Fix timeout in TransitiveGroups TestSuite

[vdelecroix]

The test suite of src/sage/groups/perm_gps/permgroup_named.py takes an infinite amount of time.

The responsible doctest is

(line 1924):    TestSuite(TransitiveGroups()).run()

And it is blocked because of

sage: T = TransitiveGroups()
sage: elt = T.an_element()
sage: T(elt)     # hangs

This comes from #22382.

[SimonKing]

T.an_element() works. What doesn't work is creating an element of T from an element of T.

The traceback after <Ctrl-c> pointed me to these lines:

/home/king/Sage/git/sage/local/lib/python2.7/site-packages/sage/sets/disjoint_union_enumerated_sets.py in _element_constructor_facade(self, el)
    575             return el
    576 
--> 577         for P in self._family:
    578             try:
    579                 return P(el)

sage: T._family Lazy family (<lambda>(i))_{i in Non negative integers} }}}

It is an infinite family - of course it hangs.

Obvious solution: First check whether the parent of the given element belongs to the family. But alas, the element of T has no reasonable parent at all. I get the following rather strange sequence of lines:

sage: t = T.an_element()
sage: t.parent() in T._family
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-14-7530d2cfb513> in <module>()
----> 1 t.parent() in T._family

/home/king/Sage/git/sage/src/sage/structure/parent.pyx in sage.structure.parent.Parent.__contains__ (/home/king/Sage/git/sage/src/build/cythonized/sage/structure/parent.c:10458)()
   1140             False
   1141         """
-> 1142         P = parent(x)
   1143         if P is self or P == self:
   1144             return True

/home/king/Sage/git/sage/local/lib/python2.7/site-packages/sage/structure/element.pxd in sage.structure.element.parent (/home/king/Sage/git/sage/src/build/cythonized/sage/structure/parent.c:28336)()
     71         return type(x)
     72     else:
---> 73         return p()
     74 
     75 

TypeError: descriptor 'parent' of 'sage.structure.sage_object.SageObject' object needs an argument
sage: t
Transitive group number 1 of degree 0
sage: t.parent()
<class 'sage.groups.perm_gps.permgroup_named.TransitiveGroup_with_category'>

No idea how to fix that, though.

[dimpase]

I don't know what exactly family means in this context, and perhaps I talk nonsense, but shouldn't iteration through such things be handled by an iterator (and yield)?

[tscrim]

So this comes from #22382, where we actually are now doing some validity checks. So when the family of sets in the union is infinite, we get the hanging behavior when an element is not in the set. We can reverse this behavior for this case, but that is a somewhat separate issue as in this case, the actual parent of the element in question is broken.

[vdelecroix]

New commits:

​c28fcf0	18220: raise an error for non exact fields
​6236b0d	22576: disable two tests

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​287dfb6

22576: disable two tests

[vdelecroix]

After disabling two tests I am able to run the doctests. However, database_gap seems to be broken (at least on my computer)

$ sage -f database_gap
$ sage -c "PrimitiveGroups(12).cardinality()"
verbose 0 (2474: permgroup_named.py, cardinality) Warning: PrimitiveGroups requires the GAP database package. Please install it with ``sage -i database_gap``.

[dimpase]

this is caused by a stale GAP workspace. Do

sage -c "gap_reset_workspace()"

[vdelecroix]

Indeed

$ sage -t --long --optional=sage,database_gap permgroup_named.py 
too few successful tests, not using stored timings
Running doctests with ID 2017-03-11-10-19-11-a1e57781.
Git branch: develop
Using --optional=database_gap,sage
Doctesting 1 file.
sage -t --long permgroup_named.py
    [445 tests, 3.90 s]
----------------------------------------------------------------------
All tests passed!
----------------------------------------------------------------------
Total time for all tests: 4.0 seconds
    cpu time: 3.2 seconds
    cumulative wall time: 3.9 seconds

[dimpase]

Don't we have # known bug tag for such broken doctests?

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​304355c

22576: disable two tests

[vdelecroix]

We do

[vdelecroix]

Thanks! I will reactivate database_gap on my patchbot as soon as it is merged.

[jdemeyer]

I think the proper way to deal with # known bug is to open a new ticket to add # known bug and keep this ticket open for the real issue.

[jdemeyer]

Besides, I'm not really following. Do I understand you correctly that the proper fix is simply to run gap_reset_workspace() whenever database_gap is installed? So it's unrelated to #22382 after all?

[vdelecroix]

Replying to jdemeyer:

I think the proper way to deal with # known bug is to open a new ticket to add # known bug and keep this ticket open for the real issue.

#22583

[vdelecroix]

Replying to jdemeyer:

Besides, I'm not really following. Do I understand you correctly that the proper fix is simply to run gap_reset_workspace() whenever database_gap is installed? So it's unrelated to #22382 after all?

You are not really understanding. This bug of this ticket is disjoint from the gap_reset_workspace() which has to do with 7.

[jdemeyer]

Replying to vdelecroix:

Replying to jdemeyer:

I think the proper way to deal with # known bug is to open a new ticket to add # known bug and keep this ticket open for the real issue.

#22583

Better do it the other way around because all the discussion is happening here.

[jdemeyer]

Replying to vdelecroix:

You are not really understanding.

Well, the comments were not clear either. It was not obvious to me that the "this" from 8 was referring not the bug on this ticket.

[dimpase]

OK, so we need to deal with sage -c "gap_reset_workspace()" issue. As Jeroen rightly says on #22570, this cannot be done by spkg-install, as it updates user-specific stuff in ~/.sage/gap/.

What else can be done? I only see an option of sage checking on startup whether sage -c "gap_reset_workspace()" must be run. Can we get away with only triggering it if a new Sage package was installed? (otherwise this looks like a time-consuming check).

Docs on how to install GAP packages manually should also indicate the need to run sage -c "gap_reset_workspace()" after a GAP package is installed or removed.

[slelievre]

Is this still a bug after database_gap has been merged partly into the gap and partly into the gap_packages packages?

[slelievre]

Replying to slelievre:

Is this still a bug after database_gap has been merged partly into the gap and partly into the gap_packages packages?

Answering my own question: yes, the following still hangs:

sage: T = TransitiveGroups()
sage: t = T.an_element()
sage: T(t)

[tscrim]

Here is a fix. The issue is that TransitiveGroups claims to be a Parent, but it is not as it doesn't have proper elements. I just hacked around that by directly implementing a __call__. It has nothing to do with the GAP package.

---

New commits:

​1b99249

Working around not being a Parent for TransitiveGroups.

[dimpase]

LGTM, thanks!