NonGNU-devel ELPA - pcre2el

When used interactively, the conversion commands can read a regexp either from the current buffer or from the minibuffer. The output is displayed in the minibuffer and copied to the kill-ring.

• When called with a prefix argument (C-u), they read a regular expression from the minibuffer literally, without further processing – meaning there's no need to double the backslashes if it's an Emacs regexp. This is the same way commands like query-replace-regexp read input.
• When the region is active, they use they the region contents, again literally (without any translation of string syntax).

• With neither a prefix arg nor an active region, the behavior depends on whether the command expects an Emacs regexp or a PCRE one.

  Commands that take an Emacs regexp behave like C-x C-e: they evaluate the sexp before point (which could be simply a string literal) and use its value. This is designed for use in Elisp buffers. As a special case, if point is inside a string, it's first moved to the string end, so in practice they should work as long as point is somewhere within the regexp literal.

  Commands that take a PCRE regexp try to read a Perl-style delimited regex literal after point in the current buffer, including its flags. For example, putting point before the m in the following example and doing C-c / p e (rxt-pcre-to-elisp) displays \(?:bar\|foo\), correctly stripping out the whitespace and comment:

  $x =~ m/  foo   |  (?# comment) bar /x
  

  The PCRE reader currently only works with / ... / delimiters. It will ignore any preceding m, s, or qr operator, as well as the replacement part of an s construction.

  Readers for other PCRE-using languages are on the TODO list.

The translation functions display their result in the minibuffer and copy it to the kill ring. When translating something into Elisp syntax, you might need to use the result either literally (e.g. for interactive input to a command like query-replace-regexp), or as a string to paste into Lisp code. To allow both uses, rxt-pcre-to-elisp copies both versions successively to the kill-ring. The literal regexp without string quoting is the top element of the kill-ring, while the Lisp string is the second-from-top. You can paste the literal regexp somewhere by doing C-y, or the Lisp string by C-y M-y.

rxt-convert-syntax (C-c / c) converts between Emacs and PCRE syntax, depending on the major mode in effect when called. Alternatively, you can specify the conversion direction explicitly by using either rxt-pcre-to-elisp (C-c / p e) or rxt-elisp-to-pcre (C-c / e p).

Similarly, rxt-convert-to-rx (C-c / x) converts either kind of syntax to rx form, while rxt-convert-pcre-to-rx (C-c / p x) and rxt-convert-elisp-to-rx (C-c / e x) convert to rx from a specified source type.

In Elisp buffers, you can use rxt-toggle-elisp-rx (C-c / t or C-c / e t) to switch the regexp at point back and forth between string and rx syntax. Point should either be within an rx or rx-to-string form or a string literal for this to work.

If you want to use emulated PCRE regexp syntax in all Emacs commands, try pcre-mode, which uses Emacs's advice system to make all commands that read regexps using the minibuffer use emulated PCRE syntax. It should also work with Isearch.

This feature is still fairly experimental. It may fail to work or do the wrong thing with certain commands. Please report bugs.

pcre-query-replace-regexp was originally defined to do query-replace using emulated PCRE regexps, and is now made somewhat obsolete by pcre-mode. It is bound to C-c / % by default, by analogy with M-%. Put the following in your .emacs if you want to use PCRE-style query replacement everywhere:

(global-set-key [(meta %)] 'pcre-query-replace-regexp)

When syntax-highlighting isn't enough to untangle some gnarly regexp you find in the wild, try the 'explain' commands: rxt-explain (C-c / /), rxt-explain-pcre (C-c / p) and rxt-explain-elisp (C-c / e). These display the original regexp along with its pretty-printed rx equivalent in a new buffer. Moving point around either in the original regexp or the rx translation highlights corresponding pieces of syntax, which can aid in seeing things like the scope of quantifiers.

I call them "explain" commands because the rx form is close to a plain syntax tree, and this plus the wordiness of the operators usually helps to clarify what is going on. People who dislike Lisp syntax might disagree with this assessment.

Occasionally you come across a regexp which is designed to match a finite set of strings, e.g. a set of keywords, and it would be useful to recover the original set. (In Emacs you can generate such regexps using regexp-opt). The commands rxt-convert-to-strings (C-c / ′), rxt-pcre-to-strings (C-c / p ′) or rxt-elisp-to-strings (C-c / e ′) accomplish this by generating all the matching strings ("productions") of a regexp. (The productions are copied to the kill ring as a Lisp list).

