[pypy-dev] incomplete builtin modules policy

[Samuele Pedroni]

Hi, we had failures re-appear for core tests because of some inprogress 
but still incomplete reimplementation of builtin modules got exposed. 
This is a problem because we can be changing things for example about 
the object model (objspace/std...) that are more related to the current 
core goal and we would like as few as possible unrelated failures to 
make it easy to spot failures produced by related changes and in general 
to track problems down.

As things are now implemented if a module _codecs or binascii is in 
pypy/lib it will be used even if the corresponding module is listed as 
needing faking in sys2/state.py, that means pypy/lib takes precedence 
over faking listing.

Given that they cause unrelated failures both binascii and _codecs need 
likely to be deactived by renaming them.

The question now is whether we want a mechanism such that locally it is 
possible to enable (e.g. even under the renamed name), so to be able to 
easely do testing for them for people working on them.

One possibility is to restore the old model where faking list take 
precedence and have support for a module that can put in place locally
under module/sys2 but is not checked in and that will be imported and 
can redefine, substract from the faking list.

The other possibility is to have such a module defines things like if 
'binascii' is requested to be imported use 'inprogress_binascii' instead.

Do we need such a feature, what approach would people prefer?

regards.