The differences as per the second patch: linux32: https://public-artifacts.taskcluster.net/LwRx7H0jQ5yzB2cakHgbPg/0/public/diff.html linux64: https://public-artifacts.taskcluster.net/IAut2huJRFmmSJbZ70RKUw/0/public/diff.html Now, for an explanation of the differences (since there are differences, and you can understand why I wanted to reduce them by explicitly disabling .init_array/.fini_array): - Differences in Build ID in notes sections and .gnu_debuglink are due to the fact that the binaries vary (note that debug sections are included in the hashing done for both, so changes in debug info can also trigger that. Changes in debug info can be change in order, change in file paths, change in accuracy, etc. ; we've never concerned ourselves much about debug info changes) - Differences in .dynstr sections are unclear as to why they happen, but all that has no real impact. The contents of .dynstr are used for the .dynsym list, and while that one actually is different because of the offset changes in .dynstr, the list of symbols is actually the same, in the same order. - Inversion of 2 bytes in lock cmpxchg and lock xadd instructions. AIUI, those are actually 2 independent prefix opcodes: F0 is for LOCK, and 66 is an operand-size override prefix, and it doesn't matter in which order they appear. - Replacement of 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 with 0f 1f 40 00 66 2e 0f 1f 84 00 00 00 00 00. Those appear to be very large nops to fill space between functions.

Comment on attachment 8939054 [details] Bug 1427316 - Use the binutils we just built to build GCC. https://reviewboard.mozilla.org/r/209472/#review215214

Comment on attachment 8939055 [details] Bug 1427316 - Check differences between builds with and without the changes. https://reviewboard.mozilla.org/r/209474/#review215216 I think it would be worthwhile to have a task that ran diffoscope between the just-produced build and the previous build (from DAG traversal). But this is obviously scope bloat.

(In reply to Gregory Szorc [:gps] from comment #5) > Comment on attachment 8939055 [details] > Bug 1427316 - Check differences between builds with and without the changes. > > https://reviewboard.mozilla.org/r/209474/#review215216 > > I think it would be worthwhile to have a task that ran diffoscope between > the just-produced build and the previous build (from DAG traversal). But > this is obviously scope bloat. I'm not really convinced. Small C++ or rust changes would cause massive diffs. And changes to xul/js would just create the same diff that landed... Only for other peripheral changes this /could/ be useful. Also note that only Linux builds have limited differences when nothing else than MOZ_BUILD_DATE and the changeset differ.

Pushed by mh@glandium.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/2ef8585fcee1 Use the binutils we just built to build GCC. r=gps

Backed out changeset 2ef8585fcee1 (bug 1427316) for failing Linux opt Executed by TaskCluster build-linux-rusttests/opt tc(BR) https://treeherder.mozilla.org/logviewer.html#?job_id=153584109&repo=mozilla-inbound&lineNumber=33344 https://hg.mozilla.org/integration/mozilla-inbound/rev/fe802396d3dd5ec2f08175845666a8283022837d https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&revision=2ef8585fcee116b5b787e4f15d99e13ab0786b52&filter-classifiedState=unclassified&filter-resultStatus=testfailed&filter-resultStatus=busted&filter-resultStatus=exception

Those failures are due to the now filed bug 1427404.

Pushed by mh@glandium.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/43d4a4750c0e Use the binutils we just built to build GCC. r=gps