Rethink source locations in the IFC to accommodate implementations which need richer source-level information

[cdacamar]

It was expressed by @ChuanqiXu9 in the last module implementors meeting that both Clang and GCC have a richer representation of source locations than was initially required from MSVC (the early adopter of the IFC). One example cited was declarations:

    int i = 0;
//  ^        ^
//  |          | Decl end
//  | Decl begin

The other vendors have a span of source locations for declarations in which represent the declaration node as written in source. In Clang it is represented as DeclStmt <line:2:5, col:14> where the span is assumed to start at column '5' and end at column '14'.

In the IFC, today we only have a single source location for nearly every node but this representation appears to be insufficient for the needs of other compiler implementations. We should rethink this or provide a richer source location structure which could allow for ranges of locations.

[DarkArc]

EDG also needs the total line count of a given file to avoid traversing all source location information when an IFC is first loaded.

[ChuanqiXu9]

Thanks. For reference, clang will give more source location encodings for different decls and stmts. For example, for if-stmt: https://github.com/llvm/llvm-project/blob/0e1a52f556a90cc7b7ce7666fc476c99cf7bfb02/clang/lib/Serialization/ASTWriterStmt.cpp#L159-L161.

Although not formally documented, we can find the recorded source locations for decls and stmts in https://github.com/llvm/llvm-project/blob/main/clang/lib/Serialization/ASTWriterDecl.cpp and https://github.com/llvm/llvm-project/blob/main/clang/lib/Serialization/ASTWriterStmt.cpp by searching AddSourceLocation in these 2 files.