Stefan Behnel wrote: > Dag Sverre Seljebotn wrote: > >> let's get as many as possible to test with the refnanny turned on >> (and let's make it easy, say, let's include the code to stick in ones >> setup.py in the RC release notes). >> > > Would there be a way to enable that automatically for an alpha release? I > know, this requires the external refnanny module, but wouldn't it be > possible to enable it by default, and then add a fallback dead-nanny into > the generated code that would just ignore calls and would be used if the > refnanny itself isn't found? That would eat some performance but would > allow users to enable the refnanny by simply having it lying around in > their PYTHONPATH, instead of requiring a separate build option. > A basic "setup.py install" also installs the refnanny (in Cython.Runtime.refnanny), so in many circumstances it would be there. It would be possible to generate C sources so that CYTHON_REFNANNY=1 by default as well, and has to be set to =0.
I'm -1 on a transparent fallback if the refnanny module is not there though, too much that starts clouding the "exactly what went on" which is so important when getting email reports. Also having actual behaviour differ in RC and final release like this makes me uneasy (what if people reports performance regressions?...i.e. one must be aware of it). All in all, +0 on CYTHON_REFANNAY default to 1 and -1 on fallback module from me. Dag Sverre _______________________________________________ Cython-dev mailing list [email protected] http://codespeak.net/mailman/listinfo/cython-dev
