Re: [Mingw-users] Case sensitiveness, Make 3.79, and Make 3.81

[Yongwei W. <wuy...@gm...>]

On 29/09/2007, Charles Wilson <cwi...@us...> wrote:
> Keith Marshall wrote:
>
> > so, Chuck's notion that the newer `all' file, corresponding to the phony
> > target of the same name, (identically here), will cause `foo' to be
> > remade doesn't seem to hold true; indeed it would appear to contradict:
>
> Sorry -- it looks like I misanalyzed the code.  I should have done what
> Keith did and just tested the proposition in actual practice. An ounce
> of experience beats a pound of theory, every time. [and on that point,
> see zlib example, below...]
>
> However, given the remainder of Keith's examples, I still /believe/ that
> while I was wrong on the specifics I was right on the generality: make
> -- even on MSYS *and* on cygwin [1] -- should not pretend the underlying
> filesystem is case-sensitive when it actually isn't.

*I* did not say it should (though I do not think it bad; see bottom of
my message).  However, please notice that your statement "even PHONY
targets are impacted by the underlying file system" is not true.

> [1] cygwin: it is true that the make-3.81 distributed by cygwin pretends
> to be case sensitive.  The only excuse for that, IMO, is the following:
>
> (a) most of the time, open-source projects are not so short-sighted as
> to rely on case sensitivity to distinguish files and targets -- so
> cygwin-make's insanity doesn't actually hurt anything.
>
> (b) in those rare cases where they do, and it is not feasible to patch
> the Makefile[.in|am] (and send that patch upstream), cygwin provides the
> "managed mount" feature to simulate case-sensitivity.

Cygwin is intended to emulate Unix behaviour.  When the Makefile
contains rules like what I originally presented, which did not require
the file system to be case-sensitive, it *should* work with any
complaints.

> If cygwin's make did not at least pretend to case-sensitivity, then the
> workaround for problematic packages mentioned in (b) would not be
> possible.  The cygwin team made the decision that the potential for
> confusion in distributing a make that is insane (e.g. believes in
> non-truth) is outweighed by the advantages allowed when using managed
> mounts.
>
> As MSYS does not provide managed mounts, that rationale is unsupportable
> for the msys version of make.

I can argue that MSYS is exactly like the Cygwin case.  I may even
further argue the native Win32 Make should work the same way as
expected in my example, since, *ahem*, the Unix principle never
prevents one from shooting on his feet.  As a tool, its duty is to do
its job as the *user* expects, when possible.

Don't say we are on Windows.  GNU is Unix!

This said, I think making phony targets different from file targets is
the most ideal way.  Since I am not planning to spend the labour
myself right now, I won't insist here it must be done.  I am still
ready to argue whether it is the best way :-).

> Regarding Earnie's post:
> > IMO, --case-insensitive-file-system should not be used for the MSYS
> > build of make.  The reason is that we are trying to allow the creature
> > comforts of POSIX as much as possible for ``configure && make'' and
> > this switch doesn't allow for the creature comfort.
>
> It is not a creature comfort if you have two non-PHONY targets:
>
> DEBUG:
>         touch DEBUG
>
> debug:
>         touch debug
>
> that clobber each other.

Agreed on this part.  However, whatever way Make behaves in this
example, the user has to modify his make rules.  Making Make
case-insensitive does not bring the user real benefits!

*Are we expecting the user to (be able to) write Debug when he wants
DEBUG?*  This looks to me the current --case-insensitive-file-system
is doing.

Best regards,

Yongwei

-- 
Wu Yongwei
URL: http://wyw.dcweb.cn/