fix(debuginfo): Improve .o handling on MachO (#173)
#173
Conversation
Currently `raw_section` finds the given section by assuming that it's within a segment named `__TEXT` or `__DWARF`, finding such a segment, and then searching within that segment. However, this fails for `.o` files, because they have a single nameless segment. As https://github.com/aidansteele/osx-abi-macho-file-format-reference says: > For compactness, an intermediate object file contains only one > segment. This segment has no name; it contains all the sections > destined ultimately for different segments in the final object file. This commit changes `raw_section` so it ignores segment names, and simply searches through all segments. As well as working with `.o` files, the resulting code is a little shorter and simpler.
Because they occur in `.o` files. Without the first change in this commit, debuginfo for entire `.o` files can be missed. Without the second change, debuginfo for the first function (which can start at 0x0) can be missed.
|
This is my first PR for this project, so please let me know if I've overlooked anything w.r.t. contributing. I did run |
jan-auer
left a comment
There was a problem hiding this comment.
Thanks for the PR, @nnethercote! Once we get my question below addressed, I'm happy to merge.
This is my first PR for this project, so please let me know if I've overlooked anything w.r.t. contributing.
All good, thanks for asking!
|
@jan-auer: Any new thoughts here? The project I'm working on currently uses a fork of Symbolic that contains this PR, because it needs to work directly with |
|
My tool is now being used in production for Firefox. It's still relying on the version of |
|
Sorry for the late reply, I took a closer look and ran some tests now. Performance checks out. It seems that this was an outdated concern, and I can frankly no longer find the cases where this would've hit us. For some reason I remember this also being a correctness concern, but this might be false memory. Thanks for this patch. Will look into merging all PRs now and then releasing a new version to crates.io as soon as we get rid of all the git dependencies. |
.o handling on Mac.o handling on MachO (#173)
| // TODO: *technically* there could be a relocatable section placed at VA 0 | ||
| let low_pc = match low_pc { | ||
| Some(low_pc) if low_pc != 0 => low_pc, | ||
| Some(low_pc) => low_pc, |
There was a problem hiding this comment.
@nnethercote I'm afraid we'll need to make this check conditional, as you suggested. See #173
I am using
symbolic_debuginfoin a symbolicator used for stack traces produced by Firefox.On Mac, we don't build dSYMs by default because doing so is slow. So I want to emulate
atos, which is able to read debuginfo from a binary by consulting the binary's symbol table and then reading debug info from the object files and archive files mentioned.I have this working, but it requires a couple of small changes to
symbolic_debuginfo, due to slight differences between.ofiles and executables/libraries.