[metapost] Uninitialized variable sometimes causing malformed TFM files
Marcel Fabian Krüger
tex at 2krueger.de
Sun Jun 16 15:40:34 CEST 2019
On Wed, Mar 20, 2019 at 08:26:22PM +0100, Marcel Krüger wrote:
> ---- On Wed, 20 Mar 2019 19:01:08 +0100 luigi scarso <luigi.scarso at gmail.com> wrote ----
> >
> >
> > On Wed, Mar 20, 2019 at 6:54 PM Marcel Krüger <tex at 2krueger.de> wrote:
> > Thanks. There is one other thing, I do not really know if I would describe it as a backward compatibility bug or a feature request:
> >
> > For non-"scaled" numbersystems, code like
> >
> > filldraw stroke z1e--z2e;
> > (example taken from cmbase.mf, `left_bracket`)
> >
> > fail, because z1e-- is interpreted as z 1e- - instead of z1e --. So z1e--z2e is equal to z1-z2e, leading to lots of errors.
> >
> > I understand that the new numbersystems and the exponential syntax can lead to problems of this kind, but I think this could be avoided:
> > Maybe a `e+` / `e-` could only be interpreted as exponential notation if it is followed by a digit? I do not think
> > anyone would write "1e-" or "3e+" for 1 or 3, so it should not lead to any breakage.
> > On the other hand it would catch almost all cases where this is invoked accidentially.
> >
> > A possible implementation of this change is attached.
> >
> >
> > hm it looks like a serious bug, and I fear I will not able to fix it for next TL --- but I hope to put it into trunk immediately after the TL code is frozen.
> >
>
> Right, I am sorry for bringing this up so late in the TL cycle.
>
> One additional note on this: The behaviour described above (`1e-` being interpreted as `1`) seems to be specific to the `double` mode.
> The `decimal` mode just fails because it is not able to parse the number, which further reduces the risk of breaking some code by changing the current
> behaviour.
>
> Also I attached a fixed version of the earlier patch, there was a smaller error.
>
> Best regards,
> Marcel Krüger
>
> >
> > --
> > luigi
> >
Hi,
are there still plans to fix this?
Best regards,
Marcel
More information about the metapost
mailing list