global cputime (inclusive some subprocesses like Singular)

[malb]

I always found this annoying:

sage: t = cputime()
sage: P = PolynomialRing(QQ,8,'x')
sage: I = sage.rings.ideal.Katsura(P)
sage: I.groebner_basis()
sage: print cputime(t)
0.217967

so here's my proposal for a fix:

sage: t = cputime_gobal()
sage: P = PolynomialRing(QQ,8,'x')
sage: I = sage.rings.ideal.Katsura(P)
sage: I.groebner_basis()
sage: print cputime_global(t)
5.647973

I am not sure if the design is particularly nice though.

Component: misc

Keywords: global cputime subprocesses

Author: Martin Albrecht

Reviewer: Simon King, Nick Alexander

Merged: sage-4.1.alpha0

Issue created by migration from https://trac.sagemath.org/ticket/4761

[williamstein]

comment:2

This needs to be rewritten since it makes the faulty assumption that there is at most one interface to singular/magma, etc., at a given moment, and also ignores many of the interfaces needlessly (?). To emphasize the first point, if I do:

sage: s = Singular()
sage: t = Singular()
sage: s('2+2')
4
sage: t('2+2')
4

then none of the computations above will be seen by cputime_global().

Regarding the second point, here is a one liner that allows you to get a list of all interface objects:

sage: print [s() for s in sage.interfaces.quit.expect_objects if s()]
[Mathematica, Singular, Axiom, Gap, Genus2reduction, GP/PARI interpreter, Kash, Lisp Interpreter, Magma, Macaulay2, Maple, Maxima, Matlab, Mupad, Mwrank, Octave, Sage, LiE Interpreter, R Interpreter, Maxima, Maxima, R Interpreter]

If you start more, they get registered in the sage.interfaces.quit.expect_objects list.

For example, this illustrates getting all currently running interface objects:

sage: gp('2+2')
4
sage: s = Singular()
sage: s('2+3')
5
sage: magma('5')
5
sage: [s() for s in sage.interfaces.quit.expect_objects if s() and s().is_running()]
[GP/PARI interpreter, Magma, Singular]

You could do this before and after the computation, and for newly started interface objects, the cputime for that one would be the total cputime (i.e., it defaults to 0).
You could also just get all interface objects, even ones that aren't yet running, but that is NOT as useful, imho, since calling cputime() on them starts them up (which is terrible)...

Anyway, hopefully the above has enough "interface fu" that you can easily write the ultimate version of cputime. And maybe it should just replace cputime? I.e.,

sage: cputime()   # what you write
sage: cputime(subprocesses=False)  # the original cputime?

It probably depends on how much overhead there is... I could easily imagine rewriting the cputime methods for each interface so that they are nearly instant unless the interface has actually done something -- i.e., put a cached value in the cputime() method for each interface that is cleared precisely when you send code to be evaluated.

[malb]

comment:4

Cool. I admit I should have searched a bit longer if there is a list of running interfaces already. Most cputime() implementations throw NotImplementedError but I assume we can just ignore those cpu times or fix that.

[malb]

comment:5

I updated the patch based on was' comments. Also, the new patch doesn't introduce a new function to the global namespace.

[williamstein]

comment:6

REFEREE REPORT:

• typo -- "be started an terminated at any given time.": The "an" should be "and".

• typo -- "somehwat random": should be "somewhat random".

• Design issues. I think if t is output from cputime(subprocesses=True), then cputime(t) should automatically have subprocesses=True, since with it False, isn't it just giving nonsense? Or is it really giving the Sage part? If the later, ignore this comment, but please document this in the function. (Yes, it's the latter.)

• Maybe a sentence that explained that so far at least all we're timing is the pexpect interface subprocesses that have a working cputime method. That any other subprocess (e.g., gfan) isn't timed. Then people know exactly what they would have to do to make sure there expect interface gets included.

• The __float__ method for GlobalCputime doesn't have the examples in an EXAMPLES block and they aren't indented properly.