Experimental: fix a bug in our compilation of switch statements....

Experimental: fix a bug in our compilation of switch statements.

As described in issue 241

http://code.google.com/p/v8/issues/detail?id=241

we can crash when compiling switch statements if we get into a state
where we are skipping comparisons (due to an unconditionally true
match) and still compiling bodies (due to continued fall-through) and
then hit a default case.

The issue is fixed by splitting the loop over the cases so that we do
not reuse the same loop code for both states (compiling comparisons
and skipping comparisons).

The code for the end of the switch has also been modified to avoid a
jump to the next statement for the last case (at the cost of extra
complexity to handle the new possible compilation states that arise).

Committed: http://code.google.com/p/v8/source/detail?r=1379

[William Hesse, 2009-02-26 14:09:06]

LGTM.

http://codereview.chromium.org/27200/diff/16/1012
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/27200/diff/16/1012#newcode2168
Line 2168: 
Could you put a comment here, indicating the possible states of (1) fallthrough
links to and from the default clause
(2) fallthrough link to remaining cases, that are only reached by fallthrough
(3) next_test link to the default clause, as bound fallthrough or as linked
target.

http://codereview.chromium.org/27200/diff/16/1012#newcode2174
Line 2174: VisitStatements(cases->at(index)->statements());
Should you break on no frame here, or is VisitStatements guarded by a
sufficiently fast check for a valid frame?

http://codereview.chromium.org/27200/diff/16/1012#newcode2180
Line 2180: // The last clause compiled was unconditionally true.  We still
The last test compiled?
Can't many fall-through clauses be compiled after this test?

[Kevin Millikin (Chromium), 2009-02-26 15:00:04]

http://codereview.chromium.org/27200/diff/16/1012
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/27200/diff/16/1012#newcode2168
Line 2168: 
On 2009/02/26 14:09:06, William Hesse wrote:
> Could you put a comment here, indicating the possible states of (1)
fallthrough
> links to and from the default clause
> (2) fallthrough link to remaining cases, that are only reached by fallthrough
> (3) next_test link to the default clause, as bound fallthrough or as linked
> target.

Not here but I've clarified some of the other comments.

http://codereview.chromium.org/27200/diff/16/1012#newcode2174
Line 2174: VisitStatements(cases->at(index)->statements());
On 2009/02/26 14:09:06, William Hesse wrote:
> Should you break on no frame here, or is VisitStatements guarded by a
> sufficiently fast check for a valid frame?

Done.

http://codereview.chromium.org/27200/diff/16/1012#newcode2180
Line 2180: // The last clause compiled was unconditionally true.  We still
On 2009/02/26 14:09:06, William Hesse wrote:
> The last test compiled?
> Can't many fall-through clauses be compiled after this test?

Done.