Re: debian testing

From: Daniel Black <daniel.subs_at_internode.on.net>
Date: Mon, 23 Nov 2009 10:09:26 +1100

On Monday 23 November 2009 08:26:50 Murray S. Kucherawy wrote:
> I completed the merge manually, but a few issues remain.
>
> On Sun, 22 Nov 2009, Daniel Black wrote:
> > The last commit I did regarding AC_SEARCH_LIBS(milter) introduced a
> > little configure uglyness:
> > "
> > checking for milter library and includes... checking for library
> > containing smfi_register... -lmilter
> > /usr
> > "
> >
> > (but it was needed to make it compile on a lenny machine)
>
> I had to comment this out. The test that it does links against libmilter
> which expects to be built multi-threaded, but the configure script runs a
> build without the -pthread option to gcc; so even though it actually does
> find the symbol you're looking for, the test fails because various pthread
> functions couldn't be found, and the script aborts.
ok - added some code (commented) that should fix that.

>
> I'll see if I can get it working later today but it's commented out for
> now. Is there a case where the existing test finds libmilter but that
> doesn't contain smfi_register()?
I don't imagine so. Its probably the oldist milter symbol. If it does find
such a version its not that useful to us/

>
> > 2.
> >
> > Currently we're checking strongly against libtre-0.8.0 by looking for
> > tre/tre.h (which is only in the 0.8.0 version. Can we use 0.7.5 as this
> > is the stable version on lenny?
> > yes - change check in configure.ac and source code to use tre/regex.h
> > no - change the error to explictly mention we require the 0.8.0 version
>
> This should be done through build-config.h and the AC_HEADER check macros,
> but I'm fine with supporting either version as long as it doesn't make our
> code too #ifdef-heavy because of lots of tre API differences.
>
> > 3.
> >
> > We almost could make this work on the libmilter-8.13 found in debian etch
> > and potentially other old stable machines. The only compile time symbol
> > not found was mfi_setsymlist to do with macro checking. A configure check
> > for this and subsequent disabling of macro features could increase
> > portability.
>
> I've added the #ifdef for SMFIF_SETSYMLIST to opendkim.c, so that should
> cover this case.

Great - an AC_SUBST(MACRO_MANNOTICE) so the user isn't expecting it to work
could be good too.

>
> Other points:
>
> The configure.ac LIB_FEATURE macro still doesn't seem to be working;
> build-config.h doesn't contain a QUERY_CACHE definition with or without
> --enable-query_cache.
>
ok fixed this.

> It also defines LIBOPENDKIM_PARSETIME even when I don't request it.
> Probably tied to LIB_FFR_FEATURE.

fixed too.

remaining bug LIBOPENDKIM_FEATURE_STRING is constant of all features (enabled
or not)

>
> This evening I'll try various other --enable and --with tests to see if
> anything else has gone awry.
>
> I really want to start betas ASAP, but not with this many issues
> outstanding.
>
understand.
Received on Sun Nov 22 2009 - 23:10:36 PST