Re: [PATCH v2] Fix crash when a stencil routine is missing

[Patrick McCarty]

On Mon, Jul 06, 2009 at 03:08:50PM -0700, Graham Percival wrote:
> On Mon, Jul 06, 2009 at 07:09:11AM +0200, Werner LEMBERG wrote:
> > > and I'm not certain that we have enough Frogs to eat all the issues
> > > that would be added if we went this route.
> > 
> > Perhaps it makes sense to classify regression tests like this:
> > 
> >   a) intentionally emit warnings and errors
> >   b) unintentionally emit warnings
> >   c) run fine
> 
> I don't see a huge urgency for this patch, so I would rather wait
> until b) was non-existent.  I totally agree that we should split
> a) and c), though.

I think there is a slight confusion about what my patch implements.

The new option, -dwarning-as-error, does *not* turn every warning into
an error.  It only turns specific warnings (that don't exist without
this patch) into errors.

The purpose is more of a long-term solution, so you're right that it's
not at all urgent.

For example, with this patch, if you compile a file containing three
"embedded-ps" stencil expressions with the SVG backend, you'll get
this output instead of a crash:

  $ lilypond -dbackend=svg test.ly
  GNU LilyPond 2.13.4
  Processing `ps.ly'
  Parsing...
  Interpreting music... 
  Preprocessing graphical objects...
  Solving 1 page-breaking chunks...[1: 1 pages]
  Drawing systems...
  Layout output to `ps.svg'...
  warning: missing stencil expression `embedded-ps'
  warning: missing stencil expression `embedded-ps'
  warning: missing stencil expression `embedded-ps'

With the -d option, you get this:

  $ lilypond -dbackend=svg -dwarning-as-error test.ly
  GNU LilyPond 2.13.4
  Processing `ps.ly'
  Parsing...
  Interpreting music... 
  Preprocessing graphical objects...
  Solving 1 page-breaking chunks...[1: 1 pages]
  Drawing systems...
  Layout output to `ps.svg'...
  error: missing stencil expression `embedded-ps'

and an invalid SVG file.

But, since the regression test suite only exercises the PostScript
backend, which doesn't currently have any missing any *defined*
stencil expressions, we won't be seeing any failures right now.

> Let's wait until Patrick reports how many regtests unintentionally
> emit warnings, and see how much work is involved in fixing those.

There are no regtests that emit the new warning.  There are plenty of
regtests that emit warnings, but not the warning in question.

If we want to implement an option to turn *all* warnings into errors,
that would require a different patch.

So, maybe we should change the -d option name after all.  :-)

Thanks,
Patrick