Create a new RegExp object for every evaluation of a RegExp literal.

Create a new RegExp object for every evaluation of a RegExp literal.
Changes necessary to following ES5 semantics and matching Safari - in ES3
the same RegExp object was generated by each evaluation of the RegExp literal.
Fixes bug 704.

[Lasse Reichstein, 2010-08-06 09:16:09]

@kmillikin, please check the virtual frame usage in ia32 and x64 for any
misunderstandings.

[Rodolph Perfetta, 2010-08-06 10:19:13]

http://codereview.chromium.org/3034060/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/2#newcode5356
src/arm/codegen-arm.cc:5356: __ tst(tmp, Operand(kSmiTagMask));
the eor and tst could be merged in a teq:
teq(tmp, Operand(kSmiTagMask));

[Lasse Reichstein, 2010-08-06 10:32:28]

http://codereview.chromium.org/3034060/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/2#newcode5356
src/arm/codegen-arm.cc:5356: __ tst(tmp, Operand(kSmiTagMask));
If I understand it corretly, the eor/tst only tests one bit if the result after
eor, whereas teq tests the entire word for being zero after eor. It wouldn't
give the same result if there are other bits set in the result of the and.

[Rodolph Perfetta, 2010-08-06 11:01:23]

http://codereview.chromium.org/3034060/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/2#newcode5356
src/arm/codegen-arm.cc:5356: __ tst(tmp, Operand(kSmiTagMask));
You are right, I went a bit quickly over, and only checked to SmiTagMask. Sorry.

[Erik Corry, 2010-08-06 11:16:28]

LGTM but we need a test that feeds Smis into the regexp cache routines (test
equivalent).

http://codereview.chromium.org/3034060/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/2#newcode3312
src/arm/codegen-arm.cc:3312: literal = frame_->PopToRegister(r0);
I find it confusing that the literal variable is reused here for a new register.
 Can we either call it something else or add a comment to say that we are
reusing the variable?

http://codereview.chromium.org/3034060/diff/1/2#newcode3316
src/arm/codegen-arm.cc:3316: __ str(tmp, FieldMemOperand(r0, i));
This is the same code as in FastCloneShallowArrayStub.  We should move this sort
of memcpy with fixed size into the macro assembler.  There we could also make it
more efficient.  With 2 scratch registers and ip we could do it with 3-register
ldm and stm which is more compact and on some chips faster.

http://codereview.chromium.org/3034060/diff/1/2#newcode5347
src/arm/codegen-arm.cc:5347: Register tmp = r2;
Please use the scratch registers provided by the virtual frame.

http://codereview.chromium.org/3034060/diff/1/2#newcode5350
src/arm/codegen-arm.cc:5350: Label done;
Please add a comment to the effect that the state of the CC flags at this label
is significant.

http://codereview.chromium.org/3034060/diff/1/4
File src/arm/full-codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/4#newcode1140
src/arm/full-codegen-arm.cc:1140: __ str(r2, FieldMemOperand(r0, i));
And here

http://codereview.chromium.org/3034060/diff/1/14
File src/runtime.cc (right):

http://codereview.chromium.org/3034060/diff/1/14#newcode7600
src/runtime.cc:7600: RUNTIME_ASSERT(size <= Heap::MaxObjectSizeInNewSpace());
size actually needs to be smaller than the minimum free space in new space after
a GC, which is less thatn the MaxObjectSizeInNewSpace.

http://codereview.chromium.org/3034060/diff/1/16
File src/string.js (right):

http://codereview.chromium.org/3034060/diff/1/16#newcode625
src/string.js:625: // Reuse lastIndex field for split limit when type is
"split".
This was wrong before, huh?

[Lasse Reichstein, 2010-08-06 13:03:10]

http://codereview.chromium.org/3034060/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/2#newcode3312
src/arm/codegen-arm.cc:3312: literal = frame_->PopToRegister(r0);
Comment added.

http://codereview.chromium.org/3034060/diff/1/2#newcode3316
src/arm/codegen-arm.cc:3316: __ str(tmp, FieldMemOperand(r0, i));
Macro added. For now it just does the same, but with a larger overhead.

http://codereview.chromium.org/3034060/diff/1/2#newcode5347
src/arm/codegen-arm.cc:5347: Register tmp = r2;
Changed to scratch0/scratch1.

http://codereview.chromium.org/3034060/diff/1/2#newcode5350
src/arm/codegen-arm.cc:5350: Label done;
Added comment

http://codereview.chromium.org/3034060/diff/1/4
File src/arm/full-codegen-arm.cc (right):

http://codereview.chromium.org/3034060/diff/1/4#newcode1140
src/arm/full-codegen-arm.cc:1140: __ str(r2, FieldMemOperand(r0, i));
Too.

http://codereview.chromium.org/3034060/diff/1/14
File src/runtime.cc (right):

http://codereview.chromium.org/3034060/diff/1/14#newcode7600
src/runtime.cc:7600: RUNTIME_ASSERT(size <= Heap::MaxObjectSizeInNewSpace());
Added constant for Heap::InitialSemiSpaceSize() * 3/4

http://codereview.chromium.org/3034060/diff/1/16
File src/string.js (right):

http://codereview.chromium.org/3034060/diff/1/16#newcode625
src/string.js:625: // Reuse lastIndex field for split limit when type is
"split".
Apparently we didn't hit it.