Re: ok, new libtool for cygwin updates

["Robert Collins" <robert dot collins at itdomain dot com dot au>]

Hi edward,
    I'm not sure whether you want blow by blow bug reports, or a
summary?

So far, I've pulled down CVS HEAD automake, extracted your archives
automake directory into the automake source tree, done
autoconf
autoheader
./configure --prefix=/usr
make
make check

(this was to get your hacked automake installed rather than fiddle round
with calling a non-installed one..

and had the following tests fail:
XFAIL: cond3.test
FAIL: pr19.test
FAIL: pr87.test
FAIL: subdirbuiltsources.test
XFAIL: yaccvpath.test

I don't know if that's expected on cygwin, with or without your patch,
but I figured you'd like to know.

I'm running a fairly standard cygwin install, a few mount points (all
binary at the moment) here and there, and CYGWIN=ntsec. Oh, no
networking, all file paths are local.

Let me know if you want a cygcheck etc.

I'm going to have a quick look at why these fail, and then onto libtool.

Rob

----- Original Message -----
From: "edward" <tailbert@yahoo.com>
To: <cygwin-apps@cygwin.com>
Cc: <libtool@gnu.org>
Sent: Friday, March 09, 2001 6:30 PM
Subject: ok, new libtool for cygwin updates


> well peeps.
>
> i actually browsed through the libtool mail archives and read the note
about
> cygwin specific things (especially the mail/cygwin32 file).
>
> here is a set of updates to libtool.m4, ltmain.in (and automake.in)
that
> does just about all of it, as far as I can tell. the libtool check
suite
> passes completely (don't forget to use the hacked automake).
>
> libtool highlights:
>
> * use libFOO.dll.a for import libs, libFOO.a for static libs,
> cygFOO-version.dll for dlls
> * install cygFOO-version.dll into lib/../bin/cygFOO-version.dll
> * actually use .lai files! sets dlname to ../bin/cygFOO-version.dll
>
> note that the key thing i tested for was the creation of dll's using a
user
> generated def file, although the libtool test suite *does* pass. note
that
> handling dependencies modules is still not robust. it works if the
module is
> already in memory, otherwise no. still need to modify libltdl to
handle
> cygwin cases specifically. bleh. so if the libtool test suite fails on
mdemo
> on the execute from installed test, there you go (just nuking the .la
file
> works just fine by the way. windows already knows about dependency
libs).
>
> as far as the automake patches go, it's mostly to allow the libtool
test
> suite to pass. i did make a change to the way .exe targets are
treated.
> instead of the automake hack of re-writing all foo_PROGRAM rules to
append
> EXEEXT, i modified it to generate an internal rule called
am_foo_PROGRAM.
> this is used for targets like clean. for the standard targets, libtool
> breaks horribly with the original hack due to the generation of script
> wrappers for apps that use shared libraries, if the EXEEXT hack is
kept in.
> this isn't the best thing i can think of, but it should do for now.
>
> automake highlights:
>
> * generate internal macros am_foo_PROGRAMS (e.g. am_bin_PROGRAMS)
which hold
> .EXEEXT'd versions of foo_PROGRAMS. this is used only in the clean
target at
> the moment.
> * you also get my partially specified conditional target generation
(see
> automake mail list for details)
>
> instead of posting patches, i am posting all of libtool/libtool.m4,
> libtool/ltmain.in and automake/automake.in, because you may already
have
> patched versions of these laying around. this should allow you to drop
those
> into your testing environment and see if it works.
>
> again, these are against the latest CVS versions
>
> ps. i've removed the previous set of test libtool stuff from my
homepage.
>
> pps. don't forget to regenerate the Makefile.in files in the libtool
test
> suites (demo, depdemo, mdemo, etc.) using the hacked automake.
otherwise
> mdemo-*.tests will fail.
>
> ppps. some final notes. from what i can tell, the usage
of -no-undefined is
> mandatory on platforms like aix and windows when generating dlls. this
is
> why the mdemo test on foo1.la builds a static archive. but in simple
cases
> like foo1.la, dlopen seems to work. cheers.
>
> cheers,
> edward
>