Re: diffutils __pycache__ failure.

[Bruno Haible]

Collin Funk wrote:
> >> I have no clue if this has a noticeable performance impact or not.
> > 
> > Can you measure it, please? For example, with
> >   GNULIB_TOOL_IMPL=py time ./test-all.sh
> > 
> > I measure a difference in the 2% range, but it's not clear to me whether
> > -B slows down or speeds up things :)
> 
> Sure, here is the results using the -B flag. I'm removing the
> __pycache__ directory before using -B flag to make sure it doesn't get
> read.
> 
> Using 'env GNULIB_TOOL_IMPL=py ./test-all.sh' in import-tests:
> 
>       no -B flag: 0m16.699s
>       -B flag: 0m20.892s
> 
> Using 'env GNULIB_TOOL_IMPL=py ./test-all.sh' in create-tests:
> 
>       no -B flag: 2m45.046s
>       -B flag: 2m46.674s

Thanks for measuring it. So, the -B flag causes a slowdown.

> The create-tests spend most of their time in autoconf and friends if I
> remember correctly.
> 
> The import tests feel noticeably slower with -B to me.

This is explained by the fact that the import tests do nearly 100
gnulib-tool invocations: The same just-in-time compilation must happen
in memory 100 times. This explains the 4 seconds of slowdown.

Bruno