Re: Automake 1.4l released

[Tom Tromey]

>>>>> "Earnie" == Earnie Boyd <address@hidden> writes:

Earnie> Automake is a sister tool to Autoconf and should maintain the
Earnie> same effort to maintain portability.

That's true.  But we're talking about the capability to run `make
distcheck' on a platform where the semantics are not Unix-like in an
unanticipated way.

I don't have a problem working around bugs in vendor tools.  We do
that all the time in automake.  However, I prefer that free software
be fixed at the source as well.  That is, we might implement a
workaround in automake, but I dislike using that as an excuse to leave
other free tools unfixed.

In this case the Cygwin kernel divergeces from traditional Unix usage
in a subtle way -- on Unix you can use utime() on a `a-w'
(readonly-by-permission) file.  On Cygwin (I'm guessing!) you can't.

I don't know Cygwin very well.  And I certainly wouldn't try to tell
the Cygwin developers what to do.  If this is just an oversight, then
my preference would be to fix it in Cygwin.  If it is fundamental to
Cygwin for some reason, then I would suggest we fix it in `cp'.
Automake probably isn't the only tool that relies on `cp -p'.

Earnie> For other reasons, you are causing a HOALOW for little gain on
Earnie> the tool side.  The gain for Automake is what, "To test the
Earnie> distribution for CDROM release"?

I don't know what "HOALOW" means.

The gain for automake is that we can now test for a real bug which is
easy to accidentally introduce into a distribution.  The `distcheck'
target exists *only* to do testing to make sure the distribution is
correct.  You make it sound like this isn't very important, like we
should simply disable this feature.  Well, I disagree with that.

Earnie> At least test to see if `touch'ing a read only file will cause
Earnie> problems and work around them automagically if it does.

This makes a presumption about the implementation of `touch'.  Some
versions of `touch' are implemented using open()/write().  Those will
fail.

The real question is about the behavior of `cp -p', which really
should not depend on the mode of the source file (provided it is
readable).  Sure, we could test to make sure `cp -p' works.  And we
might, because there is obviously a bug here.  But adding a workaround
to Automake doesn't mean the bug is gone.

Tom