runtime: gdb tests fail on NetBSD

[bradfitz]

The netbsd-amd64-8branch (NetBSD 8.0+) builder is back and NetBSD is kinda working for the first time in ages.

But only kinda.

The runtime tests fail with a timeout about two thirds of the time.

Examples:

https://build.golang.org/log/9ff11d9a995eb3b2555ce8f88bb3280c91639386
https://build.golang.org/log/d9a762c61874b62e34cec30f79ddc21a190922f7
https://build.golang.org/log/7ab9f1d000be47604a25b1ebf17fc55f5799d0ec
https://build.golang.org/log/6678e6c8dfb7533054836eb140f9eff8416e3a30

Many more at https://build.golang.org.

It'd be nice to have NetBSD happy for Go 1.10, considering the Go 1.9 news that NetBSD support was dead (https://golang.org/doc/go1.9#known_issues) which prompted a number of people to help fix things up.

/cc @aclements @ianlancetaylor @bsiegert

[gopherbot]

Change https://golang.org/cl/80136 mentions this issue: runtime: skip GDB tests on NetBSD

[ianlancetaylor]

In each failure a gdb test is timing out. Since the gdb tests are not mission critical, I think we may as well just skip them for now.

[krytarowski]

What are these GDB tests doing? If threading than this is not supposed to work. I work on improvements with ptrace(2) reliability.

[ianlancetaylor]

@krytarowski See runtime/runtime-gdb_test.go. I don't know what the actual problem is. I suggest reverting the CL and running go test runtime.

[bradfitz]

@krytarowski, if by "threading" you mean tracing a threaded program, then almost certainly yes. Go programs use threads behind the scenes like crazy.

[krytarowski]

Threads are the reason why golang breaks under GDB/NetBSD.

[aclements]

@krytarowski, thanks. Is this something that's expected to fixed in NetBSD in the near future? If not, I think we should go ahead and close this issue, since there's not much we can do about it.

Also, is there a bug tracking this in NetBSD? It would be nice to reference something canonical from the test skip message.

[bradfitz]

I found at least a recent presentation from @krytarowski: http://netbsd.org/~kamil/ptrace-netbsd/presentation.html

[krytarowski]

http://www.netbsd.org/~kamil/eurobsdcon2017.html a more recent one!

Upstream related Problem Report:

http://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=52548 or a prettier link: http://gnats.netbsd.org/52548

I've just finished porting TSan (upstream is ongoing). I'm going to switch to MSan soon and focus on LLDB and ptrace(2) fixes. This will take at least several months (I've estimated 11 months to go..).

[bsiegert]

http://gnats.netbsd.org/52548 seems to be the most closely related bug.

[gopherbot]

Change https://golang.org/cl/81195 mentions this issue: runtime: more specific reason for skipping GDB tests on NetBSD

[krytarowski]

We have added some quick fixes in gdb from the base that might fix basic tracing golang programs. The fixes landed -current.. 8.0 is getting closer so such upgrades might be too late now in -8. The corrections involve upgrade to GDB-8.0.1

Skipping the tests as of now is reasonable.

[bsiegert]

@krytarowski Would gdb-8.1 from pkgsrc work? We can get a current gdb installed on the builders, and perhaps bail out in the tests if the GDB is below 8.0.1.

[krytarowski]

Right now the basesystem and pkgsrc GDB are unrelated. The one in pkgsrc is as close to upstream as possible, while the basesystem one contains our patches from the past 20 years.

At the end both should end up as pristine upstream ones, working on it. ETA: after finishing LLDB and fixing ptrace(2).