Re: Font lock in org+elisp confused with ?\[

[Arthur Miller]

<tomas@tuxteam.de> writes:

> On Thu, Apr 01, 2021 at 09:18:30AM +0200, Arthur Miller wrote:
>> Tom Gillespie <tgbugs@gmail.com> writes:
>> 
>> > This is a known issue with org babel blocks. It is due to the fact
>> > that org babel translates the font locking for the language but not
>> > the syntax propertization. Another frequent cause is the bash case
>> > statement. The end result is that unmatched parens leak out from the
>> > babel blocks and wreak havoc elsewhere in the org file unless you
>> > balance out the parens e.g. in a comment. I have a hacked fix for
>> > this, but it has horrible performance, especially with line numbers
>> > enabled. I think that a proper solution would run arbitrary syntax
>> > propertization on subsets of a buffer without having to continually
>> > check where those subsets start or end. Best,
>> > Tom
>> 
>> Thanks for the explanation!
>> 
>> I don't know how org-babel works, I guess it opens another buffer,
>> font-locks syntax there and copy it over? What was the reason why they
>> don't copy other properties? Performance or something else?
>
> Hm. I've just poked around a bit. If you are talking about babel
> blocks, I think the entry point is `org-src-font-lock-fontify-block',
> defined in `org-src.el'.
>
> And indeed, this function basically does this (with a lot of
> corner-case chasing around, as is usual in our trade):
>
>   (let ((lang-mode (org-src-get-lang-mode lang)))
>         ...
>         (with-current-buffer
>             (get-buffer-create
>                (format " *org-src-fontification:%s*" lang-mode))
>                ...))
>
> The rest is left to the reader's fantasy, as they say :)

:-) Left as an exercise to reader's fantasy?

I tried to copy 'syntax-table property, but it didn't solve anything.

I think the problem is some bad regexp somewhere when it come to
indentation, it misses what it looks at somewhere (obviously),
but I have no idea where. I am not sure it is org fault.

I see gnus is confused too, so it is probably deeper, maybe in elisp
mode?

I took a profiler snapshot to get a nice stack-trace while moving the
cursor just from one place after parenthesis to the closing parenhesis
on this line:

(if (= (char-before (point)) ?\[)

I have attached profiler-repport.

I also noticed that the function got called ~1000 times, while there are
only ~140 src blocks.

profile-repport
Description: Text document

> Cheers

Happy Eastern!