[PATCH] better implementation for __HALT_COMPILER();

Hi,

With the first release of pecl/phar, I have started thinking about the
implementation of __HALT_COMPILER();

Currently, this code results in a fatal error:

file1.php:
<?php
__HALT_COMPILER();
?>

file2.php:
<?php
__HALT_COMPILER();
?>

file3.php:
<?php
include 'file1.php';
include 'file2.php'; // fatal error - can't declare __HALT_COMPILER();
more than once in the same request
?>

The current implementation simply defines COMPILER_HALT_OFFSET as
the first __HALT_COMPILER();, independent of the file.

However, it occurred to me that it might be easy to fix this by
introducing a hash of halt offsets by file. After talking to Sara on
IRC, she proved she is truly an internals goddess by suggesting instead
to use a mangled constant name (COMPILER_HALT_OFFSET + filename) and
use the existing global constants hash instead. So, with that helpful
tip, I whipped up a patch to provide file-specific
COMPILER_HALT_OFFSET. This does change behavior in that
COMPILER_HALT_OFFSET can only be used within the same file that it
is defined. However, this is extremely unlikely to break any existing
scripts since it was designed for self-contained installers, and to my
knowledge, Phar/PHP_Archive and some auxiliary thing called PHK are the
only things using it. The benefits to users of phar-like things would
be immense, with no performance degradation for anything except for the
worst case constant lookup where a user requests the value of an
undefined constant (UPPERCASE constant doesn't exist, lowercase constant
doesn't exist).

Anyways, the patch is very short, so I've attached it. This patch is
against PHP_5_2 (since I consider it a bug fix) and I will happily whip
up a patch against HEAD if it is acceptable. My assumption here, of
course, is that this would go in 5.2.2 at the earliest since Ilia was
waiting for a CVS lock to release in order to put up 5.2.1 while I was
coding the patch.

Thanks,
Greg