* Match literals in a case independent way....

* Match literals in a case independent way.
* When matching constant length sequences, do case independent
  characters first, then case dependent characters, then character
  classes

Committed: http://code.google.com/p/v8/source/detail?r=827

[plesner_quenta.org, 2008-11-24 09:49:20]

Lgtm

http://codereview.chromium.org/11352/diff/1/9
File src/jsregexp.cc (right):

http://codereview.chromium.org/11352/diff/1/9#newcode858
Line 858: inline bool case_independent() { return case_independent_; }
This name is slightly confusing.  Maybe is_case_independent or ignore_case would
be clearer.

http://codereview.chromium.org/11352/diff/1/9#newcode887
Line 887: case_independent_ = case_independent;
Shouldn't this be set by the constructor?

http://codereview.chromium.org/11352/diff/1/9#newcode1082
Line 1082: Vector<const uc16> quarks,
lol

http://codereview.chromium.org/11352/diff/1/9#newcode1110
Line 1110: ASSERT(unibrow::Ecma262UnCanonicalize::kMaxWidth == 4);
ASSERT_EQ

http://codereview.chromium.org/11352/diff/1/9#newcode1116
Line 1116: if (((exor - 1) & exor) == 0) {
Maybe a comment that says that this checks that exactly 1 bit is set in exor.

http://codereview.chromium.org/11352/diff/1/9#newcode1118
Line 1118: uc16 upper = (c1 | c2);
The results of uncanonicalized are always given in sorted order so upper should
always equal c2.  Consider replacing this with an assertion that c2 > c1 and
then just use c2.

http://codereview.chromium.org/11352/diff/1/9#newcode1130
Line 1130: // FALL THROUGH!
Why are you yelling?

http://codereview.chromium.org/11352/diff/1/9#newcode1132
Line 1132: macro_assembler->CheckCharacter(chars[0], &ok);
We could consider using the OrThenCheckNotCharacter trick for pairs of
characters in this and the previous case.

http://codereview.chromium.org/11352/diff/1/9#newcode1166
Line 1166: CharacterRange& range = (*ranges)[i];
ranges->at(i) also works.

http://codereview.chromium.org/11352/diff/1/9#newcode1185
Line 1185: if (range_count != 0) {
You return earlier if this is not the case.

http://codereview.chromium.org/11352/diff/1/9#newcode1198
Line 1198: if (!cc->is_negated()) {
Is there a reason to not use char_is_in_class here?

[Erik Corry, 2008-11-24 13:16:51]

Having looked at the non-ascii case independent pairs I found we needed another
macro assembler instruction to cover a good chunk of the remaining equivalence
classes.  This version includes the new instruction.

http://codereview.chromium.org/11352/diff/1/9
File src/jsregexp.cc (right):

http://codereview.chromium.org/11352/diff/1/9#newcode1132
Line 1132: macro_assembler->CheckCharacter(chars[0], &ok);
On 2008/11/24 09:49:21, plesner wrote:
> We could consider using the OrThenCheckNotCharacter trick for pairs of
> characters in this and the previous case.

There are 837 set of canonical characters in the current unicode spec.  With the
'or' trick we cover over 500, leaving 319.  If we add a subtract-then-or trick
we cover another 90, leaving 229.  Of these only 15 are triplets and one is a
quad.  I don't think it's worth spending much time on such rare cases.

http://codereview.chromium.org/11352/diff/1/9#newcode1198
Line 1198: if (!cc->is_negated()) {
On 2008/11/24 09:49:21, plesner wrote:
> Is there a reason to not use char_is_in_class here?

Yes, the sense is reversed.