(stdout,"tools param is not fully supported yet\n"); // Extract tools from the request body json tools = json_value(body, "tools", json::array()); + } // Debug: Log system prompt when tools are detected - if (!tools.empty() && server_verbose) { - LOG_VERBOSE("Tool calls detected in request", { - {"tool_count", tools.size()}, - {"model", json_value(body, "model", std::string(DEFAULT_OAICOMPAT_MODEL))} - }); - - // Extract and log system prompt from messages - if (body.contains("messages") && body["messages"].is_array()) { - for (const auto& msg : body["messages"]) { - if (msg.contains("role") && msg["role"] == "system" && msg.contains("content")) { - std::string content_str; - if (msg["content"].is_string()) { - content_str = msg["content"]; - } else if (msg["content"].is_array()) { - // Handle content blocks format - for (const auto& block : msg["content"]) { - if (block.contains("type") && block["type"] == "text" && block.contains("text")) { - if (!content_str.empty()) content_str += " "; - content_str += block["text"]; - } - } - } - - if (!content_str.empty()) { - LOG_VERBOSE("System prompt with tools", { - {"system_prompt", content_str.substr(0, 500) + (content_str.length() > 500 ? "..." : "")} - }); - } - break; // Only log first system message - } - } + else { + throw std::runtime_error("tools param requires --jinja flag"); } } @@ -472,7 +452,7 @@ static json oaicompat_completion_params_parse( std::string model_name = json_value(body, "model", std::string(DEFAULT_OAICOMPAT_MODEL)); // Apply chat template to the list of messages with tools - llama_params["prompt"] = format_chat(model, chat_template, body.at("messages"), tools, model_name); + llama_params["prompt"] = format_chat(model, tmpl, body.at("messages"), tools, model_name); // Handle "stop" field if (body.contains("stop") && body.at("stop").is_string()) { @@ -491,6 +471,13 @@ static json oaicompat_completion_params_parse( throw std::runtime_error("response_format type must be one of \"text\" or \"json_object\", but got: " + response_type); } } + // Apply chat template to the list of messages + if (use_jinja) { + llama_params["prompt"] = tmpl.apply(body.at("messages"), tools, /* add_generation_prompt= */ true); + } + else { + llama_params["prompt"] = format_chat(model, tmpl, body.at("messages")); + } // Handle "n" field int n_choices = json_value(body, "n", 1); diff --git a/include/llama.h b/include/llama.h index 0c26868eb..c68fa2298 100644 --- a/include/llama.h +++ b/include/llama.h @@ -554,8 +554,9 @@ extern "C" { LLAMA_API bool llama_supports_mlock (void); LLAMA_API bool llama_supports_gpu_offload(void); - LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx); + + LLAMA_API const struct llama_model * llama_get_model(const struct llama_context * ctx); LLAMA_API uint32_t llama_n_ctx (const struct llama_context * ctx); LLAMA_API uint32_t llama_n_batch (const struct llama_context * ctx); LLAMA_API uint32_t llama_n_ubatch (const struct llama_context * ctx); @@ -566,6 +567,8 @@ extern "C" { LLAMA_API enum llama_vocab_type llama_vocab_type (const struct llama_model * model); LLAMA_API enum llama_rope_type llama_rope_type (const struct llama_model * model); + LLAMA_API const struct llama_vocab* llama_model_get_vocab(const struct llama_model* model); + LLAMA_API const char* llama_model_chat_template(const struct llama_model* model, const char* name); LLAMA_API int32_t llama_n_vocab (const struct llama_model * model); LLAMA_API const struct llama_vocab* llama_get_model_vocab(const struct llama_model* model); LLAMA_API int32_t llama_n_ctx_train(const struct llama_model * model); diff --git a/models/templates/Mistral-Small-3.2-24B-Instruct-2506.jinja b/models/templates/Mistral-Small-3.2-24B-Instruct-2506.jinja new file mode 100644 index 000000000..19a3eaee4 --- /dev/null +++ b/models/templates/Mistral-Small-3.2-24B-Instruct-2506.jinja @@ -0,0 +1,124 @@ +{%- set today = strftime_now("%Y-%m-%d") %} +{%- set default_system_message = "You are Mistral Small 3, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.