Merge pull request #12748 from ethereum/markCreationAssembly

Store whether an evmasm Assembly is creation code.

Author: ekpyron

Changelog.md

@@ -8,6 +8,7 @@ Compiler Features:
 * JSON-AST: Added selector field for errors and events.
 
 Bugfixes:
+ * Yul IR Code Generation: Optimize embedded creation code with correct settings. This fixes potential mismatches between the constructor code of a contract compiled in isolation and the bytecode in ``type(C).creationCode``, resp. the bytecode used for ``new C(...)``.
 
 
 ### 0.8.12 (2022-02-16)

docs/yul.rst

@@ -1124,6 +1124,11 @@ regular strings in native encoding. For code,
 
 Above, ``Block`` refers to ``Block`` in the Yul code grammar explained in the previous chapter.
 
+.. note::
+
+    An object with a name that ends in ``_deployed`` is treated as deployed code by the Yul optimizer.
+    The only consequence of this is a different gas cost heuristic in the optimizer.
+
 .. note::
 
     Data objects or sub-objects whose names contain a ``.`` can be defined
@@ -1172,17 +1177,17 @@ An example Yul Object is shown below:
 
             // now return the runtime object (the currently
             // executing code is the constructor code)
-            size := datasize("runtime")
+            size := datasize("Contract1_deployed")
             offset := allocate(size)
             // This will turn into a memory->memory copy for Ewasm and
             // a codecopy for EVM
-            datacopy(offset, dataoffset("runtime"), size)
+            datacopy(offset, dataoffset("Contract1_deployed"), size)
             return(offset, size)
         }
 
         data "Table2" hex"4123"
 
-        object "runtime" {
+        object "Contract1_deployed" {
             code {
                 function allocate(size) -> ptr {
                     ptr := mload(0x40)
@@ -1204,7 +1209,7 @@ An example Yul Object is shown below:
                 // code here ...
             }
 
-            object "runtime" {
+            object "Contract2_deployed" {
                 code {
                     // code here ...
                 }

libevmasm/Assembly.cpp

@@ -415,9 +415,8 @@ map<u256, u256> const& Assembly::optimiseInternal(
 	for (size_t subId = 0; subId < m_subs.size(); ++subId)
 	{
 		OptimiserSettings settings = _settings;
-		// Disable creation mode for sub-assemblies.
-		settings.isCreation = false;
-		map<u256, u256> const& subTagReplacements = m_subs[subId]->optimiseInternal(
+		Assembly& sub = *m_subs[subId];
+		map<u256, u256> const& subTagReplacements = sub.optimiseInternal(
 			settings,
 			JumpdestRemover::referencedTags(m_items, subId)
 		);
@@ -436,7 +435,7 @@ map<u256, u256> const& Assembly::optimiseInternal(
 				m_items,
 				_tagsReferencedFromOutside,
 				_settings.expectedExecutionsPerDeployment,
-				_settings.isCreation,
+				isCreation(),
 				_settings.evmVersion
 			}.optimise();
 
@@ -537,8 +536,8 @@ map<u256, u256> const& Assembly::optimiseInternal(
 
 	if (_settings.runConstantOptimiser)
 		ConstantOptimisationMethod::optimiseConstants(
-			_settings.isCreation,
-			_settings.isCreation ? 1 : _settings.expectedExecutionsPerDeployment,
+			isCreation(),
+			isCreation() ? 1 : _settings.expectedExecutionsPerDeployment,
 			_settings.evmVersion,
 			*this
 		);

libevmasm/Assembly.h

@@ -48,7 +48,7 @@ using AssemblyPointer = std::shared_ptr<Assembly>;
 class Assembly
 {
 public:
-	explicit Assembly(std::string _name = std::string()):m_name(std::move(_name)) { }
+	Assembly(bool _creation, std::string _name): m_creation(_creation), m_name(std::move(_name)) { }
 
 	AssemblyItem newTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(Tag, m_usedTags++); }
 	AssemblyItem newPushTag() { assertThrow(m_usedTags < 0xffffffff, AssemblyException, ""); return AssemblyItem(PushTag, m_usedTags++); }
@@ -117,7 +117,6 @@ class Assembly
 
 	struct OptimiserSettings
 	{
-		bool isCreation = false;
 		bool runInliner = false;
 		bool runJumpdestRemover = false;
 		bool runPeephole = false;
@@ -157,6 +156,8 @@ class Assembly
 	std::vector<size_t> decodeSubPath(size_t _subObjectId) const;
 	size_t encodeSubPath(std::vector<size_t> const& _subPath);
 
+	bool isCreation() const { return m_creation; }
+
 protected:
 	/// Does the same operations as @a optimise, but should only be applied to a sub and
 	/// returns the replaced tags. Also takes an argument containing the tags of this assembly
@@ -214,6 +215,8 @@ class Assembly
 	mutable std::vector<size_t> m_tagPositionsInBytecode;
 
 	int m_deposit = 0;
+	/// True, if the assembly contains contract creation code.
+	bool const m_creation = false;
 	/// Internal name of the assembly object, only used with the Yul backend
 	/// currently
 	std::string m_name;

libsolidity/codegen/CompilerContext.cpp

@@ -572,8 +572,7 @@ void CompilerContext::updateSourceLocation()
 evmasm::Assembly::OptimiserSettings CompilerContext::translateOptimiserSettings(OptimiserSettings const& _settings)
 {
 	// Constructing it this way so that we notice changes in the fields.
-	evmasm::Assembly::OptimiserSettings asmSettings{false, false,  false, false, false, false, false, m_evmVersion, 0};
-	asmSettings.isCreation = true;
+	evmasm::Assembly::OptimiserSettings asmSettings{false,  false, false, false, false, false, m_evmVersion, 0};
 	asmSettings.runInliner = _settings.runInliner;
 	asmSettings.runJumpdestRemover = _settings.runJumpdestRemover;
 	asmSettings.runPeephole = _settings.runPeephole;

