Upgrade GCC to 4.7.2

[jdemeyer]

spkg: ​http://boxen.math.washington.edu/home/jdemeyer/spkg/gcc-4.7.2.p1.spkg (diff​)

apply 14378_gcc_version.patch​ to SAGE_ROOT.

gcc-4.7.2.p1 (Jeroen Demeyer, 4 April 2013)

• Trac #14378: Use $SAGE_SPKG_INST variable.
• Disable bootstrap-debug on all systems.
• Remove libitm which is broken on OS X 10.4 PPC.

[jpflori]

Could you include the Solaris fix to prevent debug things as we discussed on sage-???. This is the same fix as for old OS X.

[leif]

Afterwards create an optional GCC 4.8.0 spkg ... :-)

[Needs a C++ compiler to bootstrap, I know, but nvm.]

---

Haven't looked or tried whether that's still an issue, but at some point in the past the GCC spkg failed to build when encountering parts of CLooG and/or PPL. (We may add --without-* and/or --disable-lto.)

[jpflori]

I think we should also keep the 4.6.x series as optional spkg. Jeroen proposed to do it IIRC, so consider this message to be just a reminder.

[jpflori]

And I agree with the patch you posted for GCC versions checks. Just a little typo:

echo >&2 "gcc <= 4.5 fails to compile PARI/GP om ia64." 

[jdemeyer]

Replying to jpflori:

Could you include the Solaris fix to prevent debug things as we discussed on sage-???. This is the same fix as for old OS X.

Sorry, I don't know what you mean.

[jpflori]

This fix: ​https://groups.google.com/d/msg/sage-release/eNtnQ5jWuag/Yyce8bZgyUIJ

[jpflori]

FYI, I rebuilt GCC more "correctly" using Sun as and ld and the BUILD_CONFIG trick was not needed anymore, not that I'm suggesting to do as before and only use it on OS X 10.4.

[jdemeyer]

I have tested this on the buildbot and there are no problems at all.

[vbraun]

Looks good to me.

How are we keeping gcc 4.6 as optional spkg? Rename it to gcc46-4.6.3.spkg?

[jdemeyer]

Replying to vbraun:

How are we keeping gcc 4.6 as optional spkg? Rename it to gcc46-4.6.3.spkg?

Just put it with the optional packages, no need to rename it.

[vbraun]

Is there a way to disambiguate between standard and optional spkg when running sage -f <package>?

[jdemeyer]

Replying to vbraun:

Is there a way to disambiguate between standard and optional spkg when running sage -f <package>?

You can always do

./sage -i spkg/standard/gcc-4.7.2.p1.spkg

or

./sage -i http://www.sagemath.org/packages/optional/gcc-4.6.3.spkg

No need to differentiate then.

Or also

./sage -i gcc-4.6.3.spkg

[jdemeyer]

This doesn't actually build on OS X 10.4 PPC (as reported on sage-devel). But it seems that the problem lies with libitm, which is a non-critical part of GCC.

[jdemeyer]

Removed libitm, testing now on OS X 10.4 PPC.

[jpflori]

For future reference, the bug linked in the sage-devel post at ​https://groups.google.com/d/msg/sage-devel/DhJcPuFIlPo/wu3sv6nQLIgJ seems to be the upstream report: ​http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52482 (as reported in the post itself, nothing new here).

[jpflori]

Looks fine, compile on a usual Linux and works there. Also seems to be going fine on a Linux ppc, so positive review.

[jdemeyer]

Follow-up: #14441

[jhpalmieri]

I can't get this to build in iras: see ​log file. Note that it actually creates the various binaries in local/bin, so that further attempts (using, say, the 4.7.3 package at #14441 fail because "configure: error: C compiler cannot create executables". I'm trying to clear out local/bin and try 4.7.3 again. Other suggestions?

[jdemeyer]

Replying to jhpalmieri:

I can't get this to build in iras: see ​log file.

Could you please try a serial build (no MAKE -jN) and post the log file of that if it fails.

I'm trying to clear out local/bin and try 4.7.3 again.

Removing just the gcc, g++ and gfortran executables should be sufficient to force the use of the system gcc to build GCC.

[jhpalmieri]

Replying to jdemeyer:

Replying to jhpalmieri:

I can't get this to build in iras: see ​log file.

Could you please try a serial build (no MAKE -jN) and post the log file of that if it fails.

A serial build of 4.7.2 succeeded. Should the spkg force that on ia64 machines?

[leif]

Replying to jhpalmieri:

Replying to jdemeyer:

Replying to jhpalmieri:

I can't get this to build in iras: see ​log file.

Could you please try a serial build (no MAKE -jN) and post the log file of that if it fails.

A serial build of 4.7.2 succeeded. Should the spkg force that on ia64 machines?

Without having looked at the log: Are you sure this is related to the architecture (and not just caused by the usual clock skew on Skynet)?

[jhpalmieri]

Replying to leif:

Replying to jhpalmieri:

Replying to jdemeyer:

Replying to jhpalmieri:

I can't get this to build in iras: see ​log file.

Could you please try a serial build (no MAKE -jN) and post the log file of that if it fails.

A serial build of 4.7.2 succeeded. Should the spkg force that on ia64 machines?

Without having looked at the log: Are you sure this is related to the architecture (and not just caused by the usual clock skew on Skynet)?

I have no idea. How would I check?

[leif]

Replying to jhpalmieri:

Replying to leif:

Replying to jhpalmieri:

Replying to jdemeyer:

Replying to jhpalmieri:

I can't get this to build in iras: see ​log file.

Could you please try a serial build (no MAKE -jN) and post the log file of that if it fails.

A serial build of 4.7.2 succeeded. Should the spkg force that on ia64 machines?

Without having looked at the log: Are you sure this is related to the architecture (and not just caused by the usual clock skew on Skynet)?

I have no idea. How would I check?

(Move the whole Sage installation and) build the GCC spkg (in parallel) below /tmp (which on Skynet is always on local media I think)?

[jhpalmieri]

Replying to leif:

(Move the whole Sage installation and) build the GCC spkg (in parallel) below /tmp (which on Skynet is always on local media I think)?

I tried that, but it failed with "no space left on device". There are close to 40000 files and/or directories in /tmp on iras, most or all of which should be cleaned out...

[leif]

Replying to jhpalmieri:

Replying to leif:

(Move the whole Sage installation and) build the GCC spkg (in parallel) below /tmp (which on Skynet is always on local media I think)?

I tried that, but it failed with "no space left on device". There are close to 40000 files and/or directories in /tmp on iras, most or all of which should be cleaned out...

OMG, I see...

I deleted my own left-overs, but probably the buildbot, Volker and Jeroen should do so as well...

(FWIW, I reported Sage still leaving a lot of temporary crap a couple of times on sage-devel.)

[vbraun]

Oh boy, no tmpwatch %-)