On Fri, 22 Mar 2013 10:18:37 -0400 Stefan Monnier <monnier@iro.umontreal.ca> wrote: 

>> OK, but I can't think of any cases in CFEngine specifically where I
>> don't want to skip over _ with M-f.  It's always part of a word.
>> The language itself is designed that way and has a pretty rigid syntax.

SM> CFengine has no notion of words, it has a notion of identifiers, which
SM> in Emacs are called symbols and can be skipped with C-M-f or
SM> forward-symbol.

>> Or are you saying users don't expect _ to be part of a
>> word and it's unconventional to make it so?

SM> Exactly.  Many users actually like _ to be part of a word, but these are
SM> user-preferences (and we should indeed define a words-are-symbols-mode
SM> for those users since what they really want is for M-f, M-t, M-DEL to
SM> move by symbols rather than by words).

Yes!  In particular, context expressions in CFEngine are things like
"a_b_c.d_e" and it's a pain to skip over them with the usual `M-f' or
CUA keybindings.  It would be nice to be able to set this as a
preference for all modes derived from `prog-mode', IIUC what you mean.

SM> Basically, the definition of a word is something that major modes should
SM> not touch, because it is a notion from human language, and not from the
SM> programming language in use in a particular buffer.

SM> The major mode should define the syntax of symbols (aka
SM> "identifiers") instead.

OK, I think I understand.  You're saying "words" are not part of the
syntax for a programming language, generally, and are more of a
text-mode concept.  Symbols usually are what I want if I'm parsing a
programming language.

I modified cfengine.el to use \_< and \_> for all the regular
expressions; can you take a look at the attached patch and see if it's
closer to the expected behavior?

Thanks
Ted