Re: [boost] Re: status/Jamfile messed up for type_traits...

From: David Abrahams (dave_at_[hidden])
Date: 2002-11-04 09:03:12

"John Maddock" <jm_at_[hidden]> writes:

> > > Don't apologise, as you will have noticed the type_traits tests are in a
> > > state of flux right now, they will settle down soon (honestly), and yes
> that
> > > build everything test will be removed from status/Jamfile, it will stay
> in
> > > libs/type_traits/test/Jamfile though as it is sometimes useful to build
> and
> > > check everything at once with the more conforming compilers.
> >
> > I don't see how that will make any difference. It certainly won't
> > speed up whole-system testing. If you don't refactor the rule
> > invocations it's going to be a maintenance nightmare wherever you put
> > that code. Type traits are central to the operation of many boost
> > libraries, and if people make changes which break them on many
> > compilers (ahem! cough! -- they're broken now!) , somebody has to dive
> > in and start running the tests.
> 
> Actually it's a *lot* quicker to build and run the tests as one big test
> suite than as lots of separate ones

I wasn't suggesting that you break the tests into separate suites. I
was suggesting that you write a simple Jam rule that encapsulates
everything you're repeating for each test in the Jamfile.

> as for maintenance the Jamfile is automatically generated by a
> little shell script to keep maintenance down.  

Now that's just silly, and I daresay somewhat distressing. We went to
all this effort to provide a reasonably powerful programming language
in Jam which allows you to avoid writing the same thing over and over
again; you could clean this all up with a few lines of Jam code, and
instead you add a layer of scripting on top which generates the
"usual-quality" machine-generated code. I just can't understand why
that's preferable.

> BTW there should be very few tests failing now, is_convertible may
> be broken on Intel C++, but I'm waiting for the next lot of test
> results to come back on that one.  If there are still failures that
> you're not expecting it's probably because the new tests are quite a
> bit more thorough than the old ones and are picking up new (old)
> compiler bugs (shrug).

More-thorough tests are good. Messy machine-generated Jamfiles are
bad.

-- 
                    David Abrahams
dave_at_[hidden] * http://www.boost-consulting.com