···<date: 2014-04-06, Sunday>···<from: Hans Hagen>···

> On 4/5/2014 3:12 PM, Philipp Gesang wrote:
> > ···<date: 2014-04-03, Thursday>···<from: Hans Hagen>···
> >
> >> On 4/3/2014 6:29 PM, Dohyun Kim wrote:
> >>> This is a bug report based on the issue discussed at
> >>> https://github.com/lualatex/luaotfload/issues/212
> >>>
> >>> \font\tenrm{file:lmroman10-regular.otf:mode=node;script=latn}\tenrm
> >>> \setbox0\vbox{%
> >>>     x\penalty-10000
> >>>     $a$x$a$
> >>> }
> >>> \unvbox0
> >>> \setbox0\lastbox
> >>> \unhbox0
> >>> \end
> >>>
> >>> The plain tex code shown above fails with a lua error:
> >>> luatex-fonts-merged.lua:9616: attempt to index local 'current' (a nil value)
> >>>
> >>> My guess is: as math-on (math-off too) is a "discardable" item after a
> >>> linebreak, the first math-on node has gone away. So the
> >>> math-on/math-off pair has become broken, by which the behavior of
> >>> "node.end_of_math" is confused.
> >>
> >> mtxrun --script plain test.tex
> >>
> >> runs ok here so i need an example that fails with context or luatex-plain
> >
> > It fails with the luatex-plain format from the minimals as of
> > yesterday and Luatex 4972:
> >
> >      error: .../context/tex/texmf-context/tex/context/base/font-ota.lua:144: attempt to index local 'current' (a nil value)
> >      .
> >      \bye ^^@-\par
> >                  \vfill \supereject \end
> >      l.17 \bye
> >
> > Though the issue arises only in node mode.
> >
> >> normally the font callback is called on a hlist and there are always
> >> paired math nodes
> >>
> >> (what is line 9616 in the current beta?)
> >
> > Line 9626 with yesterday’s code. Attached are patches for
> > font-ota.lua and luatex-fonts-otn.lua that guard all invocations
> > of end_of_math() with checks for the subtype.
> 
> I think that the removed math node is a bad thing so that will change in 
> a next luatex release (i tested a patch but it will not go in the tl 
> release). (Just like the math node at the end of a line it will have 
> zero surround in order to make the backend happy.)

I’d be interested in testing the patch as well.

> In order to avoid the 'current' issue I'll use an overload of 
> end_of_math in the generic code.
> 
> The real solution is to make the built-in end_of_math more strict 
> (trivial patch ready and tested so that can end up in the next luatex).
> 
> In the context code i won't intercept it because end_of_math will do 
> that at some point.

Good news. Though I’m a little worried about the fontloader in
Context and the generic format diverging too much.

Thanks for looking into this,
Philipp