Re: "make dist" calls "automake" ??

[Guido Draheim]

Tom Tromey wrote:
> 
> >>>>> "Akim" == Akim Demaille <address@hidden> writes:
> 
> Akim> Please, don't do that.  Using missing is much more appropriate.
> Akim> For instance, only missing will help independently of the
> Akim> version of Automake used by the user.  If it's old enough, then
> Akim> Makefile.in is produced, and if automake fails (e.g., version
> Akim> too old), then Makefile.in is preserved.
> 
> But the point is that in the particular case we are talking about,
> preserving the Makefile.in is *wrong*.  You'd end up with a broken
> distribution.
... 'cause the makefile.in must be copied...

> 
> Since the code is gone, this doesn't matter, except as a philosophical
> issue.
...  there are other places...

> 
> And philosophically I don't see any problem with requiring that `make
> dist' be done by a person with the right tools available in $PATH.  In
> general we don't require this, but `dist' is a bit special.
> 
In general, people will not expect it to be required - if none of the
autotools input files have changed, it looks a bit outerspace to have
the autotools output files be remade. This is a philosophic theme, and
I stand in the position that this make-like timestamp remake-rule
should be honoured as far as possible. Yes, it may be *acceptable* to 
require special requirements around calls to "make dist", however it 
should ensure to guard if users fail to see this requirement and keep
to the intuitive remake-rule which reads: remake is done when needed
and not when cleaning things up, not even for the case of just
packing things into a tarball.

As a sidenote - this problem may have been present long before and may
still be in current versions, but it was not noticable as the differences
of behaviour between versions of autotools was not that great. E.g.
the call to autoheader of any version may have yielded the same result
but now autoheader removes the correct one and places a config.h.in
that has no defs because it did not find configure.in. While the call
to "automake" would not affect the original, the "autoheader" even 
does so (as noted in that other thread).

Personally, I'd like to have make-dist disabled without maintainer-mode
to prevent inappropriate usage - if it needs to change basic files
coming from autotols. Actually, "philosophically", have automatic 
remake  of autotools files (via make timestamp rules or other rule 
implications) be disabled always without maintainer-mode, just as it 
had been within many (all?) other rules for a looooong time. IMHO, the 
easiest might be to add an option to `missing` instead of the "-run" 
that says (something like) "-warn" - and when called, it would be talking
the user saying "the make program did see a need to call xxx but it 
is currently disabled (without maintainer-mode?)". This is just a
(global) update to the am_maintainer_mode macroisms and their usage in
makefile.in generation. Even more, it would also shield about other 
"over"-usage of autotools within generated Makefiles, reaching a bit
ahead while we have to check it does not break some user's rules that
reference autotools (for whatever reasons).
Whaddaya think? Any chance to go this way?

cheers, Guido