Re: forward-comment and syntax-ppss

["Andreas Röhler" <andreas.roehler@online.de>]

On 06.12.2016 23:56, Stefan Monnier wrote:
>> I'm not quite sure what you're saying here.  Normally, a piece of
>> software not working properly is an extremely strong argument against it
>> (though it's an equally strong argument for fixing it).
> It's probably a strong argument against installing it willy nilly, yes.
>
>> What, exactly, is this patch intended to do?  That's not obvious.
> It gets rid of some pathological cases where backward scanning currently
> may re-scan the whole buffer over and over again, each time from
> point-min, resulting in excruciatingly slow backward movement (these
> cases are fairly rare, luckily).
>
>> I think this would be a last resort, when nothing else can work.  The
>> syntactic primitives should not have high level lisp code poking about
>> in their innards.
> I don't much care for such philosophy, because I consider syntax-ppss to
> be low-level Lisp code.
>
>> When it comes to narrowing a buffer, I think that the syntactic
>> context of a buffer position should not depend on whether the buffer
>> has been narrowed.
> As you know by now this is just one opinion, and other users of
> narrow-to-region beg to differ, and hence we will need to provide a way
> for users of narrow-to-region to communicate their intentions, since
> without that extra information, there can be no reliable caching of
> parsing.
>
>> If there are use cases this inconveniences, I think that those use
>> cases' use of narrowing is mistaken, and some other tool should be
>> used in its place.
> Fine by me.
>
>>> So I see this patch as an opportunity to improve syntax-ppss.
>> I don't think anybody will object to improving syntax-ppss.  But, at the
>> same time, calling syntax-ppss from the C syntax routines should be
>> avoided, as far as possible.  Such calls are likely to transform
>> side-effect free functions (such as forward-comment) into side-effect
>> encumbered functions, which is likely to make them less useful, and to
>> make Emacs more buggy, less flexible and more difficult to use.
> We've happily crossed that bridge already (tho not by calling
> syntax-ppss but by calling syntax-propertize).
>
>> And, just to emphasise, I think it's clear that syntax-ppss has no
>> business in forward-comment.  syntax-ppss is slow (compared to the
>> current implementation of forward-comment),
> The patch I showed only uses syntax-ppss when we would otherwise
> re-parse the whole buffer, so it shouldn't slow things down too much.
>
> BTW, when writing syntax-ppss I did not really intend to "speed things
> up", but to get rid of pathologically slow behavior.  In many cases
> calling parse-partial-sexp from point-min will be faster than calling
> syntax-ppss (mostly because syntax-ppss is all written in Elisp, so just
> the time to consult its cache can be longer than the time to compute the
> answer from scratch), but its behavior should stay "good enough" in
> a longish buffers (where calling parse-partial-sexp from point-min
> becomes excessive).
>
> IOW, syntax-ppss will be slower than parse-partial-sexp when
> parse-partial-sexp is fast, but it will be faster when
> parse-partial-sexp is slow.
>
>> and there's no need for it, given the existence of the
>> comment-cache branch.
> This branch has fundamentally the same caching problems as syntax-ppss
> (i.e. in case of narrowing it may sometimes return what the user
> intended and sometimes not), unsurprisingly.  And it also suffers from
> "Such calls are likely to transform side-effect free functions (such as
> forward-comment) into side-effect encumbered functions".
> And of course, it's a much larger change, with a redundant cache.
>
> So, from where I sit I'd put it the other way around:
> "there's no need for the comment-cache branch, given the existence of
> this syntax-ppss-based simple patch".
>
> Of course, you could argue that your patch works with cc-mode whereas
> mine might not.  It's no surprise, since you actually changed cc-mode
> to accommodate your comment-cache, like you could do to accommodate
> this patch.
>
>
>          Stefan
>

syntax-ppss has major design issues. Just look at this comment:

           ;; Use the `syntax-begin-function' if available.
           ;; We could try using that function earlier, but:
           ;; - The result might not be 100% reliable, so it's better to use
           ;;   the cache if available.
           ;; - The function might be slow.
           ;; - If this function almost always finds a safe nearby spot,
           ;;   the cache won't be populated, so consulting it is cheap.