1502499 - Ensure searchfox fallback local-build path still works after nasm dependency is added

Status: RESOLVED WONTFIX

(Webtools :: Searchfox, enhancement)

Description

[Kartikaya Gupta (email:[e-mail])]

After bug 1501796 lands we should check to make sure the local-build codepath for searchfox still works. In theory it should since it uses mach bootstrap to install dependencies before doing the build, but it hasn't really been exercised since we switched to downloading artifacts from taskcluster.

Comment 1

[Kartikaya Gupta (email:[e-mail])]

We're running `mach bootstrap` with the `--no-system-changes` argument which means that it doesn't install any stuff from apt-get. i.e. no nasm. So that needs fixing. But if we run without `--no-system-changes` then bootstrap goes wild with power and tries to set up mercurial, cinnabar, etc etc and we don't need any of that stuff. It would be nice if bootstrap had some in the middle option that just installed the proper dependencies and toolchain, but didn't mess with version control.

Comment 2

[Kartikaya Gupta (email:[e-mail])]

Actually I was wrong. Even though it presents setting up mercurial/cinnbar/etc as the default option, using --no-interactive doesn't actually choose those options. So if we drop --no-system-changes it should basically just work.

Comment 3

[Kartikaya Gupta (email:[e-mail])]

I tested that and it looks like after nasm is installed the build actually fails due to linker errors. *shrug*

10:28.37 media/ffvpx/libavutil/libmozavutil.so
10:28.42 /usr/bin/ld.gold: error: relocation section 7 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 10 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 13 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 7 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 10 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 13 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 7 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 10 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 13 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 9 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 12 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 15 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 7 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 10 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 13 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 7 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 10 uses unexpected symbol table 0
10:28.42 /usr/bin/ld.gold: error: relocation section 13 uses unexpected symbol table 0
10:28.42 clang: error: linker command failed with exit code 1 (use -v to see invocation)

Comment 4

[Kartikaya Gupta (email:[e-mail])]

I think going forward there's not much point maintaining the "local build" codepath for mozilla-central repos. We get too much value out of using the taskcluster artifacts (specifically getting data for all tier-1 platforms instead of just linux). The local build codepath is liable to break silently unless we're continually testing it, and it's not worth the effort.

Dropping this codepath also provides some value in terms of code cleanup - I had put all the taskcluster downloading and analysis merging stuff in the setup script but we can move some of it into the build script instead. That will provide a concrete advantage because it will allow the upload script to run sooner (because it runs after setup but before build) which makes it easier to share gecko-dev and gecko-blame to the mozilla-releases repos.