I'm expecting some howls of protest on this patch (if nobody is howling,
nobody is listening.)

The basic justification for adding functionality here is that legacy
apps that steal focus can cause users to tie themselves into Metacity
configuration knots:

 - There is some legacy app that is stealing focus away from
   my terminal: let me use focus_new_windows=strict

 - Now when I launch windows from the terminal, they pop behind and
   flash in the taskbar. I need more config options to fix this!

So the idea here is to short-circuit the chain at the beginning and
provide a means for site-adminstrators to fix troublesome applications
without having to configure and hack-out all the focus-stealing-prevention
goodness.

So, the second second of howls will likely be about the way I did it -
arbitrary boolean expressions? s-exps? 400 lines of parsing code? So,
let me provide some background to my thinking here.

 * Since we're dealing by hypothesis with a legacy app that can't
   be modified, we need a fair bit of matching power - we can't just
   expect the app to be modified to export a suitable WM_ROLE; this
   is why I wanted to allow matching on boolean combinations of WM_NAME
   and WM_CLASS.

 * I decided against some g_strsplit()-parsable syntax of
   'class:Mylegacyapp,name:New mail;...' or something because it was going
   to be a) horrible and b) not extensible.

 * I went with s-exps for the standard reason: I wanted something that
   was expressive without having to invent a syntax and provide a
   yacc grammar or complicated hand-built parser.

 * The use of globs rather than regular expressions for pattern matching
   has no real justification other than I wanted a patch I could backport
   to glib-2.12, which doesn't have GRegex. The syntax is extensible to
   allow regular expressions to be added with a few more lines, if there
   was really a reason.

 * Those familiar devilspie may notice that the expressions are incompatible
   with the sxpressions it uses - where devilspie has:

    (is (window-class) "Mylegacyapp")

   I have:

    (eq class "Mylegacyapp")

   Basically the reasoning was that I couldn't be 100% compatible with
   devilspie (needed boolean operators, couldn't use regular expressions),
   so I decided to just do what made sense to me and kept the parsing
   code simple.

 * So why not just make this a devilspie functionality? Make devilspie
   set _NET_WM_USER_TIME? Doesn't work - because we are dealing with new
   windows and trying to change their props out of devilspie would be
   inherently racy.

 * The patch isn't quite as big as it looks - about a third of the code
   in window-matcher.c is unit tests that are #ifdef'ed out.

Clearly this patch isn't in "the spirit of Metacity", but the basic idea
is, as described above, to workaround broken apps and keep focus handling
working as designed.