Update PPL to v1.0

[vbraun]

spkg: ​http://boxen.math.washington.edu/home/jdemeyer/spkg/ppl-1.0.p0.spkg

Apply trac_14232_ppl_doctest_fixes.patch​

Apply 14232_bin_hgignore.patch​ to local/bin

Apply 14232_ppl_deps.patch​ to $SAGE_ROOT

[jdemeyer]

How about

sage: C_Polyhedron(1, 'empty').max_space_dimension()
357913940            # 32-bit
1152921504606846974  # 64-bit

I believe that's the standard way.

[vbraun]

Done!

[vbraun]

Updated patch

[vbraun]

Fixed the # long doctests, too

[vbraun]

Straightforward version bump, still needs review

[dimpase]

looks good.

[jdemeyer]

This exposes #11391 on Ubuntu 13.04.

[jdemeyer]

Working on a solution, hang on...

[jdemeyer]

Added fix for #11391 and #12672, needs review.

spkg diff: ppl-1.0.p0.diff​

[vbraun]

Looks good to me.

I've updated the compilerwrapper to revert LD_LIBRARY_PATH to SAGE_ORIG_LD_LIBRARY_PATH, see #10572. This would be a generic way to avoid similar headache in the future, and slowly work towards getting rid of the whole LD_LIBRARY_PATH hack.

[vbraun]

looks good to me

[jdemeyer]

Why do we configure PPL with --disable-fpmath? This was already present in the initial PPL spkg:

 * configure --disable-fpmath works now (since we don't use the
  floating point math stuff).

[jdemeyer]

In order to avoid race conditions which happen when GCC is ran while installing PPL, we need 14232_ppl_deps.patch​. This needs review.

I also made a further small change to the spkg to support the PPL_CONFIGURE environment variable, this also needs review.

[vbraun]

The floating-point stuff in PPL needs a funky rounding mode set, so its not going to work as expected in Sage. Which is why I disabled it in the configure.

This ticket is just a textbook case of the perils of abusing LD_LIBRARY_PATH.

[jdemeyer]

Added GLPK to PPL dependencies since PPL seems to use GLPK.

14232_ppl_deps.patch​ needs review.

[jdemeyer]

It seems that there was a patch on #10039 to add the GLPK dependency, but by mistake, this patch wasn't merged.

[dimpase]

Replying to jdemeyer:

It seems that there was a patch on #10039 to add the GLPK dependency, but by mistake, this patch wasn't merged.

indeed. See ​https://www.cs.unipr.it/mantis/view.php?id=500 which seems to say that the dependency on GLPK will eventually be gone, but not yet...

[leif]

Meanwhile at PPL 1.1, we have:

-rw-r--r-- 1 leif leif 57351174 2015-04-29 16:21 local/lib/libppl.a
-rw-r--r-- 1 leif leif 83958276 2015-04-29 16:21 local/lib/libppl_c.a
-rwxr-xr-x 1 leif leif     1236 2015-05-02 18:29 local/lib/libppl_c.la
lrwxrwxrwx 1 leif leif       17 2015-04-29 16:21 local/lib/libppl_c.so -> libppl_c.so.4.0.0
lrwxrwxrwx 1 leif leif       17 2015-04-29 16:21 local/lib/libppl_c.so.4 -> libppl_c.so.4.0.0
-rwxr-xr-x 1 leif leif 29967312 2015-04-29 16:21 local/lib/libppl_c.so.4.0.0
-rwxr-xr-x 1 leif leif     1156 2015-05-02 18:29 local/lib/libppl.la
lrwxrwxrwx 1 leif leif       16 2015-04-29 16:21 local/lib/libppl.so -> libppl.so.13.0.0
lrwxrwxrwx 1 leif leif       16 2015-04-29 16:21 local/lib/libppl.so.13 -> libppl.so.13.0.0
-rwxr-xr-x 1 leif leif 19604376 2015-04-29 16:21 local/lib/libppl.so.13.0.0

I.e., while the library version of libppl got bumped, libppl_c is still at 4 (the same version as in PPL 0.x!).

So we still have to build PPL "serially" (not in parallel with other packages) to avoid potential race conditions with old system GCCs linked against that version; GCC 4.8 and any later version use ISL instead of PPL. (And PPL was optionally used since GCC 4.4 IIRC, so the "critical" versions are 4.4.x to 4.7.x.)

[leif]

(Once we require full C++11 support / GCC >= 4.8.1, we can change that.)

[vbraun]

The real issue is a) abuse of LD_LIBRARY_PATH and b) necroposting on long-closed tickets ;-)

[leif]

Replying to vbraun:

necroposting on long-closed tickets ;-)

ROFL. I was just wondering whether it's still an issue (i.e., whether we could remove PPL from the sequential toolchain build) -- it is, so I didn't open a ticket for changing that; and as of Sage 6.7.beta3, build/Makefile still refers to this ticket:

# TOOLCHAIN consists of dependencies determined by build/install,
# including for example the GCC package.
toolchain: $(TOOLCHAIN)

# Build all packages that GCC links against serially, otherwise this
# leads to race conditions where some library which is used by GCC gets
# reinstalled. Since system GCCs might use Sage's libraries, we do this
# unconditionally. We still use the dependency checking from $(MAKE),
# so this will not trigger useless rebuilds.
# See #14168 and #14232.
toolchain-deps:
        $(MAKE) $(INST)/$(ZLIB)
        $(MAKE) $(INST)/$(SAGE_MP_LIBRARY)
        $(MAKE) $(INST)/$(MPFR)
        $(MAKE) $(INST)/$(MPC)
        $(MAKE) $(INST)/$(PPL)

You can open a new ticket for letting build/Makefile refer to some further new ticket of course.