Re: Creating subdirs of distdir

[Ralf Wildenhues]

Hi Stepan,

* Stepan Kasal wrote on Tue, May 23, 2006 at 10:43:50PM CEST:
> On Tue, May 23, 2006 at 09:31:18PM +0200, Ralf Wildenhues wrote:
> 
> > I think we can kill the per-file forks before the mkdir_p, too.
> 
> You mean the $(srcdir) and $(top_srcdir) stripping.  But it takes
> place only in certain situations:
> - if the make implementation inserts $(srcdir) (GNU make doesn't)

Right.

> - or if the author of Makefile.am usees $(srcdir) or $(top_srcdir)
>   (well educated maintainers don't)

This is wrong.  Well-educated maintainers do this in some situations:
when the file in question is also listed as prerequisite somewhere (and
even $(EXTRA_DIST) is listed as prerequisite somewhere), so that stuff
*really* works with non-GNU make.

In fact, typical Makefiles generated by Automake contain at least one
such file: $(srcdir)/Makefile.in.

> So when you replaced all the per-file forks by a constant number of
> three forks, you actually slowed down the most common usecase.

Even if that were a common case, that would not be clear (it's not clear
3 forks and a sed script are worse that a bunch of shell matching).
But it's best not to rely on intuition when optimizing, so here we go
(`\time make distdir' of the Automake package itself, on GNU/Linux):

before:
3.33user 1.44system 0:05.46elapsed 87%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+740075minor)pagefaults 0swaps

your patch:
1.79user 0.58system 0:03.36elapsed 70%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+238784minor)pagefaults 0swaps

my patch:
0.65user 0.56system 0:01.87elapsed 65%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+211732minor)pagefaults 0swaps

> I think it should be either reverted, or wrapped in a case command,
> see below.

I haven't applied anything here; it is Alexandre's decision to apply
(or tell me to decide).

> Moreover, your sed program is not portable; you cannot use semicolon
> after the `t' command nor after the `:' command.

Right, thanks for catching.

> What about the following (quick untested sketch):

Some comments inline:

> > +   list='$(DISTFILES)'; \
> > +## The `case' command eliminates unnecesary forks in most of the cases
> > +## when $list would stay intact.
> > +     case $$list in $(srcdir)/*|$(top_srcdir)/*) \

(trivial omissiong of two leading `*')

> > +     list=`for file in $$list; do echo $$file; done | \
> > +       | sed -e "s|^$$srcdirstrip/||;t" \
> > +             -e "s|^$$topsrcdirstrip/|$(top_builddir)/|;t"`;; \
> > +##         (The second `t' command clears the flag for the next round.)

This comment and the sed script isn't right: the second `t' won't be
visited if the first one hit.

> > +   esac; \
> > +##
> > +## Make the subdirectories for the files.
> > +## (The DISTDIRS list can be incomplete, because files in subdirectories 
> > can
> > +## be specified for `dist' conditionally.)
> > +##
> > +   case $$list in */*) \
> > +     ( cd "$(distdir)"; \

At this point, $list is not guaranteed to be newline-separated now.

> > +       $(mkdir_p) `echo "$$list" | sed -n 's,/[^/]*$$,,p' | sort -u`; \
> > +     ) ;; \
> > +   esac; \
> > +##
> > +##
> > +   for file in $$list; do \

Cheers,
Ralf