You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@serf.apache.org by "Daniel Sahlberg (Jira)" <ji...@apache.org> on 2026/01/04 21:42:00 UTC

[jira] [Created] (SERF-211) Finalise async resolver

Daniel Sahlberg created SERF-211:
------------------------------------

             Summary: Finalise async resolver
                 Key: SERF-211
                 URL: https://issues.apache.org/jira/browse/SERF-211
             Project: serf
          Issue Type: Bug
            Reporter: Daniel Sahlberg


The async resolver code contains a bunch of FIXMEs:

 

resolve.c

/*
 * FIXME: EXPERIMENTAL
 * TODO:
 *  - Wake the poll/select in serf_context_run() when new resolve
 *    results are available.
 */

 

init_once.c

/* FIXME: Do we really want to support APR-0.9.x? */  {color:#FF0000}(no?){color}

return APR_EGENERAL; /* FIXME: Just abort()? */ {color:#FF0000}(as the previous comment says: It can't happen, was there some stupid compiler complaining?){color}

 

serf.h

/* FIXME: EXPERIMENTAL */

(four times, on the declarations of the async functions/callbacks)

 

 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Re: [jira] [Created] (SERF-211) Finalise async resolver

Posted by Branko Čibej <br...@apache.org>.
On 4. 1. 26 23:21, Daniel Sahlberg wrote:
> Den sön 4 jan. 2026 kl 22:56 skrev Branko Čibej<br...@apache.org>:
>
>> On 4. 1. 26 22:44, Daniel Sahlberg wrote:
>>> So I added this and linked from SERF-208. I think removing the FIXME's
>> is a
>>> blocker (not that it is very difficult), except possibly the TODO.
>> As I commented in the issue: we can decide that the async resolver is
>> not experimental. It's a new API with limited testing.
>>
> What is needed if we decide this is experimental? Is it enough to just
> document it as experimental?


It should be. we could add a #define to the headers and do some magic so 
that changes in the ABI could be hidden from the source level.


>> The TODO will have to stay for now because I haven't been able to find a
>> way to reliably awake the context. This is because adding or removing
>> file descriptors to the pollset can be completely outside Serf's
>> control, handled by the application. Otherwise we'd just add a wakeup
>> pipe and be done with it.
>>
>> The thing to watch here (and document) is that any new resolved
>> addresses may not be noticed until the poll timeout has expired, so if
>> an application uses a really long poll delay, this could be a bit of a
>> pain. That's not really recommended for async I/O, of course -- there
>> should ideally be no delay.
>>
>> -- Brane
>
> In [1] you mention testing this with Subversion. What would be needed for
> SVN to use this?

I haven't looked at that.


> /Daniel
>
>
> [1]https://lists.apache.org/thread/tgv733ss8l5pgxtjsz8zmmc6nx49lbbb

Re: [jira] [Created] (SERF-211) Finalise async resolver

Posted by Daniel Sahlberg <da...@gmail.com>.
Den sön 4 jan. 2026 kl 22:56 skrev Branko Čibej <br...@apache.org>:

> On 4. 1. 26 22:44, Daniel Sahlberg wrote:
> > So I added this and linked from SERF-208. I think removing the FIXME's
> is a
> > blocker (not that it is very difficult), except possibly the TODO.
>
> As I commented in the issue: we can decide that the async resolver is
> not experimental. It's a new API with limited testing.
>

What is needed if we decide this is experimental? Is it enough to just
document it as experimental?


>
> The TODO will have to stay for now because I haven't been able to find a
> way to reliably awake the context. This is because adding or removing
> file descriptors to the pollset can be completely outside Serf's
> control, handled by the application. Otherwise we'd just add a wakeup
> pipe and be done with it.
>
> The thing to watch here (and document) is that any new resolved
> addresses may not be noticed until the poll timeout has expired, so if
> an application uses a really long poll delay, this could be a bit of a
> pain. That's not really recommended for async I/O, of course -- there
> should ideally be no delay.
>
> -- Brane


In [1] you mention testing this with Subversion. What would be needed for
SVN to use this?

/Daniel


[1] https://lists.apache.org/thread/tgv733ss8l5pgxtjsz8zmmc6nx49lbbb

Re: [jira] [Created] (SERF-211) Finalise async resolver

Posted by Branko Čibej <br...@apache.org>.
On 4. 1. 26 22:44, Daniel Sahlberg wrote:
> So I added this and linked from SERF-208. I think removing the FIXME's is a
> blocker (not that it is very difficult), except possibly the TODO.

As I commented in the issue: we can decide that the async resolver is 
not experimental. It's a new API with limited testing.

The TODO will have to stay for now because I haven't been able to find a 
way to reliably awake the context. This is because adding or removing 
file descriptors to the pollset can be completely outside Serf's 
control, handled by the application. Otherwise we'd just add a wakeup 
pipe and be done with it.

The thing to watch here (and document) is that any new resolved 
addresses may not be noticed until the poll timeout has expired, so if 
an application uses a really long poll delay, this could be a bit of a 
pain. That's not really recommended for async I/O, of course -- there 
should ideally be no delay.

-- Brane

Re: [jira] [Created] (SERF-211) Finalise async resolver

Posted by Daniel Sahlberg <da...@gmail.com>.
So I added this and linked from SERF-208. I think removing the FIXME's is a
blocker (not that it is very difficult), except possibly the TODO.

/Daniel


Den sön 4 jan. 2026 kl 22:42 skrev Daniel Sahlberg (Jira) <ji...@apache.org>:

> Daniel Sahlberg created SERF-211:
> ------------------------------------
>
>              Summary: Finalise async resolver
>                  Key: SERF-211
>                  URL: https://issues.apache.org/jira/browse/SERF-211
>              Project: serf
>           Issue Type: Bug
>             Reporter: Daniel Sahlberg
>
>
> The async resolver code contains a bunch of FIXMEs:
>
>
>
> resolve.c
>
> /*
>  * FIXME: EXPERIMENTAL
>  * TODO:
>  *  - Wake the poll/select in serf_context_run() when new resolve
>  *    results are available.
>  */
>
>
>
> init_once.c
>
> /* FIXME: Do we really want to support APR-0.9.x? */
> {color:#FF0000}(no?){color}
>
> return APR_EGENERAL; /* FIXME: Just abort()? */ {color:#FF0000}(as the
> previous comment says: It can't happen, was there some stupid compiler
> complaining?){color}
>
>
>
> serf.h
>
> /* FIXME: EXPERIMENTAL */
>
> (four times, on the declarations of the async functions/callbacks)
>
>
>
>
>
>
>
> --
> This message was sent by Atlassian Jira
> (v8.20.10#820010)
>