On Mar 7, 2010, at 3:20 AM, Florent Hivert wrote:
Hi,
A 30-second skim through the list gives me the impression that
there are
probably 3 or 4 issues total that are causing all of these
failures. Of
course I could be wrong, and who knows how hard it will be to fix
those
homology ones (= chomp didn't build correctly?). This one jumped
out at me
though:
I don't understand why it jumped to you.
File
"/export/home/drkirkby/32/sage-4.3.4.alpha0/devel/sage/sage/
categories/finite_semigroups.py",
line 232:
sage: sorted(S.j_transversal_of_idempotents())
Expected:
['a', 'ab', 'ac', 'acb', 'b', 'bc', 'c']
Got:
['a', 'ab', 'ac', 'acb', 'b', 'c', 'cb']
How could it get that wrong?
What to you mean by that ? As indicated by the name, the function
compute a
transversal of an equivalence relation (the j-relation) so the
choice is not
canonical at all. It actually depends on the implementation of Cayley
graphs. Please see #8445 which should fix this issue.
I was reading that last line as [..., 'c', 'bc']. Makes more sense now.
The only long term solution I see is a continuous build bot that
runs on a
Solaris box (among others) and bounces tickets that fail. I plan on
putting
one together if no one beats me to it, but not until this summer.
+10 to that but not only for Solaris.
Yep, that's my intent.
I mean, neither authors of patches nor
reviewers have the time/will/possibility to check on all the possible
architectures. This should be automated.
Yep. Also, it shifts the review focus away from making sure the patch
applies and passes tests to actually refereeing the code,
documentation, and examples themselves.
My dream solution (and I can tell you
I'm not the only one dreaming about that) is the following:
When you put a ticket on trac, you must clearly indicate in a field
(not in
the text) which patches should be applied in which order. A sensible
default
may be helpful here. You must also indicate on which other tickets
this patch
depend. Then in a fully automated manner a few time after, either:
- a green light turn on indicating that all tests passed on all the
architecture.
- a red light turn on with a link to the log of the failure.
Note that I've no idea how hard it is to implement in trac,
I think it's totally doable, though it might be worth switching to
something like Reitveld. As part of my dream solution a patched copy
of Sage would be left lying around for any potential referees to try
out (and in such a way that they could easily add their own examples
to the doctests).
neither if we have the necessary hardware to support this load.
Collectively we probably do--or at least the architectures that are in
common use will be more completely tested.
- Robert
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
