Improving the regression testing infrastructure

[Eric Pouech]

Francois Gouget a écrit :
> 
> On Fri, 29 Mar 2002, Eric Pouech wrote:
> 
> > > at least until I can find a good way to test
> > > CreateProcess...
> > a bit out of topic (from the question), but I have an
> > embryonic CreateProcess test, so don't bother with that
> > (and there is a lot to test here)
> 
>    This is interesting. How do you test it?
>    AFAICS, The problem with the current infrastructure is that you
> cannot have a second executable to start. So you have to start yourself,
> and then convince the test framework to let you do your thing rather
> than just running another test. You cannot just write a perl/bash script
> either as this would not work on Windows.
>    If you put submit a CreateProcess test I would be happy to add checks
> for the quote / command line parsing.
as of today, the test restarts itself, with a bit of a different command
line
it's written in C, but didn't test it on windows yet (I know a few
issues
already)

the global test goes as :
1/ the parent creates the child
2/ the parent waits for the child to terminate
3/ the child process stores most information it can gather in a
temporary 
   file
4/ the child terminates
5/ the parent does the final checking

I did choose the temporary file (instead of registry for example) to be
as
independant as possible of other APIs (even if the format of the file is
a 
.INI one, and as of today, the parent reads it with private profile API)

one of the checking that won't be possible is all the path lookup
(except
if we decide to copy the exec in different places :-(

I'll put something out when I have the test working in Windows

A+