Build PCRE without JIT if needed

[dimpase]

The JIT feature of PCRE is buggy on some systems: it already required a fix for Cygwin and it gives a "bus error" on Solaris.

To work around this, I propose to always run the PCRE testsuite. This takes only a few seconds anyway. If the testsuite fails, recompile PCRE without --enable-jit.

(Skip this, at the very least, on Cygwin since the testsuite fails there anyways, albeit for unrelated reasons; see #24756).

CC: @mkoeppe @dimpase @EmmanuelCharpentier

Component: packages: standard

Author: Jeroen Demeyer

Branch: db16bc9

Reviewer: Dima Pasechnik

Issue created by migration from https://trac.sagemath.org/ticket/24628

[jdemeyer]

Author: Jeroen Demeyer

[dimpase]

comment:3

I believe PCRE is used in Polymake, and in Sage's polymake interface. I guess without JIT they (as well as R, as R is using it too) might get slower...

[jdemeyer]

Branch: u/jdemeyer/build_pcre_without_jit_if_needed

[jdemeyer]

Commit: db16bc9

[jdemeyer]

comment:5

Replying to @dimpase:

I believe PCRE is used in Polymake, and in Sage's polymake interface. I guess without JIT they (as well as R, as R is using it too) might get slower...

Slower but functional should be acceptable. Note that my patch still uses JIT on systems where it is supported.

---

New commits:

db16bc9

Build PCRE without JIT if it does not work

[jdemeyer]

comment:7

real    9m46.453s
user    7m32.365s
sys     2m2.237s
Copying package files from temporary location /datapool/jeroen/sage/local/var/tmp/sage/build/pcre-8.40.p1/inst to /datapool/jeroen/sage/local
Successfully installed pcre-8.40.p1
Deleting temporary build directory
/datapool/jeroen/sage/local/var/tmp/sage/build/pcre-8.40.p1
Finished installing pcre-8.40.p1.spkg

[dimpase]

Reviewer: Dima Pasechnik

[dimpase]

comment:8

looks good to me

[vbraun]

Changed branch from u/jdemeyer/build_pcre_without_jit_if_needed to db16bc9

[embray]

Changed commit from db16bc9 to none

[embray]

comment:10

Wish someone had CC'd me on this (I don't read every ticket that gets opened). The test suite for PCRE has always failed on Cygwin (with or without JIT) so this broke the build on Cygwin.

[dimpase]

comment:11

IMHO on Cygwin, and, in fact on any system that provides PCRE,
we should just use the system's one.

[embray]

comment:12

Replying to @dimpase:

IMHO on Cygwin, and, in fact on any system that provides PCRE,
we should just use the system's one.

Agreed. Working on that :)

[EmmanuelCharpentier]

comment:13

Replying to @embray:

Replying to @dimpase:

IMHO on Cygwin, and, in fact on any system that provides PCRE,
we should just use the system's one.

Agreed. Working on that :)

This is a general need : unless we have specific needs not solvable in an interface, we should use whatever is installed systemwide.

A mechanism to do that in the toplevel configuration file would be useful. IIRC, Erik has undertaken to refactor parts of this cofiguraion system (that's why I haven't yet progresed much on OpenSSL, BTW : I'm waiting for the dust to settle a bit...), this might be one of his objectives.

However, I can think of difficulties to extende such a mechanism to library interfaces...

[embray]

comment:14

I already have a prototype for this that I'm pretty happy with, but I've been waiting, as you say, for the dust to settle on #21524 before moving forward on it. That ticket is ready for review BTW.