bug#24172: 25.1; Doc of parse-sexp-ignore-comments: what does a value of nil mean?

[Lars Ingebrigtsen]

(Looks like the debbugs address in the Cc was mangled, so I'm just
resending the entire email to have it end up in the bug tracker.)

Alan Mackenzie <acm@muc.de> writes:

> Hello, Lars.
>
> In article <mailman.2415.1564240928.2688.bug-gnu-emacs@gnu.org> you wrote:
>> Michael Heerdegen <michael_heerdegen@web.de> writes:
>
>>> in the docstring of `parse-sexp-ignore-comments' (or at least in the
>>> manual), I miss a description about what a value of nil exactly means:
>>> How are comments treated then?  Are they treated as separate units that
>>> can then be parsed as well, but separately from code, or are they treated
>>> as indistinguishable from code?
>>>
>>> For example, if parsing starts from within a comment, and parsing finds
>>> the end of the comment and is not yet finished, is parsing just
>>> continued inside the following code, or does it fail?
>
>> After doing some testing, it seems that if it's nil, the commands
>> affected by the setting just treat the commented-out text as if it
>> wasn't commented out.
>
>> So the answer to your last question seems to be "yes".
>
>> ;; (foo
>
>> (bar zot))
>
>> However, pretty much the same thing happens with a non-nil value, too --
>> with point before (foo C-M-f will advance past zot)).
>
>> So it doesn't treat comments as whitespace, really -- it only does that
>> if point is outside (before, at the end of a line, etc) the comment to
>> begin with.  Seems like you could write an essay about it, but perhaps
>> it's not worth listing the eccentricities here which I guess could change.
>
> parse-sexp-ignore-comments (or, rather, parse_sexp_ignore_comments) is
> tested, and acted upon, by precisely one C function, scan_lists in
> syntax.c.  scan_lists is used by just two primitives, scan-lists and
> scan-sexps.  These, in turn, are called by forward-list, etc.
>
> When this variable is nil, scan_lists fails to recognise comment
> delimiters - it just goes past them as though they were random
> punctuation characters.  By contrast, when parse-sexp-ignore-comments is
> t, scan_lists calls a comment scanning function when it encounters a
> comment delimiter.
>
> Noteworthy is that parse-partial-sexp doesn't respect the setting of
> parse-sexp-ignore-comments.  This might be considered a bug.
>
> All in all, this variable seems not to be a good idea.  It is not tested
> consistently by the syntax routines (see above), must be set explicitly
> to t by any major mode with comments, the nil value is rarely used, and
> it is not clear whether this nil value is actually ever useful.
>
> It would seem that if a major mode does not want comments to be
> recognised, it would be better not to give any character comment syntax
> in its syntax table.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no