[Gsll-devel] Status of status.text

[Liam Healy]

Yeah status.text is way out of date, at the very least there are 4000+
tests now, not ~1200.

Most of the recent changes have been to consolidate foreign array
handling into GSD, based on static-vectors where available.  On
the surface there isn't much change, but internally there are substantial
changes as you've noticed.  The major benefit from a user's perspective
is that it should be a lot easier to use grids in different libraries, say
computing an array in GSL and passing it to another foreign library.

As for the two errors you are seeing, these are exactly the errors
I see.  The basic problem with the test suite is that most of it was
made up by me without proper epsilons.  After I discovered that GSL
had a complete suite of tests, I started porting those tests, and they
include an epsilon for all float results.  Therefore, if one of those tests
fails then it ought to fail in GSL too.

So the sensible thing is to eliminate my made-up tests and replace them
with the GSL tests.  That's what really needs to be done with those
failing tests.  That's the long-range plan, but it's very tedious work, so
I will put a burst of effort in and then get fed up with it.  That
happened in early June, and is the reason the number of tests grew by
~2000; my meager random distribution tests were failing so I did
a wholesale port of the GSL tests.

Liam

On Tue, Jul 20, 2010 at 8:14 PM, Jason Nielsen <jdn at math.carleton.ca> wrote:
> Dear Liam,
>
> I just pulled the latest version of gsll.  Apparently you have busy as
> there are quite a few changes.  That is great, keep up the good work!
>
> With all the recent e-mail on the test suite I'm wondering how up to date
> is the status.text file?  When I run the test suite on an x86-64 box
> running Ubuntu 10.04 with SBCL 1.0.39 and a custom built gsl library
> (mostly some CFLAGS tweaks) I get the following two failures:
>
> EXPONENTIAL: (MULTIPLE-VALUE-LIST
>               (LET ((RNG (MAKE-RANDOM-NUMBER-GENERATOR +MT19937+ 0)))
>                 (LOOP FOR I FROM 0 TO 10
>                       COLLECT (SAMPLE RNG :EXPONENTIAL :MU 10.0d0))))
> failed:
> Expected ((0.0025828444588394794d0 18.145581427987647d0
>            12.636598054339759d0 0.5424387252062355d0 14.624994234158105d0
>            7.236607929535993d0 0.4345362449683603d0 2.95303920904529d0
>            6.161052939065796d0 3.011686333539114d0 2.7451079819355364d0))
> but saw ((82.61578216370394d0 1.77823538531874d0 3.3214653339021365d0
>           29.412645812312775d0 2.6351843583734817d0 6.635371453335953d0
>           31.57709150859036d0 13.637712930094189d0 7.766237034468241d0
>           13.468928238240315d0 14.268822088229632d0))
> EXPONENTIAL: 11 assertions passed, 1 failed.
>
> and:
>
> LU: (MULTIPLE-VALUE-LIST (TEST-LU-SOLVE-DIM (CREATE-COMPLEX-MATRIX 7)))
> failed:
> Expected (#2m(#C(24.0717272023734d0 -9.84612797621247d0)
>               #C(-269.338853034031d0 87.5455232472528d0)
>               #C(2966.61356736296d0 -1026.24473923993d0)
>               #C(-18207.3812124749d0 5673.8447304241d0)
>               #C(55769.3879019068d0 -16154.0963210502d0)
>               #C(-78894.1207561151d0 19505.3812987858d0)
>               #C(39554.8551241728d0 -7765.93696255317d0)))
> but saw (#2m(#C(-81.70598196064235d0 -81.70599595009693d0)
>              #C(3354.209927435102d0 3354.278264591346d0)
>              #C(-31954.85554225817d0 -31956.353685805567d0)
>              #C(119027.61090077131d0 119037.36678048137d0)
>              #C(-203627.71695137964d0 -203653.92708809028d0)
>              #C(160626.5466779804d0 160657.2640539496d0)
>              #C(-47344.00882810219d0 -47357.00253607087d0)))
> LU: 11 assertions passed, 1 failed.
>
> TOTAL: 4014 assertions passed, 2 failed, 0 execution errors.
>
> a few quick tests of the LU decomp routine seem to suggest that it is
> working fine so I'm wondering if this is the kind of "numeric fuzz" you
> were alluding to in a previous post.
>
> With regards,
> Jason
>
>
> _______________________________________________
> Gsll-devel mailing list
> Gsll-devel at common-lisp.net
> http://common-lisp.net/cgi-bin/mailman/listinfo/gsll-devel
>