Partial solution to analyze potentially constant expressions

Partial solution to analyze potentially constant expressions

This change adds code that detects if an expression can definitively
never be constant. We use this to analyze the initializer expressions
in const constructors. This check is not entirely water tight, but
together with the checks in canonicalization code catches most
illegal initializer expressions.

The const constructor of class Symbol turns out to be illegal. The
name verification check can't be part of the constructor code.

R=iposva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=24749

[Ivan Posva, 2013-07-03 20:23:42]

LGTM with comments.

-Ivan

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc
File runtime/vm/ast.cc (right):

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode333
runtime/vm/ast.cc:333: case Token::kADD:
Still legal?

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode345
runtime/vm/ast.cc:345: return true;
Shouldn't the operand also be potentially const?

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode411
runtime/vm/ast.cc:411: return true;
Really? Thanks for the offline explanation, it would be helpful to have it here
as a comment. Even more helpful would be if we could somehow assert that we are
only accessing parameters in the context of const constructor initializer
expressions. Not sure how to do that though.

https://codereview.chromium.org/18649003/diff/1/tests/corelib/corelib.status
File tests/corelib/corelib.status (right):

https://codereview.chromium.org/18649003/diff/1/tests/corelib/corelib.status#...
tests/corelib/corelib.status:12: symbol_test/none: fail # bug 11669
Consistent uppercase Fail and Pass here and below.

[hausner, 2013-07-03 20:47:19]

Committed patchset #2 manually as r24749 (presubmit successful).

[hausner, 2013-07-03 21:20:54]

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc
File runtime/vm/ast.cc (right):

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode333
runtime/vm/ast.cc:333: case Token::kADD:
On 2013/07/03 20:23:42, Ivan Posva wrote:
> Still legal?

Historical artifact that I should have removed with the elimination of unary
plus a couple of weeks ago. Removed.

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode345
runtime/vm/ast.cc:345: return true;
On 2013/07/03 20:23:42, Ivan Posva wrote:
> Shouldn't the operand also be potentially const?

Actually, yes. Fixed.

https://codereview.chromium.org/18649003/diff/1/runtime/vm/ast.cc#newcode411
runtime/vm/ast.cc:411: return true;
On 2013/07/03 20:23:42, Ivan Posva wrote:
> Really? Thanks for the offline explanation, it would be helpful to have it
here
> as a comment. Even more helpful would be if we could somehow assert that we
are
> only accessing parameters in the context of const constructor initializer
> expressions. Not sure how to do that though.

Added comment and checked that the local variable/parameter is final.

https://codereview.chromium.org/18649003/diff/1/tests/corelib/corelib.status
File tests/corelib/corelib.status (right):

https://codereview.chromium.org/18649003/diff/1/tests/corelib/corelib.status#...
tests/corelib/corelib.status:12: symbol_test/none: fail # bug 11669
On 2013/07/03 20:23:42, Ivan Posva wrote:
> Consistent uppercase Fail and Pass here and below.

Done.