Fix gp2c spkg-configure to let configure succeed on Fedora 32

[dcoudert]

On some systems, e.g. Fedora 32, /bin is a symlink to /usr/bin
and the PATH has the symlink /bin in front of /usr/bin.

In this situation, with the version of build/pkgs/gp2c/spkg-configure.m4
from #30231 (merged in Sage 9.2.beta8), running ./configure fails
while checking for gp2c.

This ticket amends build/pkgs/gp2c/spkg-configure.m4 to solve
that problem.

The problem occurs even without trying to build gp2c, as currently
configure checks for system packages including of disabled optional
packages (see #29498 for fixing that).

CC: @dimpase @slel @orlitzky

Component: packages: optional

Keywords: gp2c

Author: David Coudert, Michael Orlitzky

Branch/Commit: fa3c4fc

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/31051

[dcoudert]

New commits:

524f4e6

change AC_MSG_ERROR to AC_MSG_WARN in spkg-configure.m4 of gp2c

[dcoudert]

Branch: public/misc/31051_gp2c

[dcoudert]

Commit: 524f4e6

[slel]

Changed keywords from none to gp2c

[dimpase]

Reviewer: Dima Pasechnik

[dimpase]

comment:3

OK

[dimpase]

comment:5

Do I get it right that the compilation goes through with the change on Fedora 32, right?

[dcoudert]

comment:6

Yes, I was able to compile with this change (plus the use of upstream url for libsemigroups).

[mkoeppe]

comment:7

I don't think this is the correct fix. When gp2c is enabled to be installed, then this should be a hard error

[orlitzky]

comment:8

Replying to @mkoeppe:

I don't think this is the correct fix. When gp2c is enabled to be installed, then this should be a hard error

True, and building gp2c should also fail since we don't know where pari.cfg is. Like the comment says, "If we can't find pari.cfg, gp2c isn't going to work."

What was the original ./configure command, and what did it output? What does it output after the message was changed to a warning? Something is fishy.

[dcoudert]

comment:9

I tried to upgrade from 9.3.beta3. After git pull on develop branch, I tried make build and it fails during configuration. the config.log can be found in the message https://groups.google.com/g/sage-release/c/dmqD6CIRUVo/m/9oCozIMkAwAJ.
and in this message https://groups.google.com/g/sage-release/c/dmqD6CIRUVo/m/CM7JJqJNAwAJ you can find the config.log with this modification.

I agree that this might be a symptom of another issue.

[orlitzky]

comment:10

Ok, there are two things weird:

1. Why is it trying to install the (optional) gp2c package when you haven't enabled it with --enable-gp2c?
2. Why can't it find your system pari.cfg?

The real problem is the first item... but, I'm curious. Your config log shows...

configure:17361: checking for gp
configure:17379: found /bin/gp

Is that really where gp lives, or is it a symlink to /usr/bin/gp? Perhaps /bin is a symlink to /usr/bin? We use that path when guessing the location of pari.cfg, so having it return /bin/gp instead of /usr/bin/gp would throw a wrench into things. It's an easy fix however.

(But regardless, you haven't enabled gp2c at all, so you shouldn't be hitting this problem.)

[dcoudert]

comment:11

You get it right, /bin is a symlink

gelati:/home/dcoudert>ls -ll /bin 
lrwxrwxrwx. 1 root root 7 Jan 28  2020 /bin -> usr/bin
gelati:/home/dcoudert> ls -l /usr/bin/gp*  
lrwxrwxrwx  1 root root       7 Nov 12 23:19 /usr/bin/gp -> gp-2.11
-rwxr-xr-x  1 root root  132936 Nov 12 23:19 /usr/bin/gp-2.11
...

I haven't enabled gp2c. May be it's a dependency of another package ? I haven't check that. I have (ok, it's certainly not the best way to get the list):

gelati:/home/dcoudert/sage> ./sage -t src/sage/graphs/print_graphs.py 
too few successful tests, not using stored timings
Running doctests with ID 2020-12-16-01-44-24-894c14a1.
Git branch: develop
Using --optional=benzene,bliss,buckygen,build,cryptominisat,csdp,dochtml,dot2tex,fedora,gap_packages,glucose,libsemigroups,mcqd,memlimit,pip,plantri,python_igraph,rubiks,sage,sage_numerical_backends_cplex,sage_spkg,tdlib,texttable
Doctesting 1 file.
sage -t --random-seed=0 src/sage/graphs/print_graphs.py
    [12 tests, 0.02 s]
----------------------------------------------------------------------
All tests passed!
----------------------------------------------------------------------
Total time for all tests: 0.0 seconds
    cpu time: 0.0 seconds
    cumulative wall time: 0.0 seconds

[orlitzky]

Changed branch from public/misc/31051_gp2c to u/mjo/ticket/31051

[orlitzky]

comment:12

This should fix the second problem, and allow --enable-gp2c to work on systems where /bin is a symlink to /usr/bin. I still have no idea about the larger issue, though.

---

New commits:

fa3c4fc

Trac #31051: support /bin -> /usr/bin symlink in gp2c spkg-configure.

[orlitzky]

Changed commit from 524f4e6 to fa3c4fc

[mkoeppe]

comment:13

Replying to @dcoudert:

I haven't enabled gp2c. May be it's a dependency of another package ?

For all packages, including non-enabled optional packages, the presence of system packages is checked. Wishlist tickets #29372, #29498 propose to change this.

[orlitzky]

comment:14

Replying to @mkoeppe:

For all packages, including non-enabled optional packages, the presence of system packages is checked. Wishlist tickets #29372, #29498 propose to change this.

This failure is within an AS_IF([test "x$sage_spkg_install_gp2c" = "xyes"]... block. Should that be passing when gp2c is not enabled?

[dcoudert]

comment:16

It's working for me on that desktop. Thanks!

I let you decide if we can set back to positive review and how to change authors/reviewers.

[dimpase]

comment:17

Looks good to me. Please add the names :-)

Let me also point out that the fact that my fix worked too is an indication that there could be a better way to discover pari.cfg than the one currently implemented in spkg-configure of gp2c.

[orlitzky]

Changed author from David Coudert to David Coudert, Michael Orlitzky

[orlitzky]

comment:18

I would still very much prefer that we not do these checks at all when gp2c is disabled, but if everyone's content to do that in some other ticket, this is still an improvement.

Ultimately, pari itself should tell us where pari.cfg is installed, via either pkg-config or something like gp --pari-cfg. Nobody thinks enumerating all possible locations is a good way to do it, but the build would eventually fail if gp2c was enabled and if we allowed the check to proceed with a warning. The previous way of running find on all of /usr and then waiting five minutes was even worse.

[dimpase]

comment:19

pari.cfg file can be found in the PARI object directory and is installed in $prefix/lib/pari/.

this is in gp2c docs of the Par project:
https://pari.math.u-bordeaux.fr/pub/pari/manuals/gp2c/gp2c.html

Fedora is wrong about the location it installs pari.cfg to

[orlitzky]

comment:20

And those docs are wrong =)

pari.cfg is installed to $(sysdatadir), which defaults to (but is not guaranteed to be) $prefix/lib/pari

[vbraun]

Changed branch from u/mjo/ticket/31051 to fa3c4fc