[Feature Request]: Add flag for munging generated code with file and line info

[jabraham17]

When profiling Chapel code, it would be useful to be able to map generated symbol names back to the original Chapel code, without looking at the generated code.

For example,

proc foo(x: int) {
  writeln(x);
}
proc foo(x: real) {
  writeln(x);
}
proc foo(x: bool) {
  writeln(x);
}

foo(1);
foo(1.0);
foo(true);

The generated code for this code snippet with have foo_chpl, foo_chpl2, and foo_chpl3, but without looking at the generated code there is no way to know which is which. With the C backend, before each function name we print a comment with the original filename and line number (compiler/codegen/cg-symbol.cpp:2764), like

/*MySourceFile.chpl:7 */
void foo_chpl3(....) {

This information is harder to glean from the LLVM backend, and furthermore its unfortunate to require looking at the generated code to get this information

It would be useful to provide a flag like --munge-with-locations (similar to --munge-with-ids), which would embed the filename and line number in the function name, e.g. foo_chpl_MySourceFile_1

[mppf]

Some related things that you migth already be aware of:

• the stacktrace support has some function name tables
• --cpp-lines and -g
• --munge-with-ids embeds dyno's IDs into symbol names (and only works with the LLVM backend). It would be relatively straightforward to make a helper tool to go from such a munged name to a source location.

Ideally, with the LLVM backend, we would be generating sufficiently good debug information for debuggers and other tooling. Perhaps that would be enough for your use cases as well.