tinderbox information.

[Jim Gettys]

On Tue, 2004-03-09 at 15:47, John Dennis wrote:
> On Tue, 2004-03-09 at 14:46, Jim Gettys wrote:
> > Hmmm... my test for a failed build is buggy.  I'll go fix...
> > 
> > Sorry for having the tinderbox cry wolf.
> 
> Jim, when we worked on this last week I changed the succeed/fail test
> from grepping the log file to checking the error status. I see you
> changed it back. Why?

make -k always returns success.  So status can't be used if
one wants to do -k builds.  So I had to change it to do -k builds.

I think make always returns error messages with "***" which
can be used to check for any errors, so long as make is being
used to to do builds; this seems independent of the return status,
thankfully, and isn't typically generated by other commands.

My build test is currently failing as I should have done an
fgrep rather than a grep.  I'll check in a fix in a moment.

> I don't think grepping is a very robust method, first of all you can
> have all sorts of failures, not just those that output a known string,
> second, you've tied the script to a hard coded file, and we tried to
> eliminate all hard coded files, third, if you are going to grep the log
> file you should be getting the file path name from the upper control
> structure, otherwise you might not be looking at the right file and it
> prevents multiple simultaneous builds, another goal we had.
> 

Well, the log file it is checking is the tinderclient.log file,
which will be the same even in different modules.  So I don't
think it is tied to a given module, though it is depending on a
"feature" of make.  Whether this is true about make on other
platforms (non-gnu make), I don't know off hand.

Arguably, though, the tinderclient.log file should go into
the build directory (XMonolithic) so that multiple builds could
go on in the same build directory; but that is a nit.


                          - Jim

-- 
Jim Gettys <Jim.Gettys@hp.com>
HP Labs, Cambridge Research Laboratory