[FFmpeg-devel] [PATCH/RFC] Prefer getaddrinfo over gethostbyname
Mon Jan 4 13:04:00 CET 2010
On 04/01/10 12:52, Martin Storsj? wrote:
>> Maybe, the "non CONFIG_IPV6" functions can be disabled/removed one by one,
>> instead of removing the whole "non CONFIG_IPV6" at one time?
>> I mean, you can start by removing the functions that do not use getnameinfo
>> and multicast macros... Just an idea.
> I'm not sure if the functions can be upgraded one at a time or that
> requires more refactoring - I'll look into it later if you don't get there
Well, the "incremental upgrade" obviously requires some work and
modifications on the existing code. If this makes the transition more
complex instead of simpler, let's just drop the "incremental upgrade" idea.
>> But let's keep in mind that 90% of the times (or maybe even 99%) when we use
>> UDP we use numeric addresses, so there are no symbolic names to be resolved,
>> and it should be immediately clear if an address is IPv4 or IPv6.
> Well, the udp protocol is used internally within the rtp protocol,
> launched e.g. from the rtsp demuxer, where it probably would happen to use
But in this case (receiving RTP packets over UDP) you do not really need
to resolve a remote host address (I don't know, maybe the current code
does it, but it is not really needed...).
>>> But given this, I guess a better approach would be to simply drop this
>>> patch (it actually feels a bit redundant, since it does the same as the
>>> resolve_host fix in patch 2). Later (after the initial getaddrinfo support
>>> is commited?) we could work on adding more compatibility wrappers for
>>> making the "ipv6 codepath" in udp.c the only one.
>> AFAIR, this was the original plan. But I do not want to stop or slow down
>> things, so if people feel that a different plan is better, let's go for a new
>> plan. My only suggestion is that the disabling/removal of the "non
>> CONFIG_IPV6" codepath can probably be done incrementally (I hope :).
> Ok, so lets drop the current patch for the udp protocol, and instead aim
> at cleaning it up and merging the codepaths later, after the first patches
> have been commited.
If you are fine with this, good.
My impression is that at the end the codepaths merge will just consist
in removing the "#else" part of "CONFIG_IPV6".
More information about the ffmpeg-devel