Re: emacs 22 - regular-expression isearch on spaces extremely lenient

[ken manheimer]

Drew Adams <drew.adams <at> oracle.com> writes:

> [Ken Manheimer wrote:]
>
>     i recently started noticing that emacs 22 regular expression isearches
>     do not treat spaces exactly - any number of spaces in your search will
>     map to any number, same or different, of spaces in the target.
> 
>     can anyone tell me whether or not it's deliberate, and what the
>     rationale is?
> 
> (setq search-whitespace-regexp nil) will turn this off. When this is nil,
> each space you type matches literally, against one space.

thanks!  i did infer that from a bit of investigation (in isearch-search), and
am kinda mortified by how unobvious it is.  i agree very strongly with your
analysis, and feel the feature is too elusive in too many ways - all the
questions i was asking: behaviorally ("wait a minute, does it really do that,
and if so, when?"), properness ("is it really supposed to do that?"),
control/inhibition (how the hell do i turn it off?!") - that it's a Real Bad
Idea.  oh well.

i agree that making the control obvious, and defaulting it to off, is a much
better way to make the feature available.  (i could even see having a transient
message in the minibuffer whenever someone types in a space, saying how to make
it a fuzzy space with Ctrl-Q.)

> `search-whitespace-regexp' is, by default, "\\s-+", which searches for any
> amount of whitespace when you type a space. This was introduced for regexp
> search in Emacs 21, I believe. There is no such "magic-space" searching in
> Emacs 20. Doc:
> 
>  If non-nil, regular expression to match a sequence of whitespace chars.
>  This applies to regular expression incremental search.
>  When you put a space or spaces in the incremental regexp, it stands for
>  this, unless it is inside of a regexp construct such as [...] or *, + or ?.
>  You might want to use something like "[ \t\r\n]+" instead.
>  In the Customization buffer, that is `[' followed by a space,
>  a tab, a carriage return (control-M), a newline, and `]+'.

i didn't notice anything about it in the documentation for isearch, though now
that i look at the section on regular expression search i see something about
it.  this is a pitfall of rich features - too much functionality to describe
adequately without overwhelming.

> The rationale was, I believe, that some users might want that: type space to
> find any amount of whitespace, in particular, to find two words that are
> separated by a newline.

i would have suggested using "\ " or "^Q " (which both currently now have the
opposite effect).

> There was talk of using this "magic-space" searching also for plain
> incremental search in Emacs 22, but I don't think that was done.

good.

> FWIW, I agree with Miles on this - this is a misfeature, if turned on by
> default. It should be off by default, and you should be able to turn it on
> via a simple toggle during incremental search (regexp or plain).

who's miles?

> Here is what I wrote 2005/02/06 to emacs-devel on this:
> 
>             > > sometimes the actual whitespace matters.
>             > Right: in *regexp* search.
>           while people generally expect regexp searches to be a bit
>           fuzzy, they might expect a non-regexp search to be exact.
>           Since the fuzzy whitespace matching often "looks" like normal
>           matching (because the majority of whitespace is in fact a
>           single space), it might take some time to see what's going on,
>           resulting in some subtle errors. This is particularly true if
>           one embeds a search inside a keyboard macro [which I often do].
> 
>     Plain (incremental) search should be a literal search. Regexp
>     search should rigorously respect the regexp. People don't expect
>     either to be fuzzy.
> 
>     The question is "Under what circumstances should typing a space be
>     interpreted as wanting to search for any amount of whitespace?"
> 
>     This is unrelated to both plain search and regexp search. You
>     might or might not want this _input effect_ with either plain
>     or regexp search.
> 
>     This is akin to word search (as I think someone mentioned).
>     Ultimately, a word search or a space-means-whitespace search is
>     implemented with a regexp search - but the point in both cases
>     is to provide a user-friendly way to do it, instead of requiring
>     users to know about regexps.
> 
>     By default, neither `C-M-s' nor `C-s' should respect the
>     user-friendly space-input feature. Or, rather, the default
>     behavior of each should be determined by a user option - a la
>     case-fold-search. And, regardless of the value of this option,
>     you should be able to toggle space-means-whitespace
>     searching from both `C-M-s' and `C-s', via a key sequence.
> 
>     The question then becomes how to toggle this space-means-whitespace
>     searching?

thanks much for the thorough answer - it's good to have the mystery settled,
even if i don't agree with the situation.  at least i now know various ways to
work around it...

(i guess it's too late to propose inverting the mode of the feature on
emacs-devel?  i'd just copy this posting over there, asking if it's worth
reopening the discussion.  i wouldn't want to do that, though, if the discussion
ended on a conclusive pronouncement that even dismissed revisiting in the event
of complaints...)

ken manheimer
http://myriadicity.net