Marking as release-blocker for Go 1.18 because this failure mode appears to be new as of November.
It occurs during compilation, so it is not trivial to resolve by adding a test skip.

(CC @jeremyfaller for routing)

So far these failures have only been seen on the freebsd-amd64-12_2 builder, but with a sample size of N=3 it isn't obvious to me whether the underlying problem may affect other platforms.

The error is probably from the linker. May be the same cause s #49752.

I haven't been trying to reproduce this (I'm going after #46272 (comment)), but this has reproduced in 2/5 of my crashes so far with:

$ goswarm -i 25 freebsd-amd64-12_2 go/src/all.bash

@prattmic do you think this could be a result from the memory corruption from the other bug, or this is a different issue? Thanks.

It seems plausible, but I haven't seen enough details to have confidence one way or another.

Same as #49752 . We can revisit once we have FreeBSD builders with kernel fixed.

https://storage.googleapis.com/go-build-log/301db3f5/freebsd-amd64-12_2_faaeee15.log (a TryBot).

I think @heschi pushed an updated FreeBSD 13 builder today — is there a fix available for 12.2 as well?

#46272 (comment) noted that the fix is cherry-picked into FreeBSD 12 as well.

I believe https://lists.freebsd.org/archives/freebsd-snapshots/2021-December/000037.html is the latest 12 snapshot.

We're currently on 12.2, which doesn't have the fix. Would you like me to upgrade us to the latest 12.3 snapshot?

Since 12.2 has TryBot failures that we can't easily work around by adding test skips, I think we would add an updated 12.3 builder as soon as is feasible and turn off the 12.2 TryBots.

(Probably we should keep the 12.2 post-submit builders up until the 12.3 builders have run for enough CLs to be confident that there are no significant regressions from 12.2 to 12.3.)

Change https://golang.org/cl/375695 mentions this issue: dashboard,env/freebsd: add 12.3-STABLE

Change https://golang.org/cl/377474 mentions this issue: dashboard: clean up builders affected by memory corruption

I am pretty sure this falls into the same category as #49752, which is now marked fixed-- I am going to try one more goswarm test on this one to check that the problem is no longer reproducible.

OK, I ran some tests with goswarm overnight (10 motes running all.bash for roughly 20 hours) using the freebsd-amd64-12_3 builder, which incorporates the fix for the problem identified in #46272. No new failures with the "unreachable sym" mode. I am going to close out this bug.

For posterity, here are the failures that did turn up:

15 times:

--- FAIL: TestBlockProfile/debug=1 (0.02s)
    pprof_test.go:929: Bad chan send entry, expect:

7 times:

--- FAIL: TestStackGrowth (32.10s)
 .. stack_test.go:154: finalizer did not run

4 times:

 ##### Testing race detector
 --- FAIL: TestOutput (10.16s)
   output_test.go:83: failed test case test_fails_on_race, expect:

5 times:

# go run run.go -G=3 locklinear.go
exit status 2
panic: lockmany: too slow: 2000 ops: 16.141185ms; 4000 ops: 516.734873ms

then maybe a half dozen other singleton failures.

    pprof_test.go:929: Bad chan send entry, expect:

That's #6999, open since 2013 (and not FreeBSD-specific). 😩

    stack_test.go:154: finalizer did not run

That looks like #19381, but curiously it hasn't failed in the dashboard in the past few years. Maybe it's sensitive to machine load?

    output_test.go:83: failed test case test_fails_on_race, expect:

I don't see any issues filed at all for that one. 🤔

    panic: lockmany: too slow: 2000 ops: 16.141185ms; 4000 ops: 516.734873ms

#32986, open since 2019.