update M4RI to version 20110901 or later

[malb]

In #11574 it was noticed that M4RI does not provide sufficient information to allow third party software to choose the appropriate compiler flags to link against it. While a work around is provided in #11574, it is not as clean as it could be. That is, M4RI should export the flags that were used to when building it, which can then be re-used by third party code.

Install http://sage.math.washington.edu/home/malb/spkgs/libm4ri-20111004.spkg

Apply either attachment: trac_11757_m4ri_sse2.patch or attachment: trac_11757_m4ri_sse2_rebased10903.patch (on top of #10903).

Depends on #11574

CC: @nexttime @alexanderdreyer

Component: packages: standard

Author: Martin Albrecht

Reviewer: Alexander Dreyer

Merged: sage-4.7.2.alpha4

Issue created by migration from https://trac.sagemath.org/ticket/11757

[malb]

comment:1

Here's how the output looks like on bsd.math:

#define __M4RI_CFLAGS                   " -mmmx -msse -msse2 -msse3   -I/Users/malb/sage-4.7.2.alpha2/local/include -g -fPIC -Wall -pedantic -O2"

I'm not sure we want the "-I/Users/..." part in there though. But we set CFLAGS in spkg-install so I'm not sure how to avoid it.

[nexttime]

comment:2

Replying to @malb:

Here's how the output looks like on bsd.math:

#define __M4RI_CFLAGS                   " -mmmx -msse -msse2 -msse3   -I/Users/malb/sage-4.7.2.alpha2/local/include -g -fPIC -Wall -pedantic -O2"

I'm not sure we want the "-I/Users/..." part in there though. But we set CFLAGS in spkg-install so I'm not sure how to avoid it.

This would introduce new potential hardcoding issues. (It shouldn't hurt if the directory later -- after moving Sage -- does no longer exist, but it could break the installation if it does, and the headers there are meanwhile incompatible.)

You could post-process the file in M4RI's spkg-install.

Do you also add __M4RI_CC? (Haven't looked at the spkg yet.)

[nexttime]

comment:3

Since you have -I... already in CPPFLAGS, not adding it to CFLAGS in spkg-install should also work.

[nexttime]

comment:4

Replying to @nexttime:

Since you have -I... already in CPPFLAGS, not adding it to CFLAGS in spkg-install should also work.

Oh, I see M4RI doesn't honor CPPFLAGS, but it also doesn't use any of Sage's headers AFAIK, so you could simply remove -I... from CFLAGS (unless configure does weird things).

[malb]

comment:5

I've revised spkg-install accordingly and replaced the SPKG. Now we get:

#define __M4RI_CC                       "gcc -std=gnu99"
#define __M4RI_CFLAGS                   "    -g -fPIC -Wall -pedantic -O2"

[nexttime]

comment:6

Added #11574 as a dependency because of attachment: ticket:11574:m4ri_20110601.patch (API change, corrected extension module dependencies in module_list.py).

(Martin, correct me if I'm wrong...)

[nexttime]

Dependencies: #11574

[malb]

comment:7

Uh, yep, that's correct. Thanks!

[nexttime]

comment:8

Why does this still need review?

And isn't "exporting" the compiler flags meanwhile fully fixed by #11574? (Except for perhaps a .pc file.)

The description should get updated...

Or is this spkg / ticket meanwhile obsolete, superseded by the one from #11574?

I actually first thought I had merged the wrong spkg.

[malb]

comment:10

In #11574 we only work around this short coming of M4RI, in this ticket it's fixed. It's actually needs work because the Sage patch should be updated to use the exported flags (instead of setting its own).

[malb]

Attachment: trac_11757_m4ri_sse2.patch.gz

[malb]

comment:11

The updated SPKG & patch work fine for me on sage.math, bsd.math and cicero (where we had issues before).

[alexanderdreyer]

comment:13

I installed and tested spkg and patch successfully on a SLES 11 amd64 and Mac OS X 10.5 ppc (32bit). Also the patch and the spkg look sane, so positive review.

[alexanderdreyer]

Reviewer: Alexander Dreyer

[malb]

comment:14

yay!

[jdemeyer]

Rebased on top of #10903

[jdemeyer]

comment:15

Attachment: trac_11757_m4ri_sse2_rebased10903.patch.gz

[jdemeyer]

Merged: sage-4.7.2.alpha4