Re: [mosquitto-dev] Mosquitto 2.0.19 released

[Greg Troxel <gdt@xxxxxxxxxx>]

Roger Light via mosquitto-dev <mosquitto-dev@xxxxxxxxxxx> writes:

> Could you please check the current `fixes` branch? I've tested the
> idea in a virtual NetBSD so hopefully it should be fine.

For reasons I am not clear on, the make-based build of master works.
Probably somehow it does not trigger the bug.  I need to set up scripts
for cmake to point it at dependencies because apparently it doesn't just
use pkgconfig, or there's something else going on.

Therefore, I took the commit that rototilled pthread support and added
it as a patch to my pkgsrc config, and then I was able to have a clean
build.

I haven't tried to run it yet, but will do that soon.

I would suggest that you amend the commit message to explain that this
is cleaning up the mechanism by which pthread support is turned on and
off to make it more globally consistent, avoiding a bug which seems to
be rarely expressed but was seen on NetBSD with pkgsrc.  Maybe that take
of mine is confused, but the new way seems much more straightforward and
less prone to breakage if the way the OS provides pthreads is not as
expected (but still meets POSIX).   Basically using #define on a token
specified by standards smells like UB and best avoided -- but again
maybe I am confused.

Thanks for providing a fix.  It would be great to have a quick-turn
2.0.20 once the other issues settle out.


I mailed you offlist about a problem with the cmake support for
websockets_shared, which is exposing what I think is a latent bug in the
cmakefiles installed by libwebsockets, which always add -ldl even on
platforms where that doesn't exist.  For others, the essence of my
comment is that allowing individual libraries to be shared or not seems
like complexity best avoided, and that it would be nice to just rely on
pkg-config and not have cross-project cmake tentacles.  But perhaps that
is the cmake way and it's just that I have not drunk the Kool-Aid.