\nYour knowledge base was last updated on 2023-10-01. The current date is " + today + ".\n\nWhen you're not sure about some information or when the user's request requires up-to-date or specific data, you must use the available tools to fetch the information. Do not hesitate to use tools whenever they can provide a more accurate or complete response. If no relevant tools are available, then clearly state that you don't have the information and avoid making up anything. + +If the user's question is not clear, ambiguous, or does not provide enough context for you to accurately answer the question, you do not try to answer it right away and you rather ask the user to clarify their request (e.g. \"What are some good restaurants around me?\" => \"Where are you?\" or \"When is the next flight to Tokyo\" => \"Where do you travel from?\"). +You are always very attentive to dates, and when asked about information at specific dates, you discard information that is at another date. +You follow these instructions in all languages, and always respond to the user in the language they use or request. +Next sections describe the capabilities that you have. + +# WEB BROWSING INSTRUCTIONS + +You cannot perform any web search or access internet to open URLs, links etc. If it seems like the user is expecting you to do so, you clarify the situation and ask the user to copy paste the text directly in the chat. + +# MULTI-MODAL INSTRUCTIONS + +You have the ability to read images, but you cannot generate images. You also cannot transcribe audio files or videos. +You cannot read nor transcribe audio files or videos. + +# TOOL CALLING INSTRUCTIONS + +You may have access to tools that you can use to fetch information or perform actions. You must use these tools in the following situations: + +1. When the request requires up-to-date information. +2. When the request requires specific data that you do not have in your knowledge base. +3. When the request involves actions that you cannot perform without tools. + +Always prioritize using tools to provide the most accurate and helpful response. If tools are not available, inform the user that you cannot perform the requested action at the moment." %} + +{{- bos_token }} + +{%- set system_prompt = default_system_message %} +{%- set loop_messages = messages %} + +{%- if not tools is defined %} + {%- set tools = none %} +{%- endif %} + +{%- if messages|length > 0 and messages[0]['role'] == 'system' %} + {%- if messages[0]['content'] is string %} + {%- set system_prompt = messages[0]['content'] %} + {%- else %} + {%- set system_prompt = messages[0]['content'][0]['text'] %} + {%- endif %} + {%- set loop_messages = messages[1:] %} +{%- endif %} + +{%- set user_messages = loop_messages | selectattr("role", "equalto", "user") | list %} + +{%- set ns = namespace(index=0) %} +{%- for message in loop_messages %} + {%- if not (message.role == "tool" or (message.get('tool_calls'))) %} + {%- if (message["role"] == "user") != (ns.index % 2 == 0) %} + {{- raise_exception("After the optional system message, conversation roles must alternate user/assistant/user/assistant/...") }} + {%- endif %} + {%- set ns.index = ns.index + 1 %} + {%- endif %} +{%- endfor %} + +{{- '[SYSTEM_PROMPT]' + system_prompt + '[/SYSTEM_PROMPT]' }} + +{%- for message in loop_messages %} + {%- if message['role'] == 'system' %} + {%- if message['content'] is string %} + {{- '[SYSTEM_PROMPT]' + message['content'] + '[/SYSTEM_PROMPT]' }} + {%- else %} + {{- '[SYSTEM_PROMPT]' + message['content'][0]['text'] + '[/SYSTEM_PROMPT]' }} + {%- endif %} + {%- elif message['role'] == 'user' %} + {%- if tools is not none and (message == user_messages[-1]) %} + {{- '[AVAILABLE_TOOLS]' + tools|tojson + '[/AVAILABLE_TOOLS]' }} + {%- endif %} + {{- '[INST]' }} + {%- if message['content'] is string %} + {{- message['content'] }} + {%- else %} + {%- for block in message['content'] %} + {%- if block['type'] == 'text' %} + {{- block['text'] }} + {%- elif block['type'] in ['image', 'image_url'] %} + {{- '[IMG]' }} + {%- else %} + {{- raise_exception('Only text and image blocks are supported in message content!') }} + {%- endif %} + {%- endfor %} + {%- endif %} + {{- '[/INST]' }} + {%- elif message['role'] == 'assistant' %} + {%- if message.get('tool_calls') %} + {%- for tool_call in message.tool_calls %} + {{- '[TOOL_CALLS]' + tool_call.function.name }} + {%- if not tool_call.id is defined or tool_call.id is not string or tool_call.id|length != 9 %} + {{- raise_exception("Tool call IDs should be alphanumeric strings with length 9!") }} + {%- endif %} + {{- '[CALL_ID]' + tool_call.id }} + {{- '[ARGS]' + tool_call['function']['arguments']|tojson }} + {%- endfor %} + {{- eos_token }} + {%- elif message['content'] is string %} + {{- message['content'] + eos_token }} + {%- else %} + {%- for block in message['content'] %} + {%- if block['type'] == 'text' %} + {{- block['text'] }} + {%- elif block['type'] in ['image', 'image_url'] %} + {{- '[IMG]' }} + {%- else %} + {{- raise_exception('Only text and image blocks are supported in assistant content!') }} + {%- endif %} + {%- endfor %} + {{- eos_token }} + {%- endif %} + {%- elif message['role'] == 'tool_results' or message['role'] == 'tool' %} + {%- if message.content is defined and message.content.content is defined %} + {%- set content = message.content.content %} + {%- else %} + {%- set content = message.content %} + {%- endif %} + {%- if not message.tool_call_id is defined or message.tool_call_id is not string or message['tool_call_id']|length != 9 %} + {{- raise_exception("Tool call IDs should be alphanumeric strings with length 9!") }} + {%- endif %} + {{- '[TOOL_RESULTS]' + message.tool_call_id + '[TOOL_CONTENT]' + content|string + '[/TOOL_RESULTS]' }} + {%- else %} + {{- raise_exception('Only system, user, assistant, and tool roles are supported!') }} + {%- endif %} +{%- endfor %} diff --git a/scripts/get_hf_chat_template.py b/scripts/get_hf_chat_template.py new file mode 100644 index 000000000..23bb1de59 --- /dev/null +++ b/scripts/get_hf_chat_template.py @@ -0,0 +1,77 @@ +#!/usr/bin/env python +''' + Fetches the Jinja chat template of a HuggingFace model. + If a model has multiple chat templates, you can specify the variant name. + + Syntax: + ./scripts/get_hf_chat_template.py model_id [variant] + + Examples: + ./scripts/get_hf_chat_template.py NousResearch/Meta-Llama-3-8B-Instruct + ./scripts/get_hf_chat_template.py NousResearch/Hermes-3-Llama-3.1-8B tool_use + ./scripts/get_hf_chat_template.py meta-llama/Llama-3.2-3B-Instruct +''' + +import json +import re +import sys + + +def get_hf_chat_template(model_id, variant=None): + try: + # Use huggingface_hub library if available. + # Allows access to gated models if the user has access and ran `huggingface-cli login`. + from huggingface_hub import hf_hub_download + with open(hf_hub_download(repo_id=model_id, filename="tokenizer_config.json")) as f: + config_str = f.read() + except ImportError: + import requests + assert re.match(r"^[\w.-]+/[\w.