Milter-greylist

Hello, list :)

   In late summer we had a discussion, which withered away without a
firm resolution (IMHO) about my proposed patches to enable properly
building some "sequential" components in case the overall build is
executed in parallel mode.

   The discussion mostly revolved around "why" (would one build this
small project in parallel in the first place) and strayed away from
the subject matter - whether this patch helps anyone (does help me)
or breaks things for anyone (one breakage was reported and fixed,
none others are at the moment known). The new ./configure options
introduced by the patch allow to specify any strings for the parallel
and sequential make's in case that some strange make program does
break, which is better than just specifying or omitting "-j 1" as
the currently released code does.

   I consider this patch successfully tested with GNU gmake, Sun make
and Tru64 make; success is at least building with no errors even if
all goes sequentially when a parallel build is requested, and at most -
with parallelization in place (and also no errors). That is, if there
is no regression (breaking of sequential builds, for example).

   Emmanuel reasonably desires that there is a general consensus about
inclusion of a feature that was finally discussed a lot. I hereby do
solicit "votes" for inclusion. The reason I ask is that I do want to
use this functionality, and I don't want to maintain a branch of the
milter-greylist project unnecessarily. And because this may help some
other users of the product as well.

   If you don't care and it doesn't break your builds - it is likely
a "yes" vote, but do please say so ;) If this feature interferes with
your ability to build (like it did for Attila) especially if the
./configure options to override don't help, please report the details
so that the configure automagic can be fixed.

   The Makefile.in/configure.ac patch updated for the 4.5.7 tarball is
attached. You can recreate the configure script with GNU autoconf:
:; gpatch -p1 < milter-greylist-4.5.7-makeseq-makepar.patch
:; rm ./configure && autoconf < ./configure.ac
and run a configure with no options (default) or your options, and a
build (which may become parallelized if support for that is detected).

   If things do break, see (and report) if overrides help:
   --with-make=PROG   Program name or full path to use as make
   --with-makeseq="$MAKE -j 1"   Program and params to use for 
sequential (single-job) make
   --with-makepar="$MAKE -j 8"   Program and params to use for parallel 
(multi-job) make
   --with-make-default-target="all"   Default target for make without 
explicit parameter

   Please don't shoot off this feature just because you don't use it,
and yet it doesn't make your life any bit harder ;)

Thanks in advance,
//Jim Klimov


On 2013-09-27 19:08, Emmanuel Dreyfus wrote:
>> So... just in case you change your mind
>
> My opinion is that a consensus should emerge on the mailing list on that
> topic. Could you please post the patch there for proposal?

Ok, will do... no backstage deals, huh? ;)

> Hello Manu,
>
> I was away from computers for most of September as "promised"; I've
> now revised the new release (to see if my builds beed to upgrade)
> by N-way comparing the code to what I've built and the CVS version
> and the tarball.
>
> In the CVS I see that there are autowhite.{c,h} files which are both
> absent from tarball releases and are not referenced in Makefiles, in
> #include directives, etc. Are they cruft that should be moved to attic?
>
> Also, both CVS and the tarball release don't include my recent (August)
> fixes to the Makefile.in and configure.ac for more flexible parallel or
> sequential make program control... I am aware of the discussion that
> has been on-list regarding this, but believe the currently checked-in
> solution to be inferior to mine (i.e. it does not really allow to run
> a paralel make with sequential pieces on a platform which does not
> use literally "make -j 1" to disable parallelism, while my fix allows
> the building user to specify any platform-dependent strings and to
> guess proper defaults if/when we become aware of more other make
> programs that might be relevant to milter-greylist).
>
> So... just in case you change your mind, I re-submit this makefile
> patch (tested to apply against the 4.5.7 tarball).
>
> //Jim  Klimov