Re: assert in page breaking

[Joe Neeman]

On 1/24/07, Han-Wen Nienhuys <address@hidden> wrote:
> Joe Neeman escreveu:
> > On 1/24/07, Han-Wen Nienhuys <address@hidden> wrote:
> >> +  if (chunks.size () != div.size () + 1)
> >> +    {
> >> +      programming_error ("did not find a valid page breaking
> >> configuration");
> >> +      ignore_div = true;
> >> +      assert (0);
> >>
> >>
> >> this is better, but can we skip the assert()?
> >
> > If this assert fails, it's a bug in the page breaker. I personally
> > prefer the assert because it means that if the bug appears during
> > "make web," I will find it instead of it being lost in the output.
> > This check really shouldn't depend on the validity of the user input
> > -- there are plenty of other sanity checks along the way to test for
> > that.
>
> It will show up on  'make check', because it would cause a diff in the
> logfile.

Is "make check" new or have I just missed it all this time? Just for
this, I'll remove the assert with no further complaints :)

> >> The problem with assert is that it prevents people from getting a PDF
> >> and looking at their score to see what the problem with it is.
> >
> > But the user builds are done with NDEBUG, right?
>
> Yes, but does it degrade gracefully when the assert is not triggered?

Sort of. If the condition in the assert is false, it means that page
breaking has failed. I try to put together some sort of a solution but
it will probably overflow pages and all sorts of nasty stuff. I also
need to check whether the fact that there are the wrong number of
systems could cause a crash further down the line.