Bug#1049434: gtk4: FTBFS on riscv64: SIGSEGV in gtk:gtk / templates test

[Aurelien Jarno]

On 2023-08-15 22:34, Aurelien Jarno wrote:
> On 2023-08-15 20:19, Aurelien Jarno wrote:
> > Source: gtk4
> > Version: 4.10.3+ds-1
> > Severity: important
> > Tags: upstream ftbfs
> > Justification: fails to build from source
> > X-Debbugs-Cc: debian-riscv at lists.debian.org
> > 
> > Dear maintainer,
> > 
> > As part of the riscv64 bootstrap, I have been trying to build and import
> > gtk4 (involves building it with a mix of the official archive and the
> > old debian-ports archive). Unfortunately it fails to build from source
> > with a failure in the testsuite:
> > 
> > | 262/722 gtk:gtk / templates                                                                         ERROR             8.58s   killed by signal 5 SIGTRAP
> > 
> 
> I am also able to reproduce the issue on amd64 (although mostly with
> SIGTRAP, only rarely SIGSEGV), when not using a separate $HOME directory
> like done in debian/run-tests.sh.
> 
> A strace shows a lot of stat on files, and I wonder if it's not just a
> timing issue: scanning all those files or a slow architecture are
> triggering the issue.

I confirm this is a timing issue, I am able to workaround it on both
amd64 (using the above trick) and riscv64 with the following patch:

--- gtk4-4.10.5+ds.orig/testsuite/gtk/templates.c
+++ gtk4-4.10.5+ds/testsuite/gtk/templates.c
@@ -41,7 +41,7 @@ show_and_wait (GtkWidget *widget)
 {
   gboolean done = FALSE;
 
-  g_timeout_add (500, main_loop_quit_cb, &done);
+  g_timeout_add (1500, main_loop_quit_cb, &done);
   gtk_widget_set_visible (widget, TRUE);
   while (!done)
     g_main_context_iteration (NULL, FALSE);

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                     http://aurel32.net