>From the test reports, here are the probable source of fails: qsorti needs to be using the correct PDL_Index type. Question: is it possible to define a full PDL data type but have the typeloop code be aliased to another type (that is the same) rather than duplicating?
convolveND in imageND.pm eigens stuff in matrixops.pd (we really need to implement a more robust eigen solver---half of this doesn't quite work as is.) indexND interpND range I'm hoping Craig can sort out the range related problems. Of course, these bugs are erratic (prove -b t/matrixops.t passes but perl -Mblib t/matrixops.t fails). The result for your platform may vary. --Chris On Sun, May 12, 2013 at 11:39 PM, Chris Marshall <[email protected]>wrote: > I've been able to reproduce the test failures you see with > 32bit SPP. Will take a look when I get the chance. > > -chm > > > > > On Sun, May 12, 2013 at 11:31 PM, Chris Marshall > <[email protected]>wrote: > >> Thanks for the report, Rob. >> >> I'll apply the strtoll patch for MSVC. Not sure what the problems >> are in the t516.txt log but definitely want to resolve those issues >> before merging 64bit suppot into master. >> >> Cheers, >> Chris >> >> >> >> On Sun, May 12, 2013 at 10:36 PM, <[email protected]> wrote: >> >>> >>> From: Chris Marshall >>> >>> >>> This release makes generally available the new, 64bit >>>> index support for testing, evaluation, use, development >>>> and feedback. The only difference from the current PDL >>>> git master branch is the introduction of a PDL_Index >>>> type that can be 64bits if your perl IV type is. >>>> >>> >>> Bit of a mixed bag for me, so far. (The following pertains only to >>> builds on MS Windows.) >>> >>> Many (all?) MS compilers need the attached patch to slices.pd. >>> Even those MS compilers that don't *need* the patch (if there are any) >>> should accept that patch ok. >>> I don't know how to apply that fix to the relevant git repo version of >>> slices.pd. (Presumably we don't want to alter the slices.pd that's in the >>> main branch just yet.) >>> >>> One thing I did notice is that the fftw and minuit test scripts don't >>> check for $Config{WITH_FFTW} and $Config{WITH_MINUIT}. >>> Consequently those test scripts were being run against the installed >>> version of those modules - and failing, of course. >>> I'll fix those 2 test scripts in the main branch when I get a chance (if >>> no-one does it beforehand). >>> >>> Other than that, my perls whose ivsize is 8 tested fine - but I haven't >>> yet had a chance to broaden the field to include any additional modules. >>> >>> My perls whose ivsize is 4 don't fare so well. Attached is the output >>> (gzipped) of running 'dmake test' with 32-bit perl-5.16.0. (The tests fare >>> pretty much the same with 5.8.8 - which is as far back as I go.) >>> >>> Hope to get a chance to take a closer look at this soon - but I have >>> limited opportunities for that over the next 4 or 5 days. >>> In the meantime, maybe the attached files are of some assistance to you. >>> >>> More later. >>> >>> Cheers, >>> Rob >>> >>> >> >
_______________________________________________ Perldl mailing list [email protected] http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