-]+$", model_id), f"Invalid model ID: {model_id}" + response = requests.get(f"https://huggingface.co/{model_id}/resolve/main/tokenizer_config.json") + if response.status_code == 401: + raise Exception('Access to this model is gated, please request access, authenticate with `huggingface-cli login` and make sure to run `pip install huggingface_hub`') + response.raise_for_status() + config_str = response.text + + try: + config = json.loads(config_str) + except json.JSONDecodeError: + # Fix https://huggingface.co/NousResearch/Meta-Llama-3-8B-Instruct/blob/main/tokenizer_config.json + # (Remove extra '}' near the end of the file) + config = json.loads(re.sub(r'\}([\n\s]*\}[\n\s]*\],[\n\s]*"clean_up_tokenization_spaces")', r'\1', config_str)) + + chat_template = config['chat_template'] + if isinstance(chat_template, str): + return chat_template + else: + variants = { + ct['name']: ct['template'] + for ct in chat_template + } + + def format_variants(): + return ', '.join(f'"{v}"' for v in variants.keys()) + + if variant is None: + if 'default' not in variants: + raise Exception(f'Please specify a chat template variant (one of {format_variants()})') + variant = 'default' + sys.stderr.write(f'Note: picked "default" chat template variant (out of {format_variants()})\n') + elif variant not in variants: + raise Exception(f"Variant {variant} not found in chat template (found {format_variants()})") + + return variants[variant] + + +def main(args): + if len(args) < 1: + raise ValueError("Please provide a model ID and an optional variant name") + model_id = args[0] + variant = None if len(args) < 2 else args[1] + + template = get_hf_chat_template(model_id, variant) + sys.stdout.write(template) + + +if __name__ == '__main__': + main(sys.argv[1:]) diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 3aea013ae..f86137c43 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -22,7 +22,7 @@ add_library(llama unicode-data.cpp ) -target_include_directories(llama PUBLIC . ../include) +target_include_directories(llama PUBLIC . ../include ../common) target_include_directories(llama PRIVATE ../ggml/src) target_compile_features (llama PUBLIC cxx_std_11) # don't bump diff --git a/src/llama.cpp b/src/llama.cpp index 50b9ad5c2..fb9331ac5 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -123,15 +123,31 @@ // // trim whitespace from the beginning and end of a string +//static std::string trim(const std::string & str) { +// Fails for Chinese character +// size_t start = 0; +// size_t end = str.size(); +// while (start < end && isspace(str[start])) { +// start += 1; +// } +// while (end > start && isspace(str[end - 1])) { +// end -= 1; +// } +// return str.substr(start, end - start); +//} + +static bool is_utf8_whitespace(uint8_t c) { + // Basic ASCII whitespace + if (c <= 0x7F) return isspace(c); + // Else: Not whitespace (or you'd need a full Unicode table) + return false; +} + static std::string trim(const std::string & str) { size_t start = 0; size_t end = str.size(); - while (start < end && isspace(str[start])) { - start += 1; - } - while (end > start && isspace(str[end - 1])) { - end -= 1; - } + while (start < end && is_utf8_whitespace(str[start])) start++; + while (end > start && is_utf8_whitespace(str[end - 1])) end--; return str.substr(start, end - start); } @@ -400,6 +416,8 @@ enum llm_kv { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, LLM_KV_TOKENIZER_HF_JSON, LLM_KV_TOKENIZER_RWKV, + LLM_KV_TOKENIZER_CHAT_TEMPLATE, + LLM_KV_TOKENIZER_CHAT_TEMPLATE_N, LLM_KV_TOKENIZER_FIM_PRE_ID, LLM_KV_TOKENIZER_FIM_SUF_ID, LLM_KV_TOKENIZER_FIM_MID_ID, @@ -512,6 +530,8 @@ static const std::map LLM_KV_NAMES = { { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" }, { LLM_KV_TOKENIZER_HF_JSON, "tokenizer.huggingface.json" }, { LLM_KV_TOKENIZER_RWKV, "tokenizer.rwkv.world" }, + { LLM_KV_TOKENIZER_CHAT_TEMPLATE, "tokenizer.chat_template" }, + { LLM_KV_TOKENIZER_CHAT_TEMPLATE_N, "tokenizer.chat_template.