Re: [Quilt-dev] [patch 0/8] test/run: Cleanups and optimizations

[Jean Delvare]

Le Wednesday 05 February 2014 à 11:41 +0100, Andreas Grünbacher a
écrit :
> I don't have an idea how much of that time is actually spent in quilt; that
> time wouldn't change with a different test driver.

I don't know either. I really would love to have these numbers but I
just don't know how to extract the information. There are so many
layers...

The user's shell calls "make", which checks the Makefile and runs perl
on the "run" script for every test case which hasn't succeeded yet. Then
perl opens the test case file and for most commands it has to fork. For
every quilt command that means spawning a new instance of bash. And for
some commands we start with spawning a new instance of /bin/sh just to
be able to parse the command line. That's about 1000 forks for the
entire test suite.

I very much doubt that replacing "run" with the fastest piece of code
will make much of a difference in the total execution time as long as we
have that many layers and forks. The only significant performance
improvement would come from being able to run a complete test case
within a single process. That would be quite a different approach
compared to what "run" does right now. But somehow I suspect it might be
difficult to do that and that may be the reason why "run" is implemented
the way it is.

Anyway, I'm busy enough as is and will not dive into this. This would
certainly be a cool Suse hackweek project if I did not already have 11
to choose from...

-- 
Jean Delvare
Suse L3 Support