Update NTL to 6.0.0

[Volker Braun]

New upstream version.

Use git branch and tarball at:

• ​http://boxen.math.washington.edu/home/jpflori/upstream/ntl-6.0.0.tar.bz2

[Leif Leonhardy]

... and keep the $MAKE patches in mind (don't recall the ticket number right now), in case they're still necessary.

[Volker Braun]

If you mean #14692: I won't included anything from that ticket as long as it is not finished.

[Leif Leonhardy]

Replying to vbraun:

If you mean #14692: I won't included anything from that ticket as long as it is not finished.

Yes, that's what I meant.

It is finished, you just have to include the ($MAKE-related) patches.

I rebased Jean-Pierre's spkg at #2114 on mine, but then Jeroen felt he had to make trivial changes to that again, so that I would now again have to rebase mine on that of #2114, but that doesn't make sense if you're going to create a 6.0.0 anyway.

[Jean-Pierre Flori]

Yes please include #14692, there was just a trivial change to .hgignore to do and hg addremove but I completely forgot...

[Jean-Pierre Flori]

If Shoup has some interest on updating NTL again, we should definitely forward him our patches.

[Volker Braun]

Initial patch

[Volker Braun]

Please somebody go ahead and forward our patches...

[Jean-Pierre Flori]

Ok, I'll take care of forwarding the patches.

[Jean-Pierre Flori]

Groumpf, you just prevented me from posting that:

Please don't produce any spkg yet if not already done. Before forwarding the patches I have to check they still apply onto 6.0.0, so I can draft an initial spkg on top of the one from #14692 and then you can fix whatever you want on top of it to ease the git transition and so on.

So I'll just check your spkg.

[Volker Braun]

Singular doesn't build with NTL-6.0.0.

In the interest of getting this git-ready let's wait with the version bump.

[Volker Braun]

For the record, this is not fixed in Singular-3.1.6 (#14333), but is fixed in the upstream trunk ​https://groups.google.com/d/msg/libsingular-devel/aiMvEnN8qyg/cIFUsOXxK_MJ

[Jean-Pierre Flori]

Replying to jpflori:

Ok, I'll take care of forwarding the patches.

Done and taken into account by Shoup. It's on his todo list. Great news.

[Jean-Pierre Flori]

Replying to jpflori:

Replying to jpflori:

Ok, I'll take care of forwarding the patches.

Done and taken into account by Shoup. It's on his todo list. Great news.

For future reference: ​http://shoup.net/pipermail/ntl_shoup.net/2013-July/000041.html

Note that Victor's answer is not appearing there.

[Jean-Pierre Flori]

Singular update is positively reviewed. I simply gitified everything and had to add a definition in the Singular patch for NTL 6.0.0 compat. Let's get this in, I agree with Volker previous changes. The only point left is quickly check that I did not screwed up the singular patch modif; and fix all the horrible segfaults.

---

New commits:

​649e889	Update NTL to 6.0.0.
​eca3246	# Wed Jul 10 11:09:10 2013 -0400

[git]

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

​5c6cf3d	Update Singular's patch for NTL 6.0.0 compatibility.
​46c4b36	Update NTL to 6.0.0.
​e2fafd7	# Wed Jul 10 11:09:10 2013 -0400
​dd46991	Remove description of deleted patches.
​536a31f	Update Singular function interface to API changes in Singular 3.1.6.
​1753f06	Update Singular to version 3-1-6.

[Jean-Pierre Flori]

It seems we now have double frees. As a first guess, I'd say that with the previous hacks we explicitely destroyed the hacked C++ objects, and we still do, but Cython might as well now it's aware that these are C++ object, and so boom.

[git]

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

​b6d64b4

Fix double free issues with NTL 6.0.0.

[Jean-Pierre Flori]

All tests should pass now.

[François Bissey]

Just to be clear. You git branch has both #14333 and this ticket. Which now that I wrote it, seem totally normal. With git, branch depending from another ticket will include change from the depending ticket.

[Jean-Pierre Flori]

I don't really know what should happen here. For sure #14333 is a dependency. But I forgot the # in the dependencies field so maybe that's what confused the git plugin.

[François Bissey]

OK so I will try to do some testing of this. Apart from singular the following packages have a dependency to ntl:

• flint
• eclib
• linbox
• "sage-clib" and sage itself.

So my plan is to start from 6.1.beta2 add #14333, this ticket the flint upgrade and then rebuild the rest.

[Jean-Pierre Flori]

Replying to fbissey:

OK so I will try to do some testing of this. Apart from singular the following packages have a dependency to ntl:

• flint
• eclib
• linbox
• "sage-clib" and sage itself.

So my plan is to start from 6.1.beta2 add #14333, this ticket the flint upgrade and then rebuild the rest.

Any result on that? :)

[François Bissey]

I dropped the ball because of other stuff. It will be easier to start from 6.2.beta2 since singular has been merged.

[François Bissey]

Well so far it is extremely painful and not going very far. Switching branch with sage --dev checkout brought me in a state where I ended downloading older versions of sage's dependencies. I have a thing about complaining about documentation, I'll certainly try to write some stuff down if I ever figure out things.

[François Bissey]

Oh and the tarball cannot be found for good measure.

[Jean-Pierre Flori]

Typo indeed.

[François Bissey]

Replying to jpflori:

Typo indeed.

Can't believe I didn't notice that.

[François Bissey]

Can you rebase the branch on 6.2.beta2? I have done:

• ./sage -dev checkout --ticket 14876
• downloaded ntl-6.0.0.tar.bz2 to upstream
• tried to rebuild flint...

cat build/pkgs/flint/package-version.txt 
2.3.p1

So that branch is still on the old versions of flint, singular and eclib. I need to be able to rebuild them to fully test the ticket.

[git]

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

​8c9d711

Merge remote-tracking branch 'trac/develop' into ticket/14876

[François Bissey]

Do you know how I should checkout the changes? ./sage -dev checkout --ticket 14876 still leaves me with flint 2.3.p1.

[Jean-Pierre Flori]

I'd say

git fetch --all
git checkout -b test_ntl trac/u/jpflori/ticket/14876

Of course, you need to be tracking everything from the track git branch for that to work. I don't remmeber the syntac, but something like

git fetch trac u/jpflori/ticket/14876
git checkout -b test_ntl trac/u/jpflori/ticket/14876

might do the trick if you don't.

I don't know how to do that with the dev scripts, sorry :(

[François Bissey]

Forget about the dev scripts for now.

git fetch --all
git checkout u/jpflori/ticket/14876

worked. pity I didn't think of switching to a test branch earlier. So far I did ./sage -f flint,eclib,singular,linbox and ./sage -b successfully. Running ./sage -tp 12 --all --long now.

[François Bissey]

----------------------------------------------------------------------
All tests passed!
----------------------------------------------------------------------
Total time for all tests: 1234.6 seconds
    cpu time: 12669.7 seconds
    cumulative wall time: 13905.8 seconds

Yes!

[François Bissey]

I just had a thought. Will the dependency of ntl need an artificial bump to be rebuilt or is the upgrade system able to cope with that?

[Volker Braun]

Dependents of NTL will be rebuilt automatically.

I tried upgrading and it didn't produce a working Sage, but complete rebuild does. Possibly because wrong header files are used at one point, one of the limitations of our build system.

[Jean-Pierre Flori]

For those interested, 6.1 has been released: #15938 (nothing to see there yet).