build: refactor Linux binary stripping to align with upstream #47932
Conversation
| ":strip_libEGL_shlib", | ||
| ":strip_libGLESv2_shlib", | ||
| ":strip_libffmpeg_shlib", | ||
| ":strip_libvk_swiftshader_shlib", |
There was a problem hiding this comment.
Why are we shipping these libraries as part of mksnapshot package ? This is not new in this PR, seems like we have been doing this for a while. These are chromium specific and mksnapshot should not be aware of them.
There was a problem hiding this comment.
@deepak1556, it looks like these are getting pulled in because the contents of the zip gets pulled from the GN runtime deps for the target. These come in via the dep chain:
//electron:electron_mksnapshot_zip -->
//tools/v8_context_snapshot:v8_context_snapshot_generator -->
//third_party/blink/public:blink
There was a problem hiding this comment.
Ah I thought we were using the target from v8 to build mksnapshot, didn't realize there was another one from tools. However I am pretty sure mksnapshot shouldn't need those libraries to run, the generated ELF didn't show any dependency on them in the dynamic section. Can we double check and remove those, it would save package size.
Description of Change
We occasionally see a transient issue in our published Linux releases where the
electronbinary gets relinked after we've stripped it when we rune build --target electron:electron_dist_zipdue to inconsistent timestamps on an input file. The binary stripping being out-of-band with the rest of these steps means we're always going to be at risk of that target causing a relink of the binary.Upstream Chromium has their own GN actions for doing stripping for their Linux releases. They do all of the same steps we currently do in CI (copy debug symbols, strip binary, add debug link), with the one exception that we pass
--compress-debug-sectionstoobjcopywhen copying the debug symbols. To retain that step that required me copy-paste-modifyingbuild/linux/strip_binary.{gni,py}from upstream Chromium. I can try to upstream the extra option since it's a minimal change, and if they accept it then we can drop these files and directly use upstream's.Chromium's
strip_binaryaction outputs the stripped binaries to separate files (defaults to<name>.stripped, which I've used in this PR), so this required some changes to thebuild/zip.pyscript to take the stripped version of the file when it is present. Using upstream'sstrip_binaryalso required setting the GN argenable_linux_installer = falseso that upstream's strip actions aren't pulled into the GN build, otherwise we get a GN error about target collision having the same output file (since we're stripping the same files as upstream, but we want our own target for them because we need to output the debug symbols to thedebug/directory, in addition to setting the compress option to true).One benefit of aligning with upstream here is that we'll be using
llvm-stripandllvm-objcopyfrom the Chromium toolchain, so we'll achieve greater consistency there, and currently we skip copying debug symbols on Linux ARM/ARM64 with a note aboutobjcopynot being available - that should work now, I bellieve.Things still to test:
mksnapshotstuff works on Linux ARMChecklist
Release Notes
Notes: none