Old FSF config.(guess|sub) scripts do not recognize Cygwin64

[jpflori]

The FSF config.guess and config.sub shipped with MPIR 2.6.0 and some other packages are very old (timestamp says 2003 for the MPIR ones) and fail to recognize Cygwin64. Latest upstream files at:

• ​http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.guess;hb=HEAD
• ​http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.sub;hb=HEAD

solves this issue.

MPIR report at:

• ​https://groups.google.com/d/msg/mpir-devel/kas57nK3UCU/li5-MMsEk_4J
• will be fixe by #15015

Other spkg affected:

• sqlite fixed at #16098.
• ecl (twice because of shipped but not used old GMP...) #9493.
• IML fixed here!

[jpflori]

There is a planned update to MPIR spkg at #13948.

[jpflori]

sqlite spkg (even the latest upstream packaged in 5.10.beta*) fails for the same reason.

[jpflori]

Same problem when configuring ECL, it fails in the provided gmp subdir (that we don't even use...).

[leif]

Replying to jpflori:

Same problem when configuring ECL, it fails in the provided gmp subdir (that we don't even use...).

I proposed a patch to allow removing the GMP folder a couple of times... 8/

[jpflori]

I'm changing this ticket to only take care of the outdated fsf config.* scripts.

The further problems with MPIR should go elsewher.

[leif]

Replying to leif:

Replying to jpflori:

Same problem when configuring ECL, it fails in the provided gmp subdir (that we don't even use...).

I proposed a patch to allow removing the GMP folder a couple of times... 8/

ROFL, three years ago: #9493.

(No idea whether the patch w.r.t. install-sh still applies or is at all still necessary.)

[tscrim]

Given the above, shall we put the new scripts as a patch and start the review?

[tscrim]

Here's my attempt at a clean patch.

---

New commits:

​c0a09c5	Make sure tinfo is correctly linked in when needed on Cygwin.
​45bb195	Update ncurses to dev version 5.9-20131221.
​f5a57d0	Merge branch 'u/jpflori/ticket/15617' of trac.sagemath.org:sage into cygwin64build

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​815819c

Initial attempt.

[tscrim]

Now with the correct thing pushed.

[jdemeyer]

This patch is very bad. MPIR's config.guess is a custom script, it is not the autotools' version. See also the thread you linked to.

For example, it detects my machine as sandybridge-unknown-linux-gnu. With this patch, this becomes x86_64-unknown-linux-gnu.

Also, Initial attempt. is obviously a bad commit message.

[jpflori]

Hey Travis, I'll up my working (hopefully not only for me) branch a little later today. In this very ticket I'll only update the FSF scripts. In another one I'll put the magic needed for MPIR to build the correct ABI.

[jpflori]

Maybe some other spkgs are missing, I'll add them while buillding Sage.

---

New commits:

​f09bc42	Let MPIR build without C++ support on Cygwin.
​c81e4b9	Merge remote-tracking branch 'trac/u/jpflori/ticket/15365' into ticket/14648
​0ac6c60	Update config.(guess|sub) files to recognize Cygwin64.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​f09bc42	Let MPIR build without C++ support on Cygwin.
​c81e4b9	Merge remote-tracking branch 'trac/u/jpflori/ticket/15365' into ticket/14648
​c311252	Update config.(guess|sub) files to recognize Cygwin64.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​f09bc42	Let MPIR build without C++ support on Cygwin.
​c81e4b9	Merge remote-tracking branch 'trac/u/jpflori/ticket/15365' into ticket/14648
​b922e3b	Update config.(guess|sub) files to recognize Cygwin64.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​f09bc42	Let MPIR build without C++ support on Cygwin.
​c81e4b9	Merge remote-tracking branch 'trac/u/jpflori/ticket/15365' into ticket/14648
​9ba3d7f	Update config.(guess|sub) files to recognize Cygwin64.

[jpflori]

IML needs updating.

[jdemeyer]

Since #15580, you could also consider copying config.guess from $SAGE_ROOT/config/config.guess.

[jpflori]

Yes, that's a very good idea!

[jdemeyer]

Be sure to check which version of Automake you require (config.guess belongs to Automake) and edit configure.ac accordingly.

[jpflori]

I guess there is some magic in configure which checks for consistency between configure.ac and config.guess?

[jdemeyer]

I meant edit $SAGE_ROOT/configure.ac, in particular the line

AM_INIT_AUTOMAKE([1.9.6 foreign])

And no, I don't think autotools cares about the version of config.guess, it just requires it to be available.

[jpflori]

After a quick look through automake versions it would be 1.10.

[tscrim]

FTR, this currently doesn't work for me on my cygwin32 and ubuntu machines. I believe it's the same issue as comment:14:

checking build system type... Invalid configuration `k8-pc-linux-gnu': machine `k8-pc' not recognized
configure: error: /bin/bash ./config.sub k8-pc-linux-gnu failed
Error configuring MPIR (with CFLAGS unset).