An example in Lisp code:

(regexp-opt '("cat" "caterpillar" "catatonic"))
   ;; => "\\(?:cat\\(?:atonic\\|erpillar\\)?\\)"
(rxt-elisp-to-strings "\\(?:cat\\(?:atonic\\|erpillar\\)?\\)")
    ;; => '("cat" "caterpillar" "catatonic")

For obvious reasons, these commands only work with regexps that don't include any unbounded quantifiers like + or *. They also can't enumerate all the characters that match a named character class like [[:alnum:]]. In either case they will give a (hopefully meaningful) error message. Due to the nature of permutations, it's still possible for a finite regexp to generate a huge number of productions, which will eat memory and slow down your Emacs. Be ready with C-g if necessary.

The Emacs RE-Builder is a useful visual tool which allows using several different built-in syntaxes via reb-change-syntax (C-c TAB). It supports Elisp read and literal syntax and rx, but it can only convert from the symbolic forms to Elisp, not the other way. This package hacks the RE-Builder to also work with emulated PCRE syntax, and to convert transparently between Elisp, PCRE and rx syntaxes. PCRE mode reads a delimited Perl-like literal of the form / ... /, and it should correctly support using the x and s flags.

Example of using the conversion functions:

(rxt-pcre-to-elisp "(abc|def)\\w+\\d+")
   ;; => "\\(\\(?:abc\\|def\\)\\)[_[:alnum:]]+[[:digit:]]+"

All the conversion functions take a single string argument, the regexp to translate:

• rxt-pcre-to-elisp
• rxt-pcre-to-rx
• rxt-pcre-to-sre
• rxt-pcre-to-strings
• rxt-elisp-to-pcre
• rxt-elisp-to-rx
• rxt-elisp-to-sre
• rxt-elisp-to-strings

PCRE has a complicated syntax and semantics, only some of which can be translated into Elisp. The following subset of PCRE should be correctly parsed and converted:

• parenthesis grouping ( .. ), including shy matches (?: ... )
• backreferences (various syntaxes), but only up to 9 per expression
• alternation |
• greedy and non-greedy quantifiers *, *?, +, +?, ? and ?? (all of which are the same in Elisp as in PCRE)
• numerical quantifiers {M,N}
• beginning/end of string \A, \Z
• string quoting \Q .. \E
• word boundaries \b, \B (these are the same in Elisp)
• single character escapes \a, \c, \e, \f, \n, \r, \t, \x, and \octal digits (but see below about non-ASCII characters)
• character classes [...] including Posix escapes
• character classes \d, \D, \h, \H, \s, \S, \v, \V both within character class brackets and outside
• word and non-word characters \w and \W (Emacs has the same syntax, but its meaning is different)
• s (single line) and x (extended syntax) flags, in regexp literals, or set within the expression via (?xs-xs) or (?xs-xs: .... ) syntax
• comments (?# ... )

Most of the more esoteric PCRE features can't really be supported by simple translation to Elisp regexps. These include the different lookaround assertions, conditionals, and the "backtracking control verbs" (* ...) . OTOH, there are a few other syntaxes which are currently unsupported and possibly could be:

• \L, \U, \l, \u case modifiers
• \g{...} backreferences

• The order of alternatives and characters in char classes sometimes gets shifted around, which is annoying.

• Although the string parser tries to interpret PCRE's octal and hexadecimal escapes correctly, there are problems with matching 8-bit characters that I don't use enough to properly understand, e.g.:

  (string-match-p (rxt-pcre-to-elisp "\\377") "\377") => nil
  

  A fix for this would be welcome.

• Most of PCRE's rules for how ^, \A, $ and \Z interact with newlines are not implemented, since they seem less relevant to Emacs's buffer-oriented rather than line-oriented model. However, the different meanings of the . metacharacter are implemented (it matches newlines with the /s flag, but not otherwise).
• Not currently namespace clean (both rxt- and a couple of pcre- functions).

• Python-specific extensions to PCRE?
• Language-specific stuff to enable regexp font-locking and explaining in different modes. Each language would need two functions, which could be kept in an alist:
  1. A function to read PCRE regexps, taking the string syntax into account. E.g., Python has single-quoted, double-quoted and raw strings, each with different quoting rules. PHP has the kind of belt-and-suspenders solution you would expect: regexps are in strings, and you have to include the / ... / delimiters! Duh.
  2. A function to copy faces back from the parsed string to the original buffer text. This has to recognize any escape sequences so they can be treated as a single character.