Update GLPK to 4.65

[thansen]

GLPK 4.65 was released recently and it prints the warning "Long-step dual simplex will be used" all over the place, leading to many failing sagemath doctests in Debian (they fixed it by backporting an upstream patch)

We backport the corresponding patch to GLPK 4.65. Having this in will have the tack of #28459 easier.

tarball: ​http://users.ox.ac.uk/~coml0531/sage/glpk-4.65.tar.bz2

[jdemeyer]

Replying to thansen:

or if glpk just shouldn't print these warnings.

Reminds me of #20876.

I have no idea what the message "Long-step dual simplex will be used" means. Is it supposed to be a warning?

[thansen]

It's this change in src/draft/glpios03.c:

+ #if 0 /* 20/I-2018 */
           xprintf("WARNING: LONG-STEP DUAL SIMPLEX WILL BE USED\n");
+ #else
+          xprintf("Long-step dual simplex will be used\n");
+ #endif

So it seems that just the text of the message was changed. Does the content of the message influence whether it's printed?

[jdemeyer]

No idea...

[thansen]

Upstream (Andrew Makhorin) says this:

GLP_MSG_OFF is not a relevant option to disable terminal/stdout output. Please use glp_term_out(GLP_OFF) and glp_term_out(GLP_ON) instead--these calls disable/enable terminal output on a lower level independently on options passed to the solvers.

See ​https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=891465#21

[jdemeyer]

Replying to thansen:

GLP_MSG_OFF is not a relevant option to disable terminal/stdout output.

It's not our goal to completely disable terminal output. We still want to see error messages for example.

[jdemeyer]

It would also be good to know what the message "Long-step dual simplex will be used" means. Is it supposed to be a warning to say that we are doing something wrong?

[jdemeyer]

This suggestion from ​https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=891465#20 is obviously the right thing to do:

It would be very helpful if GLP_MSG_OFF suppressed this message as it does with other messages.

That would be essentially a one-line patch.

[thansen]

Should be fixed in the next release. The patch is here:

​https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=891465#24

[dimpase]

New commits:

​36e1ac5

update glpk to 4.65

[git]

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

​8f1bda0

update glpk to 4.65

[dimpase]

made the version 4.65.p0, not merely 4.65

[git]

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

​655aad0

make tests less sensitive to output formatting

[dimpase]

w.r.t. the error_recovery patch, ​http://trac.sagemath.org/ticket/20710#comment:18 - one can ask upstream, instead of uncondionally applying our patch, make it conditional, on some new parameter that may be manipulated at runtime, how about this?

[jdemeyer]

It's not that applying the patch upstream will break anything, they refused the patch on principle: they simply don't want to officially support the way how Sage is absusing their error handler. I highly doubt that making the patch conditional will convince upstream. We could ask again to apply the same patch, just in case that they changed their mind.

[dimpase]

It's one thing to ask to unconditionlly change the default mode of operation, and quite another to ask to accept a patch for a possibility to manipulate the default mode.

Asking to apply the same patch again would be not welcome, I think, and might lessen our chances to get what I propose in.

Could you ask a question whether they'd welcome such a patch, before we spend time working on it?

[fbissey]

Your strategy to remove blank lines doesn't work. At least with python2.7.

[dimpase]

I only tested with python3. Surely it's a quirk that can be fixed.

[dimpase]

Replying to dimpase:

I only tested with python3. Surely it's a quirk that can be fixed.

I just tested this branch with python 2.7, all good, too. I have no idea what's not working for you.

[fbissey]

Indeed I had made a mistake! It is all good.

[chapoton]

moving milestone to 9.0 (after release of 8.9)

[slelievre]

This breaks building SageMath on Cygwin64 on Windows 7:

• ​https://groups.google.com/d/msg/sage-release/oDwzJ8wzbGE/wSLXPQwpCAAJ

[jhpalmieri]

I'm seeing the message "Long-step dual simplex will be used" using a system installation of glpk, thereby causing lots of failed tests. Is there anything to be done about this? Refuse to use the system installation if it's version 4.65? Block the message somehow?

[slelievre]

For GLPK 4.65, some package distributions ship a patched version that removes these "Long-step dual simplex will be used" messages, but some don't. Ideally when deciding whether to install GLPK or use the system one, our install scripts should take that into account.

[jhpalmieri]

Is there a good way to detect that? Is there a standard patch-level or some other indication in the version number that this change has been made?

[thansen]

In Debian we patched the doctest parser to ignore these warnings:

​https://salsa.debian.org/science-team/sagemath/-/blob/master/debian/patches/dt-version-glpk-4.65-ignore-warnings.patch

[gh-timokau]

Replying to jhpalmieri:

Is there a good way to detect that? Is there a standard patch-level or some other indication in the version number that this change has been made?

Couldn't we just run something that would print the warnings and check if the warnings are printed?

[jhpalmieri]

Replying to thansen:

In Debian we patched the doctest parser to ignore these warnings:

​https://salsa.debian.org/science-team/sagemath/-/blob/master/debian/patches/dt-version-glpk-4.65-ignore-warnings.patch

This looks good. Why don't we add this to Sage?

[dimpase]

Replying to jhpalmieri:

Replying to thansen:

In Debian we patched the doctest parser to ignore these warnings:

​https://salsa.debian.org/science-team/sagemath/-/blob/master/debian/patches/dt-version-glpk-4.65-ignore-warnings.patch

This looks good. Why don't we add this to Sage?

+1

[jhpalmieri]

Replying to dimpase:

Replying to jhpalmieri:

Replying to thansen:

In Debian we patched the doctest parser to ignore these warnings:

​https://salsa.debian.org/science-team/sagemath/-/blob/master/debian/patches/dt-version-glpk-4.65-ignore-warnings.patch

This looks good. Why don't we add this to Sage?

+1

See #29317.