On Wed, Dec 19, 2012 at 8:31 PM, Philip Martin
<philip.mar...@wandisco.com> wrote:
> "C. Michael Pilato" <cmpil...@collab.net> writes:
>
>> On 12/19/2012 01:59 PM, C. Michael Pilato wrote:
>>> On 12/19/2012 01:53 PM, Daniel Shahaf wrote:
>>>> Actual result: backgrounding svn killed it.
>>>>
>>>> Expected result: after 'bg\n', keep fetching+filtering+printing log
>>>> messages in the background.
>>>>
>>>> It worked in 1.7 with neon. Can we make it work in 1.8 too?
>>>
>>> Please file an issue. With as much code and complaint flying around as we
>>> have, sometimes my view is: "No issue, no problem." But I suspect the
>>> issue needs to be filed against Serf, not Subversion. The error you show is
>>> simply what is returned from serf_context_run().
>>
>> Sorry: to clarify, it's a Subversion-generated error message which wraps
>> the status code returned from serf_context_run().
>
> Yes, I can reproduce it on my Linux machine but only if the interrupt
> occurs in an epoll_wait call:
>
> epoll_wait(3, ^Z
> [1]+ Stopped strace subversion/svn/.libs/lt-svn log
> http://svn.apache.org/repos/asf/subversion
> $ fg
> strace subversion/svn/.libs/lt-svn log
> http://svn.apache.org/repos/asf/subversion
> 7fb0234461e8, 16, 500) = -1 EINTR (Interrupted system call)
> --- SIGCONT (Continued) @ 0 (0) ---
On Ubuntu I can reproduce this issue at the first try, On Mac OS X not at all.
Attached patch to serf fixes the issue for me.
[..]
Lieven
Index: context.c
===================================================================
--- context.c (revision 1701)
+++ context.c (working copy)
@@ -289,6 +289,11 @@ apr_status_t serf_context_run(
if ((status = apr_pollset_poll(ps->pollset, duration, &num,
&desc)) != APR_SUCCESS) {
+ /* EINTR indicates a handled signal happened during the poll call,
+ ignore, the application can safely retry. */
+ if (APR_STATUS_IS_EINTR(status))
+ return APR_SUCCESS;
+
/* ### do we still need to dispatch stuff here?
### look at the potential return codes. map to our defined
### return values? ...
