Re: perl mode color highlighting working badly

[martin rudalics]

> The problem is that in perl-mode some syntax-table text-properties are set
> in font-lock-syntactic-keywords, while others are set in
> font-lock-syntactic-face-function, and that font-lock+jit-lock applies
> font-lock-syntactic-keywords to "all" the buffer but only applies
> font-lock-syntactic-face-function to the displayed part of the buffer.
> So some of those properties can be missing if you start display from
> somewhere else than the beginning of the buffer.  Hmm...


In general this problem occurs with patterns like

foo
---
bar

where syntax or appearance of `bar' depend on something font-lock
calculates when it encounters `foo' and font-lock's scan stops at the
`---'.  In the special case I believe it was displaying a file
immediately as

foo
--- window-start ---
bar

and jit-lock - triggered by redisplay - fontified the `bar' part only.
However, the problem may also occur when you change within a window

foo
---
ba

to

foo
---
bar

which to my knowledge neither font-lock nor jit-lock can handle.  Or,
insert a stretch of text containing `foo' and `bar' such that `foo' ends
up before `window-start' and `bar' within the window.  This would be
handled by pure font-lock but here `font-lock-syntactic-keywords' might
fail with jit-lock since neither `font-lock-syntactically-fontified'
gets adjusted properly nor is the inserted region fontified immediately.

Misfontifications may persist even when you "scroll `foo' into the
window" since fontification routines usually don't refontify text -
although they might set multiline text properties correctly - this could
be probably fixed in the actual example.  Misfontifications are not
corrected by stealth fontification routines either since the latter do
not refontify displayed text.

I intend to handle this with `jit-lock-context-fontify' in a reasonable
way.  To make this practicable, however, I should reduce refontification
overhead introduced by the latter first.


> Already applying font-lock-syntactic-keywords to "all" the buffer defeats
> the purpose of jit-lock.  Applying font-lock-syntactic-face-function in the
> same way would defeat it even more, and this only because perl-mode uses it
> in a sneaky way (perl-mode is the only mode which suffers from this
> problem).

jit-lock breaks font-lock's handling of syntax-table properties anyway.
Having syntactic properties behave correctly only for positions
preceding `font-lock-syntactically-fontified' means that I can't parse
text after that position reliably with `parse-sexp-lookup-properties'
non-nil.  By the way, doesn't `tex-font-lock-syntactic-face-function'
try a similar hack with "\verb?...?" constructs?  And there's also the
commented out `python-font-lock-syntactic-face-function'.