apjanke
commented
5 years ago inputname:
Aw, crap. At least some of the BISTs are depending on the internal implementation details of test.
Here's some test code from inputname.m:
## Warning: heap big magic in the following tests!!!
## The test function builds a private context for each test, with only the
## specified values shared between them. It does this using the following
## template:
##
## function [<shared>] = testfn (<shared>)
## <test>
## endfunction
##
## To test inputname, I need a function context invoked with known parameter
## names. So define a couple of shared parameters, et voila!, the test is
## trivial.
%!shared hello, worldly
%!assert (inputname (1), "hello")
%!assert (inputname (2), "worldly")
%!assert (inputname (3), "")
## Clear parameter list so that testfn is created with zero inputs/outputs
%!shared
%!assert (inputname (-1), "")
%!assert (inputname (1), "")That seems to rely on the fact that Octave's existing test implements shared variables by passing them as parameters to a testing function call that it generates dynamically, and that that testing function (with shared variables passed as named parameters) is the immediately enclosing stack frame in which statements in the test block are executed. That's incompatible with the object-managed workspace approach that Testify is using.
Testify is failing more tests than core Octave is.
And why is it reporting more PASSes, too? 18090 vs 14933.
Files with failed tests
libinterp/corefcn/pr-output.cc-tstgetframe.minputname.misocaps.misosurface.mreducevolume.mcopyobj.mhgsave.mAnd doesn't that seem like too few files to have 34 failures?