This does exactly what it says on the box. The only small gotcha is the
section index computation for offset_pair entries, which can use either
the base address section, or the section from the offset_pair entry.
This is to support both the cases where the base address is relocated
(points to the base of the CU, typically), and the case where the base
address is a constant (typically zero) and relocations are on the
offsets themselves.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
This patch doesn't change the dump format for the "raw" entries, but it creates a good opportunity to do that.
llvm/test/tools/llvm-dwarfdump/X86/debug_loclists.s | ||
---|---|---|
8–42 | I thought this layout would make the verbose/regular dump differences more obvious, but it could be just confusing... |
llvm/test/tools/llvm-dwarfdump/X86/debug_loclists.s | ||
---|---|---|
8–42 | If it *works* I think this is good. In the past I've encountered very unintuitive behavior when combining multiple --check-prefix options and -NEXT, but it looks like you are using it to the intended effect here. |
Thanks.
Do you have any opinion on what should be the format of "raw" debug loc entries? (a, b) ".sect", [a,b] ".sect", {a, b} ".sect", etc. ?
I thought this layout would make the verbose/regular dump differences more obvious, but it could be just confusing...