Rudimentary analysis

Added rudimentary analysis phase.  Still lots of work to do but now at
least some basics are there.

[Lasse Reichstein, 2008-11-10 16:35:45]

Looks reasonable to me.
I look forward to seeing the analysis being used :)

http://codereview.chromium.org/10254/diff/1/3
File src/jsregexp.cc (right):

http://codereview.chromium.org/10254/diff/1/3#newcode672
Line 672: class SubAnalysisData: public AnalysisData {
How about AnalysisDataBuilder? 
SubAnalysisData doesn't say anything about what it's good for, just that it's a
subclass.

http://codereview.chromium.org/10254/diff/1/3#newcode832
Line 832: visited_(false) { }
Why is it only ChoiceNode that has a visited field?
Other nodes might be the first to be visited twice.

http://codereview.chromium.org/10254/diff/1/3#newcode986
Line 986: fprintf(stderr, "--- %p ---\n", static_cast<void*>(this));
Debug-print?

http://codereview.chromium.org/10254/diff/1/6
File test/cctest/test-regexp.cc (right):

http://codereview.chromium.org/10254/diff/1/6#newcode610
Line 610: Execute("(a|b|c|\\w|\\s)", "", true);
Not *really* a test?

[Christian Plesner Hansen, 2008-11-11 06:12:35]

http://codereview.chromium.org/10254/diff/1/3
File src/jsregexp.cc (right):

http://codereview.chromium.org/10254/diff/1/3#newcode672
Line 672: class SubAnalysisData: public AnalysisData {
The 'Sub' was meant to convey that the result is a "child" of the prev data
given to the constructor.  But builder is a much better name.

http://codereview.chromium.org/10254/diff/1/3#newcode832
Line 832: visited_(false) { }
Yes, but it does no harm if you don't catch it until you meet the next choice
node.  In any case I expect to change the structure so this is just temporary.

http://codereview.chromium.org/10254/diff/1/3#newcode986
Line 986: fprintf(stderr, "--- %p ---\n", static_cast<void*>(this));
It turns out I can use OS::PrintError.  Also, I'll guard all this with #ifdef
DEBUG.

http://codereview.chromium.org/10254/diff/1/6
File test/cctest/test-regexp.cc (right):

http://codereview.chromium.org/10254/diff/1/6#newcode610
Line 610: Execute("(a|b|c|\\w|\\s)", "", true);
True, but damn convenient.