Re: make rule of multiple file generation (was Re: CVS commit: src/share/mk)

[christos%zoulas.com@localhost (Christos Zoulas)]

On Dec 14, 12:41pm, uebayasi%tombi.co.jp@localhost (Masao Uebayashi) wrote:
-- Subject: Re: make rule of multiple file generation (was Re: CVS commit: sr

| > So this is fixed with an additional dependency to force serialization,
| > or by fixing the script not to re-use the same temp filenames:
| > 
| > nodes.h: nodes.c
| > nodes.c nodes.h: mknodes.sh nodetypes nodes.c.pat
| >         ${_MKTARGET_CREATE}
| >     ${SCRIPT_ENV} ${HOST_SH} ${.ALLSRC:S/^nodes.c$//} ${.OBJDIR}
| >     [ -f nodes.h ]
| 
| You should have written this version at first. :)

Well, I did not undestand what the problem was at the time :-)
 
| This seems work for me.  I think I tried a similar one but clearly I figured
| out the trick of removing nodes.c from ${.ALLSRC}.
| 
| Next is usr.bin/ktruss.

That worked ok for me with -j 4.

| > Still you have not shown me why GENCMD is necessary... For me it adds
| > complexity and slows things down. Fixing parallelization in Makefiles
| > is simple and adequately done with the existing make features, or by
| 
| $GENCMD provides a way to extract only one output file.  It looks like
| complex because it's embedded in makefile.  I don't think it's really complex.
| It's just redundant. :)

Yes, it adds more explicit rules... My vote is to remove it and fix
the Makefiles/scripts that are broken. It is pretty straight-forward.

| > correcting the scripts that generate files to use unique filenames.
| 
| Actually I like this more.

We could do both, but forcing serialization is probably more efficient since
it causes the script to be invoked only once and avoids races.

christos