On 2009-01-02 17:10, Antoine Pitrou wrote:
> Antoine Pitrou <pitrou@free.fr> added the comment:
> 
>> This is safer than enabling the support unconditionally for GCC and
>> the SUN Pro C compiler, since it is rather likely that some GCC versions
>> have bugs which could render Python unusable if compiled with the
>> dispatching support enabled.
> 
> What do you mean, "unusable"? 

Well, not working. GCC versions often have optimizer bugs (esp. the
3.x series and early 4.x versions) and I wouldn't bet on them always
getting the dispatch optimizations right.

Trying to compile Python with an unconditionally enabled dispatch
patch on such a system would render Python unusable.

> 10% slower? Well, Python 3.x is already
> unusable (compared to 2.x) by that metric... Until now, only Skip has
> reported a slowdown on his PPC machine, while x86 systems have all seen
> a positive (if tiny, sometimes) improvement.
> 
> I fear that with a configure option, disabled by default, the code will
> get very poor testing and perhaps get broken in some subtle way without
> anyone noticing.

Like Skip said: the buildbots could take care of identifying such
problems.

People using the option would certainly report problems as well and
I'm sure that Linux distributions would compile Python with the switch
after verifying that their GCC version works correctly.