[jpflori]

Replying to tscrim:

FTR, this currently doesn't work for me on my cygwin32 and ubuntu machines. I believe it's the same issue as comment:14:

checking build system type... Invalid configuration `k8-pc-linux-gnu': machine `k8-pc' not recognized
configure: error: /bin/bash ./config.sub k8-pc-linux-gnu failed
Error configuring MPIR (with CFLAGS unset).

I assume the previous version of the pkg does work?

[jpflori]

(And you are using my branch and not yours?)

[tscrim]

Yes (and yes).

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​318da4d	Merge remote-tracking branch 'trac/develop' into ticket/14648
​79275bd	Use config.(guess|sub) from $SAGE_ROOT/config/.

[jpflori]

For the k8 issue, did any previous version work?

[jpflori]

Ok, I guess the problem is with ​https://github.com/wbhart/mpir/commit/3b410a80565eb71fd32a86630d550098ebc85f82

In fact, the MPIR wrapper should be modified to map k8 and so on to something else.

[jpflori]

Latest sqlite is wroking on Cygwin64 (or rather "at least it builds now"). See #16098.

[jpflori]

Latest MPIR at #15015 will work as well.

[jpflori]

ECL fixed at #9493.

So only dealing with IML here.

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​d3a4456

Use uptodate config.* script for IML.

[vbraun]

Thats a kind of vile hack to use some of the Sage autoconf output. Did you contact IML upstream?

[jpflori]

Note that this vile hack just use standard files distributed by autoconf, not some files specifically crafted (well, the exact version is) by configure when it's run.

Not contacted IML though, I though the project was dead. If not we should really work more with them to deal with the BLAS/linking hell in their build system.

[vbraun]

Well we could give it a try. If there is no response then we should just make our own tarball with updated autoconf. Its not exactly a moving target ;-)

I agree that the hack works, but its ugly and fragile if we change anything in the sage autoconf files (e.g. move to a different directory).

[jpflori]

Replying to vbraun:

Well we could give it a try. If there is no response then we should just make our own tarball with updated autoconf. Its not exactly a moving target ;-)

I'll drop them an email (without specific patches first, just getting the temperature there).

I agree that the hack works, but its ugly and fragile if we change anything in the sage autoconf files (e.g. move to a different directory).

Sure, I realize I should at least check the return code for such cases.

[jpflori]

Sent an email upstream.

[leif]

Replying to jpflori:

Sent an email upstream.

/dev/null?

[jpflori]

Maybe... At least the mail address given here: ​https://cs.uwaterloo.ca/~astorjoh/iml.html#bug Don't know where the authors are working now.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​4638ee1

Check that FSF config.* scripts are correctly copied from SAGE_ROOT/config.

[jpflori]

Replying to jpflori:

Maybe... At least the mail address given here: ​https://cs.uwaterloo.ca/~astorjoh/iml.html#bug Don't know where the authors are working now.

Got a reply and Arne would be very happy to make a new release. Now it's time to think about what we should push upstream with regard to the build system (it may be a good time to provide support for different BLAS implem than ATLAS), bug fixes and so on.

[leif]

Replying to jpflori:

Replying to jpflori:

Maybe... At least the mail address given here: ​https://cs.uwaterloo.ca/~astorjoh/iml.html#bug Don't know where the authors are working now.

Got a reply and Arne would be very happy to make a new release.

\o/ (after years of apparent death)

Now it's time to think about what we should push upstream with regard to the build system (it may be a good time to provide support for different BLAS implem than ATLAS), bug fixes and so on.

Best post to sage-devel, such that the usual suspectives will notice, and/or cc some of those involved in the previous IML tickets (#748 (!), #9568 ...).

[vbraun]

Replying to leif:

the previous IML tickets (#748

%-)

[fbissey]

Thanks for the head up on IML :) the laundry list as far as I remember is

• getting configure to accept other cblas. Cannot use gentoo patches as they rely on cblas installing .pc files which is not standard across implementation.
• remove the "repl" subfolder which doesn't do anything and create problems on OS X ​https://raw.githubusercontent.com/cschwan/sage-on-gentoo/master/sci-libs/iml/files/iml-1.0.3-repl_removal.patch
• fix undefined symbols in nullspace.c ​https://raw.githubusercontent.com/cschwan/sage-on-gentoo/master/sci-libs/iml/files/iml-1.0.3-fix-undefined-symbol.patch
• I have an extra fix in the s-o-g tree for automake 1.13 that probably should go in ​https://raw.githubusercontent.com/cschwan/sage-on-gentoo/master/sci-libs/iml/files/iml-1.0.3-automake-1.13.patch

[jpflori]

Better late than never, I've forwaded our patches (except the emamples one, what is that for? quicker test?) and two new ones for inclusion in a potential new release of IML (one for shared libs on Cygwin (who cares? :)) and on to add a --with-cblas configure flag to pass the ld flags for linking to cblas, falling back to the old atlas detection code in case of failure).

Best, JP

[jpflori]

All packages are now recent enough except for MPIR which will get updated in #15015.