Tor's LLVM modifications

This repo is my running contributions and personal changes to clangd and clang-tidy. I'll try to merge as many of my updates as are accepted upstream to https://github.com/llvm/llvm-project.

clangd features:

Fix all

swap binary operands

testing.cpp.-.llvm-customizations.WSL_.Ubuntu.2024-01-25.10-18-04.mp4

Inlay hints for lambda captures

clang-tidy checks:

performance-vector-pessimization

Planned future work:

• modernize-use-views-enumerate: transform index-based for loops with end condition of .size() that use the index and also use the value at the index to range-based for loops with | std::views::enumerate
• modernize-use-views-iota: transform index-based for loops that start from an integral start and go to an integral end to range-based for loops with std::views::iota(start, end)
• modernize-use-views-zip: transform index-based for loops with end condition of .size() that use the value from two ranges r1 and r2 at the index to range-based for loops with std::views::zip(r1, r2)
• modernize-use-views-concat: transform consecutive range-based for loops with the same element type to a single for loop with std::views::concat(r1, r2)
• modernize-use-ctad: transform make_ functions to class-template-argument-deduction: std::make_optional(2) -> std::optional(2)
  • With config option to use braced-init or paren-init
• misc-use-static-or-inline-constexpr: suggest transforming global constexpr variables in header files to inline constexpr and global variables in source files or function-level variables to static constexpr
• performance-use-array: In cases with compile-known length of std::vector, change type to array instead
• bugprone-move-reference-capture: Suggest move captures for vars captured by reference which are later moved in the lambda iff the lambda is not immediately-invoked
• bugprone-use-midpoint: Suggest to use std::midpoint(x, y) instead of manual x + y / 2
• performance-initializer-list: In cases where vector is instantiated with the initializer-list ctor and the type is not trivially copyable, suggest using emplace back instead: std::vector<T> vec{t1, t2, t3} -> std::vector<T> vec = [&]{std::vector<T> out{}; out.emplace_back(t1); out.emplace_back(t2); out.emplace_back(t3); return out;}()
• Add "swap parameters" code action to clangd
• Add "extract constraints to concept" code action to clangd
• Add "move concept to requires clause" code action to clangd
• Add code action to convert include to forward-declarations
• Extract to lambda instead of function
• Inlay hints for size of included file
• Inlay hints for implicit default initialization
• Inlay hints for deduced template parameters
• Inlay hints for default arguments (and template args)
• Only publish block inlay hints for blocks bigger than a configurable line length
• Convert block comment to multiline comment & vice versa
• Convert member function qualifiers into explicit this parameter
• Convert eager monadic functions to lazy versions
• Autocomplete for #include should use fuzzy path finder
• Compute comptime enum underlying values on hover
  • For instance, see llvm/include/llvm/ADT/SparseBitVector.h, and hover over the enum values BITS_PER_ELEMENT
• Proper schema-based autocomplete for config.yaml
• Make "add missing includes" use the same path for the includes as autocomplete's "from _.h" functionality
• Make -Wunused-but-set-variable suggest removing the only-ever-set variable