libsolidity/codegen/CompilerContext.h

@@ -65,7 +65,7 @@ class CompilerContext
 		RevertStrings _revertStrings,
 		CompilerContext* _runtimeContext = nullptr
 	):
-		m_asm(std::make_shared<evmasm::Assembly>()),
+		m_asm(std::make_shared<evmasm::Assembly>(_runtimeContext != nullptr, std::string{})),
 		m_evmVersion(_evmVersion),
 		m_revertStrings(_revertStrings),
 		m_reservedMemory{0},

libyul/AssemblyStack.cpp

@@ -38,6 +38,7 @@
 
 #include <libevmasm/Assembly.h>
 #include <liblangutil/Scanner.h>
+#include <boost/algorithm/string.hpp>
 #include <optional>
 
 using namespace std;
@@ -71,8 +72,7 @@ evmasm::Assembly::OptimiserSettings translateOptimiserSettings(
 )
 {
 	// Constructing it this way so that we notice changes in the fields.
-	evmasm::Assembly::OptimiserSettings asmSettings{false, false,  false, false, false, false, false, _evmVersion, 0};
-	asmSettings.isCreation = true;
+	evmasm::Assembly::OptimiserSettings asmSettings{false,  false, false, false, false, false, _evmVersion, 0};
 	asmSettings.runInliner = _settings.runInliner;
 	asmSettings.runJumpdestRemover = _settings.runJumpdestRemover;
 	asmSettings.runPeephole = _settings.runPeephole;
@@ -194,7 +194,10 @@ void AssemblyStack::optimize(Object& _object, bool _isCreation)
 	yulAssert(_object.analysisInfo, "");
 	for (auto& subNode: _object.subObjects)
 		if (auto subObject = dynamic_cast<Object*>(subNode.get()))
-			optimize(*subObject, false);
+		{
+			bool isCreation = !boost::ends_with(subObject->name.str(), "_deployed");
+			optimize(*subObject, isCreation);
+		}
 
 	Dialect const& dialect = languageToDialect(m_language, m_evmVersion);
 	unique_ptr<GasMeter> meter;
@@ -281,7 +284,7 @@ AssemblyStack::assembleEVMWithDeployed(optional<string_view> _deployName) const
 	yulAssert(m_parserResult->code, "");
 	yulAssert(m_parserResult->analysisInfo, "");
 
-	evmasm::Assembly assembly;
+	evmasm::Assembly assembly(true, {});
 	EthAssemblyAdapter adapter(assembly);
 	compileEVM(adapter, m_optimiserSettings.optimizeStackAllocation);
 

libyul/backends/evm/AbstractAssembly.h

@@ -98,7 +98,7 @@ class AbstractAssembly
 	/// Append the assembled size as a constant.
 	virtual void appendAssemblySize() = 0;
 	/// Creates a new sub-assembly, which can be referenced using dataSize and dataOffset.
-	virtual std::pair<std::shared_ptr<AbstractAssembly>, SubID> createSubAssembly(std::string _name = "") = 0;
+	virtual std::pair<std::shared_ptr<AbstractAssembly>, SubID> createSubAssembly(bool _creation, std::string _name = "") = 0;
 	/// Appends the offset of the given sub-assembly or data.
 	virtual void appendDataOffset(std::vector<SubID> const& _subPath) = 0;
 	/// Appends the size of the given sub-assembly or data.

libyul/backends/evm/EVMObjectCompiler.cpp

@@ -30,6 +30,8 @@
 #include <libyul/Object.h>
 #include <libyul/Exceptions.h>
 
+#include <boost/algorithm/string.hpp>
+
 using namespace solidity::yul;
 using namespace std;
 
@@ -48,7 +50,8 @@ void EVMObjectCompiler::run(Object& _object, bool _optimize)
 	for (auto const& subNode: _object.subObjects)
 		if (auto* subObject = dynamic_cast<Object*>(subNode.get()))
 		{
-			auto subAssemblyAndID = m_assembly.createSubAssembly(subObject->name.str());
+			bool isCreation = !boost::ends_with(subObject->name.str(), "_deployed");
+			auto subAssemblyAndID = m_assembly.createSubAssembly(isCreation, subObject->name.str());
 			context.subIDs[subObject->name] = subAssemblyAndID.second;
 			subObject->subId = subAssemblyAndID.second;
 			compile(*subObject, *subAssemblyAndID.first, m_dialect, _optimize);

libyul/backends/evm/EthAssemblyAdapter.cpp

@@ -122,9 +122,9 @@ void EthAssemblyAdapter::appendAssemblySize()
 	m_assembly.appendProgramSize();
 }
 
-pair<shared_ptr<AbstractAssembly>, AbstractAssembly::SubID> EthAssemblyAdapter::createSubAssembly(string _name)
+pair<shared_ptr<AbstractAssembly>, AbstractAssembly::SubID> EthAssemblyAdapter::createSubAssembly(bool _creation, string _name)
 {
-	shared_ptr<evmasm::Assembly> assembly{make_shared<evmasm::Assembly>(std::move(_name))};
+	shared_ptr<evmasm::Assembly> assembly{make_shared<evmasm::Assembly>(_creation, std::move(_name))};
 	auto sub = m_assembly.newSub(assembly);
 	return {make_shared<EthAssemblyAdapter>(*assembly), static_cast<size_t>(sub.data())};
 }