Handle field-invocation and method-extraction in the runtime when --lazy_dispatchers=false.

Handle field-invocation and method-extraction in the runtime when --lazy_dispatchers=false.

R=fschneider@google.com

Committed: https://github.com/dart-lang/sdk/commit/40fcf4a72debd2e4bed1f7a56a9a6837d5e7d827

[rmacnak, 2015-06-19 17:02:48]

rmacnak@google.com changed reviewers:
+ fschneider@google.com, srdjan@google.com

[rmacnak, 2015-06-19 17:02:49]

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
File runtime/vm/code_generator.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
runtime/vm/code_generator.cc:1080: // would perform the closure call.
In fact, every closure call goes through the runtime :(

https://codereview.chromium.org/1195573006/diff/40001/tests/language/closure_...
File tests/language/closure_self_reference_test.dart (right):

https://codereview.chromium.org/1195573006/diff/40001/tests/language/closure_...
tests/language/closure_self_reference_test.dart:22: Expect.equals(199, counter);
Entering the runtime chews up a lot of stack space.

[Florian Schneider, 2015-06-29 10:37:53]

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
File runtime/vm/code_generator.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
runtime/vm/code_generator.cc:1080: // would perform the closure call.
On 2015/06/19 17:02:49, rmacnak wrote:
> In fact, every closure call goes through the runtime :(

Yes, that's a bit unfortunate. Let's discuss how to improve on this.

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.cc
File runtime/vm/dart_entry.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:160: } else {
When is InvokeClosure used with a non-closure object?

[rmacnak, 2015-06-29 21:54:19]

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
File runtime/vm/code_generator.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/code_generat...
runtime/vm/code_generator.cc:1080: // would perform the closure call.
On 2015/06/29 10:37:53, Florian Schneider wrote:
> On 2015/06/19 17:02:49, rmacnak wrote:
> > In fact, every closure call goes through the runtime :(
> 
> Yes, that's a bit unfortunate. Let's discuss how to improve on this.

It may be enough to just emit the field-invocation methods for all signature
classes in precompilation.

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.cc
File runtime/vm/dart_entry.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:160: } else {
On 2015/06/29 10:37:53, Florian Schneider wrote:
> When is InvokeClosure used with a non-closure object?

From the runtime stub, e.g.

class FakeFunction {
  call() => ...
}
class A {
  get f => new FakeFunction();
}
new A().f()

Or if someone invokes Dart_InvokeClosure with a FakeFunction.

Maybe this should be renamed InvokeCall.

[Florian Schneider, 2015-06-30 10:28:35]

lgtm

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.cc
File runtime/vm/dart_entry.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:190: arguments.SetAt(0, getter_result);  // Safe to
reuse this?
Why shouldn't it be safe?

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:197: // No compataible method or getter so invoke
noSuchMethod.
s/compataible/compatible/

[rmacnak, 2015-06-30 22:13:49]

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.cc
File runtime/vm/dart_entry.cc (right):

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:190: arguments.SetAt(0, getter_result);  // Safe to
reuse this?
On 2015/06/30 10:28:35, Florian Schneider wrote:
> Why shouldn't it be safe?

I didn't initially notice a new one was allocated for each runtime entry.

https://codereview.chromium.org/1195573006/diff/40001/runtime/vm/dart_entry.c...
runtime/vm/dart_entry.cc:197: // No compataible method or getter so invoke
noSuchMethod.
On 2015/06/30 10:28:35, Florian Schneider wrote:
> s/compataible/compatible/

Done.

[rmacnak, 2015-06-30 22:14:41]

Committed patchset #4 (id:60001) manually as
40fcf4a72debd2e4bed1f7a56a9a6837d5e7d827 (presubmit successful).