Hello Alan,

Alan Mackenzie <acm@muc.de> writes:

> Hello, Naofumi.
>
> Thank you indeed for your bug report, and thanks even more for
> diagnosing the cause and writing a patch!
>
> I'm truly astounded that you managed to debug this problem, hidden as it
> was in the fontification parts of CC Mode, which are difficult to run
> with the debugger.  Respect!
>

Thank you, I'm glad to hear that.

> Looking at your patch, I can see one or two problems with it.  The main
> one is that after encountering a template opener "<", I'm a bit worried
> about the accuracy of checking for the matching ">".
>
> Looking further into the problem, it seems the misfontification bug was
> triggered by the commas in the template expressions.  When there's a
> comma there, the (c-forward-<>-arglist ...) call (which "; Should always
> work") fails.  This failure is caused by the dynamic binding of
> c-restricted-<>-arglists to t.
>
> When this dynamic binding is instead at nil (rather than t), the
> fontification works properly, with functions fontified as functions.
> The reason for this faulty binding was fixing a bug report where
> scrolling a buffer containing lots of templates was very slow.  Binding
> c-restricted-<>-arglists to t "solved" that slowness.
>

Indeed.  I confirmed that avoiding (c-restricted-<>-arglists t) binding
solves this misfontification problem at least, but it makes scrolling
terribly slow...

> That slowness had been caused by inefficient scanning of the template
> structures.  I've now identified and implemented an alternative
> optimisation, which doesn't cause problems with commas in templates.
>
[..snip..]
> Can I ask you, please, to apply and test my proposed patch, which I
> enclose below.  Would you please test it on your real C++ source code,
> and either confirm it really works, or say what's still faulty about it.
> If there are any problems with applying the patch, feel free to send me
> personal email.
>

I've applied your patch.  It works well and solves misfontification
problem.  At the same time, I've not seen any noticeable performance
degradation by this patch so far.  Great.

  attachment: fix_c-restricted--arglists_verification.png

I came across this problem when I tried to compare the current
tree-sitter's fontification quality and performance with familiar
cc-mode by using very-fast flat_hash_map code [1][2].  Now problematic
fontification has gone.

> So I must apologize that I won't be using your patch, but I thoroughly
> appreciate the time you must have spent debugging the problem.
>

That's just fine with me.  I don't mind at all because I'm not an
expert.  Fixing the root cause should be done.

Best regards,
  Naofumi

[1] https://github.com/skarupke/flat_hash_map/blob/master/flat_hash_map.hpp
[2] https://probablydance.com/2017/02/26/i-wrote-the-fastest-hashtable/