[Pyrex] Testing on Linux

[Greg Ewing]

Stefan Behnel wrote:

> BTW, is there a way to regenerate the C files that are used for comparison? I
> tried running the test suite on my modified Pyrex version and it goes off
> complaining at almost any test that it found differences from the expected C
> output.

Erm, yes, that is a problem. I have a system for dealing with
it on a case-by-case basis, but it's not pretty.

If you look in Pyrex/Testing.py, in munge_c_line(), you'll
find the place where I insert ad-hoc hacks to modify the
lines being compared so as to cover up certain kinds of
differences.

Whenever I make a change that results in widespread systematic
changes to the code generation, the procedure I use is:

(1) Add a hack to munge_c_line() which stops the tester from
     complaining about the difference.

(2) Run the tests.

(3) When all the tests pass (and only then!), run

       update_references

     to copy the last-generated versions of the output files
     into the References directories.

(4) Remove the hacks added in step (1).

The other technique that I use, for when only a few tests are
affected, is just to visually examine the output, make sure
the difference isn't due to a bug, and then hand-edit the
reference file to match it.

The important thing is to be careful with the reference files
at all times, and never run update_references until you're
as certain as you can be that everything is correct.

--
Greg