perl-mode "::" as word character [patch]

perl-mode "::" as word character [patch]

[Pandora]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Just dropping in for a teeny bit.  I was studying font-lock innards the
other day and noticed that the perl-mode font lock does a lot of
wrangling around the '::' sequence.  That's what's used in perl to
divide package names from package members.  But instead of sticking
\\(::\\sw\\)*s in after every \\sw, I noticed it'd be much simpler just
to consider the double colon (NOT the single colon) to be a \\sw
character.  Seems pretty cut and dried, since :: was only introduced to
perl as a namespace separator, and can be wholly considered as if it
were a word character.

Also since my experience has been that print/printf are used like
keywords, but act like functions; they could use some special
highlighting.  As for functions in general, both &\\(\\sw+\\) and
\\(\\sw+\\)( match a function name in perl, but perl-mode only
highlights the first expression as a function.

So... I went and made a patch against CVS, in case anyone else wants to
use these three quick little hacks.
Index: lisp/progmodes/perl-mode.el
===================================================================
RCS file: /cvsroot/emacs/emacs/lisp/progmodes/perl-mode.el,v
retrieving revision 1.56
diff -r1.56 perl-mode.el
207a208
>     ;; Fontify print and printf as functions, typically w/o ()
208a210,211
>     '("\\<\\(print\\|printf\\)\\>" 1 font-lock-function-name-face)
>
210c213,214
<     '("&\\(\\sw+\\(::\\sw+\\)*\\)" 1 font-lock-function-name-face)
- ---
>     '("&\\(\\sw+\\)" 1 font-lock-function-name-face)
>     '("\\(\\sw+\\)\\s *(" 1 font-lock-function-name-face)
213,214c217,218
<     '("[$*]{?\\(\\sw+\\(::\\sw+\\)*\\)" 1 font-lock-variable-name-face)
<     '("\\([@%]\\|\\$#\\)\\(\\sw+\\(::\\sw+\\)*\\)"
- ---
>     '("[$*]{?\\(\\sw+\\)" 1 font-lock-variable-name-face)
<     '("\\([@%]\\|\\$#\\)\\(\\sw+\\(::\\sw+\\)*\\)"
- ---
>     '("[$*]{?\\(\\sw+\\)" 1 font-lock-variable-name-face)
>     '("\\([@%]\\|\\$#\\)\\(\\sw+\\)"
258a263,264
>     ;; Or $PKG::member  :: is a word character, really.  c.c
>     ("\\(::\\)" (1 "w"))
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFCcyn2B/meY5RuPPQRAlJeAKCruu9fvqzETQhQQTE9TV70WSTLqgCdHsDL
R1rwSw/KRWeOvYaGIzb9+iM=
=dJBg
-----END PGP SIGNATURE-----

Re: perl-mode "::" as word character [patch]

[Stefan Monnier]

> Just dropping in for a teeny bit.  I was studying font-lock innards the
> other day and noticed that the perl-mode font lock does a lot of
> wrangling around the '::' sequence.  That's what's used in perl to
> divide package names from package members.  But instead of sticking
> \\(::\\sw\\)*s in after every \\sw, I noticed it'd be much simpler just
> to consider the double colon (NOT the single colon) to be a \\sw
> character.  Seems pretty cut and dried, since :: was only introduced to
> perl as a namespace separator, and can be wholly considered as if it
> were a word character.

Word syntax is clearly wrong.  Symbol syntax (i.e. "_") OTOH sounds right.

> Also since my experience has been that print/printf are used like
> keywords, but act like functions; they could use some special
> highlighting.  As for functions in general, both &\\(\\sw+\\) and
> \\(\\sw+\\)( match a function name in perl, but perl-mode only
> highlights the first expression as a function.

I don't think function calls should be highlighted, only function definitions.
But that's just me.

Could you (re)send a context diff rather than plain diff?


        Stefan

Re: perl-mode "::" as word character [patch]

[Pandora]

Stefan Monnier wrote:
> Word syntax is clearly wrong.  Symbol syntax (i.e. "_") OTOH sounds right.

I would be happy to agree.  But \\s_ doens't seem to match anything, and
\\sw doesn't match anything designated as ("\\(::\\)" (1 "_")).  It only
matches "w".  Word syntax may be "wrong" for English, but in perl ':' is
equivalent to A or q or whatever.

> I don't think function calls should be highlighted, only function definitions.
> But that's just me.

I'll go with the precedent, but if it's better without, you can remove
both the 'starting-with-&' and the 'followed-by-a-(' expressions.
They're both pretty equivocably defined as function calls, not function
definitions, in perl.

