Bad skip condition definition can abort entire test run

[apjanke]

The doctest code that decides whether a test is skipped is an unprotected eval.

  % determine whether test should be skipped
  % (careful about Octave bug #46397 to not change the current value of “ans”)
  eval (strcat ('DOCTEST__current_test.skip = ', ...
                 doctest_join_conditions(DOCTEST__current_test.skip), ...
                ';'));

This means that an error in a tested file's skip condition definition will bubble up to the top, and abort the entire test run. This condition occurs with the doctest code in the Octave Forge interval and symbolic packages.

Maybe this eval(), and the others in doctest_run_tests, should be protected by a try/catch that turns a bad skip or xfail definition into a test failure, instead of aborting the run?

Caused apjanke/octave-testify#44.

[mtmiller]

I think eval might also need to be fixed to do the protection itself. This was brought up in Octave #55939, if I understand the author, eval in Matlab does catch errors.

[apjanke]

That‘s a surprise to me. Matlab eval normally throws errors. I think they’re encountering an undocumented Matlab behavior related to an undocumented second argument to eval there. Normally, it's only str2num() that doesn't throw errors, and that’s probably because str2num is internally try/catching the eval(...) that it does.

[mtmiller]

Ok, unclear to me, fair to continue under the assumption that eval may raise an error.

[cbm755]

At the very least, surely syntax errors in the str must raise errors outside eval so I think we (Doctest) should do try-catch here.

Maybe do the "EXTRACTION_ERROR" on catch.

[apjanke]

The fix looks good to me.