Buildbot failing due to a link error on arm-linux-androideabi

PR #38192 was recently accepted, but didn't pass buildbot tests.
build report
log contents

The error is:

CANNOT LINK EXECUTABLE: cannot locate symbol "_ZN11collections5slice10merge_sort8collapse17hfd9d6022dd12b1bbE" referenced by "./collectionstest-b2ddd77ea87480bb"...

adb: error: remote object '/data/tmp/check-stage2-T-arm-linux-androideabi-H-x86_64-unknown-linux-gnu-collectionstest-b2ddd77ea87480bb.log' does not exist


command did not execute successfully: "adb" "pull" "/data/tmp/check-stage2-T-arm-linux-androideabi-H-x86_64-unknown-linux-gnu-collectionstest-b2ddd77ea87480bb.log" "/buildslave/rust-buildbot/slave/auto-linux-64-x-android-t/build/obj/build/tmp"
expected success, got: exit code: 1

Symbol _ZN11collections5slice10merge_sort8collapse17hfd9d6022dd12b1bbE is actually a function I introduced. It is a fn nested within another fn, and called from one place only.

[michaelwoerister]

cc me

[alexcrichton]

cc @rust-lang/compiler (in addition to @michaelwoerister)

I should also note that, after some discussion over IRC, we decided to stick one #[inline] above that function. Hopefully that fixes the linker problem.

It's best to inline the function in this particular case anyway. By the way, the reason why the function was not getting inlined without explicit #[inline] is that that it was non-generic and called from a generic function.

EDIT: Indeed, inlining did work around the linker problem.

[alexcrichton]

Buildbot's on its way out, and I think this was resolved, so closing.