Re: (fwd) Bug#557528: tin: feed_articles() segfaults on spurious SIGPIPE

On 22.11.2009, at 20:11, Urs Janßen wrote:

> I'm seeing (very) infrequent segfaults when feed-saving articles.
> Apparently, this occurs when tin gets a SIGPIPE, probably from writing
> to the network socket.  (I have a very flaky DSL, so it's quite  
> possible
> the connection sometimes get dropped just when tin is about to write.)
>
> When that happens, pclose(pipe_fp) will segfault, since pipe_fp is  
> NULL.
> ----- End forwarded message -----

When receiving SIGPIPE, got_sig_pipe becomes TRUE in  
signal.c:signal_handler().

feed.c:feed_articles() calls handle_SIGPIPE() which expands to:

if (got_sig_pipe) goto got_sig_pipe_while_piping

Changing the macro to:

if (got_sig_pipe) { \
	if (function == FEED_PIPE) \
		goto got_sig_pipe_while_piping; \
	else \
		got_sig_pipe = FALSE; \
	}

might solve this issue. The "goto" is only useful when 'pipe'ing  
articles (FEED_PIPE). Other FEED_* functions must not jumpt to the  
label.

Dennis