%s" }, { LLM_KV_TOKENIZER_FIM_PRE_ID, "tokenizer.ggml.fim_pre_token_id" }, { LLM_KV_TOKENIZER_FIM_SUF_ID, "tokenizer.ggml.fim_suf_token_id" }, { LLM_KV_TOKENIZER_FIM_MID_ID, "tokenizer.ggml.fim_mid_token_id" }, @@ -530,13 +550,11 @@ static const std::map LLM_KV_NAMES = { }; struct LLM_KV { - LLM_KV(llm_arch arch) : arch(arch) {} + LLM_KV(llm_arch arch, const char* suffix = nullptr); llm_arch arch; - - std::string operator()(llm_kv kv) const { - return ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch)); - } + const char* suffix; + std::string operator()(llm_kv kv) const; }; enum llm_tensor { @@ -634,6 +652,13 @@ enum llm_tensor { LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, }; +LLM_KV::LLM_KV(llm_arch arch, const char* suffix) : arch(arch), suffix(suffix) {} + +std::string LLM_KV::operator()(llm_kv kv) const { + return suffix ? ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch), suffix) + : ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch)); +} + static const std::map> LLM_TENSOR_NAMES = { { LLM_ARCH_LLAMA, @@ -21964,6 +21989,10 @@ void llama_free(struct llama_context * ctx) { delete ctx; } +const struct llama_vocab* llama_model_get_vocab(const struct llama_model* model) { + return &model->vocab; +} + const struct llama_model * llama_get_model(const struct llama_context * ctx) { return &ctx->model; } @@ -22152,6 +22181,24 @@ uint64_t llama_model_size(const struct llama_model * model) { return size; } +const char* llama_model_chat_template(const struct llama_model* model, const char* name) { + const auto key = name ? LLM_KV(model->arch, name)(LLM_KV_TOKENIZER_CHAT_TEMPLATE) + : LLM_KV(model->arch)(LLM_KV_TOKENIZER_CHAT_TEMPLATE); + const auto& it = model->gguf_kv.find(key); + if (it == model->gguf_kv.end()) { + // one-off fix for very popular models (so we are not flooded with issues) + // do not extend this list unless absolutely necessary + // Mistral-Small-2503 does not have built-in chat template + llama_vocab_pre_type pre_type = model->vocab.type_pre; + if (!name && pre_type == LLAMA_VOCAB_PRE_TYPE_TEKKEN && model->layers.size() == 40) { + return "mistral-v7-tekken"; + } + + return nullptr; + } + return it->second.c_str(); +} + uint64_t llama_model_n_params(const struct llama_model * model) { uint64_t nparams = 0; for (const auto & it : model->tensors_by_name) { diff --git a/tests/test-chat-template.cpp b/tests/test-chat-template.cpp index a8222caee..c7165311f 100644 --- a/tests/test-chat-template.cpp +++ b/tests/test-chat-template.cpp @@ -7,6 +7,16 @@ #include "llama.h" #include "common.h" +#include "chat-template.hpp" + +static std::string normalize_newlines(const std::string& s) { +#ifdef _WIN32 + static const std::regex nl_regex("\r\n"); + return std::regex_replace(s, nl_regex, "\n"); +#else + return s; +#endif +} int main(void) { llama_chat_message conversation[] = { @@ -143,10 +153,10 @@ int main(void) { std::vector chat2; llama_chat_msg sys_msg{"system", "You are a helpful assistant"}; - auto fmt_sys = [&](std::string tmpl) { - auto output = llama_chat_format_single(nullptr, tmpl, chat2, sys_msg, false); - printf("fmt_sys(%s) : %s\n", tmpl.c_str(), output.c_str()); - printf("-------------------------\n"); + auto fmt_sys = [&](std::string tmpl_str) { + minja::chat_template tmpl(tmpl_str, "", ""); + auto output = llama_chat_format_single(nullptr, tmpl, chat2, sys_msg, false, /* use_jinja= */ false); + printf("fmt_sys(%s) : %s\n", tmpl_str.c_str(), output.c_str()); return output; }; assert(fmt_sys("chatml") == "<|im_start|>system\nYou are a helpful assistant<|im_end|>\n"); @@ -162,9 +172,10 @@ int main(void) { chat2.push_back({"assistant", "I am assistant"}); llama_chat_msg new_msg{"user", "How are you"}; - auto fmt_single = [&](std::string tmpl) { - auto output = llama_chat_format_single(nullptr, tmpl, chat2, new_msg, true); - printf("fmt_single(%s) : %s\n", tmpl.c_str(), output.c_str()); + auto fmt_single = [&](std::string tmpl_str) { + minja::chat_template tmpl(tmpl_str, "", ""); + auto output = llama_chat_format_single(nullptr, tmpl, chat2, new_msg, true, /* use_jinja= */ false); + printf("fmt_single(%s) : %s\n", tmpl_str.c_str(), output.c_str()); printf("-------------------------\n"); return output; };