On Wed, 2021-02-03 at 18:52 +0100, Samuel Thibault wrote:
> Hello,
>
> It seems that the go issues that we were meeting (segfaults etc.)
> were due to the missing support for unwinding over signal handling
> with siginfo support. The fix is already committed upstream, I am
> uploading a fixed patch to debian-ports, and requeued the golang
> packages that seemed affected.
Do you mean the patch:
glibc (2.31-10~0) unreleased; urgency=medium
* debian/patches/hurd-i386/unsubmitted-tiocflush.diff: Cope with
BSD 4.1-ish ioctl(..., TIOCFLUSH, NULL).
That patch does not improve the go issues. I've re-run the go and libgo
tests from gcc-snapshot-20210108-1. The results are the same. For
example for libgo, see especially the SIGABRT failures:
=== libgo Summary ===
# of expected passes 169
# of unexpected failures 13
Looking at the errors more closely shows:
FAIL: cmd/go/internal/fsys: overlaid files can't be opened for write,
want nil
FAIL: html/template: SIGABRT: abort: failed to throw unwind exception
(reason=5)
FAIL: log/syslog: setsockopt: protocol not available
FAIL: net: setsockopt: protocol not available
FAIL: net/http: Segmentation fault (core dumped)
FAIL: net/http/cgi: error while loading shared libraries: libgo.so.19
FAIL: os/signal: Segmentation fault
FAIL: reflect: SIGABRT: abort: failed to throw unwind exception
(reason=5)
FAIL: runtime: SIGABRT: abort: failed to throw unwind exception
(reason=5)
FAIL: runtime/pprof: total 0 CPU profile samples collected: too few
samples
FAIL: sync/atomic: SIGABRT: abort: failed to throw unwind exception
(reason=5)
FAIL: text/template: SIGABRT: abort: failed to throw unwind exception
(reason=5)
FAIL: time: SIGABRT: abort: failed to throw unwind exception (reason=5)
