Ian Jackson <ijack...@chiark.greenend.org.uk> wrote:
>> The invocation of gpg-agent by gpg frontend has an inherent race in the
>> current implementation; When gpg frontend invokes gpg-agent, after
>> spawning gpg-agent, gpg frontend tries to connect five times with one
>> second intervals. When a machine is busy enough and scheduling of
>> processes goes not that fair, the connection to gpg-agent from gpg
>> frontend may fail.
>
> This might be relevant. My test suite failures occur on a loaded
> machine, because the test case runner parallelises separate tests.
>
> Does this produce a particular message, when it occurs ?
Yes.
> There are other possible approaches to this situation than to hope to
> win the race. But I hesitate to suggest adding additional code to
> this area, having seen what's there already...
>
> If you would like to suggest a debug printf, or name lines of source
> code, I can probably arrange to see if this is happening.
Here is the lines.
diff --git a/common/asshelp.c b/common/asshelp.c
index f3a92f9e5..8c584fb4f 100644
--- a/common/asshelp.c
+++ b/common/asshelp.c
@@ -451,6 +451,8 @@ start_new_gpg_agent (assuan_context_t *r_ctx,
break;
}
}
+ if (i >= SECS_TO_WAIT_FOR_AGENT)
+ log_error ("agent invoked successfully, but...\n");
}
}
--
