* No failures on our own tests....

* No failures on our own tests.
* 26 failures on Mozilla tests.
* Remember to include linebreaks in \s
* Interpreter takes flat 16 bit strings as input.
* Remove dubious test from mjsunit/regexp.js
  (http://code.google.com/p/v8/issues/detail?id=152)
* Add debugging help (off by default) to unicode-test.js
* The regexp-macro-assembler interface now has the concept
  of a current_character register.
* Removed CheckCharacterClass from regexp-macro-assembler
  (too high level an operation for this level).
* Introduce CheckCharacterLT and CheckCharacterGT to the
  macro assembler interface.
* Make the re2k assembler use a growable instruction buffer
  to eliminate an arbitrary size limit.
* Add --trace-regexp-bytecodes option to debug build.
* Make RegExpNode::GoTo virtual so the backtrack node can
  just inline itself.
* Add protected RegExpNode::Bind() that subclasses use when
  emitting their code.
* Limit max recursion in Emit stage to avoid stack overflow.
* Remember to reserve at least 2 registers for 0th capture.
* Bail out to JSCRE when encountering \b, ^, $.
* Fix code emission and implement guards on ChoiceNode.
  (Still doesn't use dispatch table).
* Implement code emission for TextNode.
* Remember to set up backtrack when writing capture indeces
  to capture registers so they can be unwound if neccessary.
* DispatchTableConstructor::VisitBackreference isn't yet
  implemented, but we don't crash the VM.  (Later we
  discover the regexp has backreferences and defer to jscre).
* \b in a character class means backspace.


Committed: http://code.google.com/p/v8/source/detail?r=790

[Christian Plesner Hansen, 2008-11-18 14:07:41]

Lgtm.  Comments are mostly nitpicking.

http://codereview.chromium.org/11228/diff/1/9
File src/bytecodes-re2k.h (right):

http://codereview.chromium.org/11228/diff/1/9#newcode67
Line 67: #ifdef DEBUG
Does this have to be #ifdef DEBUG?  Unused static consts have no effect anyway
right?

http://codereview.chromium.org/11228/diff/1/13
File src/interpreter-re2k.cc (right):

http://codereview.chromium.org/11228/diff/1/13#newcode42
Line 42: # define BYTECODE(name) break;                                         
       \
I would prefer if the breaks were written explicitly everywhere.  This is too
subtle for my taste.  But feel free to disagree and not change it.

http://codereview.chromium.org/11228/diff/1/13#newcode44
Line 44: if (FLAG_trace_regexp_bytecodes) {                   \
Couldn't this be factored into a function rather than duplicated?

http://codereview.chromium.org/11228/diff/1/13#newcode169
Line 169: pc += 7;
I would suggest using constants for opcode length so the literal value occurs
exactly one place.  I've previously used the approach of packaging all the
information about a bytecode set into constants on a class so you could do
OpcodeInfo<CHECK_GT>::kLength, OpcodeInfo<CHECH_GT>::kArgCount, etc., so you
would always just use pc += OpcodeInfo<[current opcode]>::kLength.  That may be
too spacy for our use though.

http://codereview.chromium.org/11228/diff/1/19
File src/regexp-macro-assembler-re2k.h (right):

http://codereview.chromium.org/11228/diff/1/19#newcode57
Line 57: uc16 limit,
These should fit on one line.

http://codereview.chromium.org/11228/diff/1/4
File test/mjsunit/unicode-test.js (right):

http://codereview.chromium.org/11228/diff/1/4#newcode9132
Line 9132: //dump_re(re);
?