Re: Multiply distributed files

[Akim Demaille]

>>> "adl" == Alexandre Duret-Lutz <address@hidden> writes:

Oops, sorry for the delays, I hadn't noticed you had answered.

>>> "AD" == Akim Demaille <address@hidden> writes:
 AD> So the "dist" immediately below is executed with a read-only source
 AD> tree.  This seems wrong to me, I suggest restoring the rights first.

 > Why?  We ought to be able to run "make dist" from an up-to-date
 > read-only source tree.

OK, sound argument.

 > Removing these rights ensures this.  Restoring them would hide some
 > bugs, like people modifying source files during "make dist".

 > Maybe a less invasive change is to restore the rights only on the
 > directories that have been copied with `cp -p'? (I think this can be
 > done immediately after the `cp -p'.)

That wouldn't do it, since you don't know if the file or the directory
will be installed first.  Actually, maybe we should use install
instead of cp, and set the rights while we set up the tarball?  What
is the contract between Automake and its users wrt permissions of
files within the tarballs?  An exact copy of what is the source?  In
this case, it defeats the point you rose above, since in some
circumstances distdir uses permissions which are not those chosen by
the user.

 > IMHO EXTRA_DISTing directories should be only used for directories
 > that are generated automatically (e.g., Doxygen documentation).  If
 > the directory contains hand-maintained files (and hence backups, etc.),
 > EXTRA_DISTing it is too error-prone.

I agree, and I know and promoted this approach.  But for the case at
hand, I still believe there is a bug from Automake, highlighted by
this discouraged use of EXTRA_DIST.