[Libwebsockets] [PATCH] Add Linux io_uring in the list of recognized libev backends

Olivier Langlois olivier at olivierlanglois.net
Sat Feb 22 20:26:01 CET 2020


it makes sense. From now on, I'll do that next time I encounter the
same type of situation.


On Fri, 2020-02-21 at 08:11 +0000, Andy Green wrote:
> On 2/21/20 4:38 AM, Olivier Langlois wrote:
> > and also check libev version and only include the 2 latest backends
> > only if
> > lws is compiled against a recent enough libev version.
> Thanks... I modified it the same way as I modified the AIO patch
> before 
> pushing that, to check at cmake if the symbol exists in the lib
> instead 
> of try to guague it by version, and pushed it on master.
> https://libwebsockets.org/git/libwebsockets/commit?id=caaedd65883d195e50abf6af03888d5ffbc0e00c
> Originally we tried to be compatible across openssl versions using
> the 
> version tests way, but there were two problems, 1) dev versions may
> have 
> the change in their code we're testing for, but not have the version 
> step we use to judge it happen until the release, and 2) stuff can 
> subsequently get deprecated meaning more breakage and work
> researching 
> the version it stopped existing in, and fixing up these tests long
> after 
> we have forgotten about this.  Mbedtls has the same problem to a
> lesser 
> extent, and a new kind of problem 3) it's quite build-time
> configurable 
> and you can switch off availability of apis easily by build-time 
> choices.  The version might tell you what's in the source but the
> user 
> or some upstream provider may have disabled the related apis for
> build.
> Empirically testing for the symbols with a compiliability check (it 
> doesn't need to run the code, so it's compatible with cross) at 
> build-time avoids those problems.
> -Andy

More information about the Libwebsockets mailing list