Thu 14 Nov 2013 06:38:26 PM UTC, comment #5:
No, linear-algebra has already been moved to hg. And turns out that this bug has already been fixed by jwe with cset a29567b4b299.
The fact that the test in cellfun is using the @kronprod/iscomplex may be a separate issue since it should be used only for the kronprod class. But I can't reproduce it. Even while after loading the linear-algebra package with the buggy @kronprod/iscomplex.m, _run_test_suite_ does not error with cellfun.
|
Thu 14 Nov 2013 05:37:54 PM UTC, comment #3:
I think the iscomplex() shadowing is a bit of a side-issue. _run_test_suite_.m is undocumented and I haven't checked it out yet, so perhaps it is just rong invocation by me.
As to my Octave setup:
I hit this on a plain, normal cross-built MinGW Octave (tip from yesterday evening; unfortunately hg id's aren't kept in e.g., octave_config_info and as I'm on Windows now I can't check).
No claims from me this MXE builds are perfect; but it is built using the standard mxe-octave build chain.
I know about expected shadowing behavior with overloaded classes. Doesn't seem to work as advertised in this case.
FYI,
and fntests.log says:
I suppose it is better to unload all OF packages before running tests for core Octave.
|
Thu 14 Nov 2013 02:15:54 PM UTC, comment #1:
Patch looks simple enough and correct.
This iscomplex, and any function under an @ directory, should only be called when the argument is of the matching class type.
If "which iscomplex" shows "built-in function" then I'm not sure why this function would be called from run_test_suite (any arguments to that call?). If "which iscomplex" shows path/to/linear-algebra-x.y.z/@kronprod/iscomplex.m instead, then there's a problem with how your Octave path is being set up.
|
Thu 14 Nov 2013 12:28:45 PM UTC, original submission:
Patch (diff -u) attached, should be self-explaining.
On a more general note, how does this iscomplex.m differ from Octave's built-in core iscomplex.m?
From what I understand it looks like it works on class KP but I see no fall-back to the core-octave one for non-KP arguments. Which explains how I hit the bug:
I was running _run_test_suite_ in a MinGW buid and got two cellfun.c FAILs, caused by shadowing of iscomplex() by this OF-linear-algebra version of iscomplex.m
|