A more serious concern though, is the foreach construct:
foreach my $var (@list) {...}
That assigns the members of list to variable var one after the other.
My attempt to match 'followed-by-a-(' considers everything left of
(@list) to be a function call.  I'm not good enough with emacs regexps
and font-lock to make this exception, so I'll leave my ...( function
highlighting commented out.
"\\(?:\\<foreach\\>\\|\\<for\\>\\)\\s *\\(\\sw+\\)\\s *(" and... then
how to make that /not/ highlighted, while every other
"\\(\\sw+\\)\\s *(" gets highlighted as a function call?

> Could you (re)send a context diff rather than plain diff?

Um, sure.

Index: lisp/progmodes/perl-mode.el
===================================================================
RCS file: /cvsroot/emacs/emacs/lisp/progmodes/perl-mode.el,v
retrieving revision 1.56
diff -a -u -r1.56 perl-mode.el
--- lisp/progmodes/perl-mode.el 22 Mar 2005 19:43:13 -0000      1.56
+++ lisp/progmodes/perl-mode.el 6 May 2005 01:07:38 -0000
@@ -205,13 +205,18 @@
     ;;
     ;; Fontify local and my keywords as types.
     '("\\<\\(local\\|my\\)\\>" . font-lock-type-face)
+    ;; Fontify print and printf as functions, typically w/o ()
     ;;
+    '("\\<\\(print\\|printf\\)\\>" 1 font-lock-function-name-face)
+
     ;; Fontify function, variable and file name references.
-    '("&\\(\\sw+\\(::\\sw+\\)*\\)" 1 font-lock-function-name-face)
+    '("&\\(\\sw+\\)" 1 font-lock-function-name-face)
+;;   How to handle foreach $var (...) {} ?
+;;    '("\\(\\sw+\\)\\s *(" 1 font-lock-function-name-face)
     ;; Additionally underline non-scalar variables.  Maybe this is a
bad idea.
     ;;'("[$@%*][#{]?\\(\\sw+\\)" 1 font-lock-variable-name-face)
-    '("[$*]{?\\(\\sw+\\(::\\sw+\\)*\\)" 1 font-lock-variable-name-face)
-    '("\\([@%]\\|\\$#\\)\\(\\sw+\\(::\\sw+\\)*\\)"
+    '("[$*]{?\\(\\sw+\\)" 1 font-lock-variable-name-face)
+    '("\\([@%]\\|\\$#\\)\\(\\sw+\\)"
       (2 (cons font-lock-variable-name-face '(underline))))
     '("<\\(\\sw+\\)>" 1 font-lock-constant-face)
     ;;
@@ -256,6 +261,8 @@
     ("\\(\\$\\)[{']" (1 ". p"))
     ;; Handle funny names like $DB'stop.
     ("\\$ ?{?^?[_a-zA-Z][_a-zA-Z0-9]*\\('\\)[_a-zA-Z]" (1 "_"))
+    ;; Or $PKG::member  :: is a word character, really.  c.c
+    ("\\(::\\)" (1 "w"))
     ;; format statements
     ("^[ \t]*format.*=[ \t]*\\(\n\\)" (1 '(7)))
     ;; Funny things in sub arg specifications like `sub myfunc ($$)'

Re: perl-mode "::" as word character [patch]

[Stefan Monnier]

>> Word syntax is clearly wrong.  Symbol syntax (i.e. "_") OTOH sounds right.
> I would be happy to agree.  But \\s_ doesn't seem to match anything, and

It matches any char marked with the _ syntax, which is the syntax to use for
chars which are not word-chars but are allowed as part of symbols.

> \\sw doesn't match anything designated as ("\\(::\\)" (1 "_")).

Of course not.  To match a symbol try "\\(\\sw\\|\\s_)+".

> It only matches "w".  Word syntax may be "wrong" for English, but in perl
> ':' is equivalent to A or q or whatever.

No, it's not the same thing.  Emacs distinguishes words from symbols and
it's quite handy sometimes.

> I'll go with the precedent, but if it's better without, you can remove
> both the 'starting-with-&' and the 'followed-by-a-(' expressions.
> They're both pretty equivocably defined as function calls, not function
> definitions, in perl.

Agreed.

> "\\(?:\\<foreach\\>\\|\\<for\\>\\)\\s *\\(\\sw+\\)\\s *(" and... then
> how to make that /not/ highlighted, while every other
> "\\(\\sw+\\)\\s *(" gets highlighted as a function call?

Aren't there more cases? like

       print HANDLE (arg1, arg2) ?

>> Could you (re)send a context diff rather than plain diff?
> Um, sure.

Thank you.


        Stefan