RFC: startup time - again

[Alexander Belchenko]

David Cournapeau пишет:
> Alexander Belchenko wrote:
>> So it's time to say that apriori bzr on windows is slow and unusable?
>> Py2exed bzr.exe has faster startup time than pure python bzr on windows,
>> but if you throw away test suite for library.zip it's also a bit faster.
>>
>> I found there is too much small things that could make bzr.exe a bit
>> faster.
>> And each of them seems too small, but in sum they are noticeable win.
> 
> I am not claiming that the comparison is really meaningful, but as a
> comparison point, for numpy, outside one or two things, that's the small
> things which mattered: 5% here, 4% there.

IIUC if you have 10 small things with 5% win in sum you likely will get 30-50% win?
Yes, that's my point too. But often developers say about one small case:
this is too small win to work on this.