[PATCH] D59553: [LLD][ELF][DebugInfo] llvm-symbolizer shows incorrect source line info if --gc-sections used

Mon Apr 15 22:23:28 PDT 2019

MaskRay added a comment.

In D59553#1463805 <https://reviews.llvm.org/D59553#1463805>, @avl wrote:

> @MaskRay Hi, Since you changed status of review into "Needs Revision" - What changes do you think should be done ?

My thoughts about this change are split. I wanted to reset the status of review into "Normal" but I didn't how to do that :( I didn't want to change it to "Accepted" so I chose "Resign". It seems it didn't restore the status back to normal and only you can do that...

What I don't like

- complexity. Two lines are two lines.
- The demonstration case is a bit contrived. Filling in a garbage collected text section with zeros/nops to overlap with another text section doesn't seem common. The problems aren't new but lld's layout mitigates many(most?) of the problems. See https://reviews.llvm.org/D60470#1462246

What I like

- `-1` or `-2` makes the resolved relocation outstanding -> a different value from 0 makes it more obvious. Avoiding -1 just because "base address selection entry" uses -1 is not a strong argument IMHO. As long as there ranges are invalid (or valid but unrealistic to collide with good ones) we are good.
- Existing tools have learned various heuristics to avoid 0 DW_AT_low_pc collision, but I am not sure what they'd do if those relocations resolve to -1. llvm-symbolizer is certainly good (I have checked a few examples - I'm learning it recently), but what about other tools?

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D59553/new/

https://reviews.llvm.org/D59553