Introduce getter p on double in the VM to control fractional precision.

Introduce getter p on double in the VM to control fractional precision.
Add test.

R=asiva@google.com, fschneider@google.com

Committed: https://github.com/dart-lang/sdk/commit/fbbedf4f184f3815afc057e41a3404858d005e10

[regis, 2016-08-03 22:30:41]

regis@google.com changed reviewers:
+ asiva@google.com, fschneider@google.com

[Florian Schneider, 2016-08-03 23:35:58]

Can you add a test case where e.g. a, b are at the precision limit, (and maybe
just above)?

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
File tests/standalone/fixed_precision_double_test.dart (right):

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:7: bool TestAssociativity(bool
chop) {
For testing here maybe pass in a function chop (or identity) in here as a
parameter f to avoid duplication of the literals/expressions:


bool TestAssociativity(Function f) {
  double a = f(1234.567);
...

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:35:
Expect.isFalse(TestAssociativity(false));
e.g. Expect.isFalse(TestAssociativity((x) => x));

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:38:
Expect.isTrue(TestAssociativity(true));
e.g. Expect.isTrue(TestAssociativity((x) => x.p));

[regis, 2016-08-04 18:22:31]

Thanks! PTAL

Since this feature will be used with near-integer values (e.g. pixels), I did
not add tests that would succeed or fail around the precision limit. Also, these
tests could break in case we change the number of bits kept for the fractional
part.

However, I added tests for simple literals, as well as Nan and infinity.

Thanks,
Regis

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
File tests/standalone/fixed_precision_double_test.dart (right):

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:7: bool TestAssociativity(bool
chop) {
On 2016/08/03 23:35:58, Florian Schneider wrote:
> For testing here maybe pass in a function chop (or identity) in here as a
> parameter f to avoid duplication of the literals/expressions:
> 
> 
> bool TestAssociativity(Function f) {
>   double a = f(1234.567);
> ...
> 
> 

Done.

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:35:
Expect.isFalse(TestAssociativity(false));
On 2016/08/03 23:35:58, Florian Schneider wrote:
> e.g. Expect.isFalse(TestAssociativity((x) => x));

Done.

https://codereview.chromium.org/2208473006/diff/1/tests/standalone/fixed_prec...
tests/standalone/fixed_precision_double_test.dart:38:
Expect.isTrue(TestAssociativity(true));
On 2016/08/03 23:35:58, Florian Schneider wrote:
> e.g. Expect.isTrue(TestAssociativity((x) => x.p));

Done.

[siva, 2016-08-04 20:20:13]

LGTM with one comment.

https://codereview.chromium.org/2208473006/diff/20001/tests/standalone/fixed_...
File tests/standalone/fixed_precision_double_test.dart (right):

https://codereview.chromium.org/2208473006/diff/20001/tests/standalone/fixed_...
tests/standalone/fixed_precision_double_test.dart:37: Expect.notEquals(1/3,
(1/3).p);
What is the significance of these tests , how does one know when a value and
value.p should be equal?


Also can we add a test which asserts that a value has been normalized i.e
simulating the asserts they would add to their code before doing the operation.
Something like
verify(double d) {
  assert(d == d.p);
}

verify(1.1.p); // would pass.
verify(1.1);  // would fail.

[Florian Schneider, 2016-08-04 20:32:50]

Lgtm.

[regis, 2016-08-04 22:16:12]

I have added tests showing the use of assert(d == d.p).

Thank you both!

https://codereview.chromium.org/2208473006/diff/20001/tests/standalone/fixed_...
File tests/standalone/fixed_precision_double_test.dart (right):

https://codereview.chromium.org/2208473006/diff/20001/tests/standalone/fixed_...
tests/standalone/fixed_precision_double_test.dart:37: Expect.notEquals(1/3,
(1/3).p);
On 2016/08/04 20:20:13, siva wrote:
> What is the significance of these tests , how does one know when a value and
> value.p should be equal?
> 
> 
> Also can we add a test which asserts that a value has been normalized i.e
> simulating the asserts they would add to their code before doing the
operation.
> Something like
> verify(double d) {
>   assert(d == d.p);
> }
> 
> verify(1.1.p); // would pass.
> verify(1.1);  // would fail.

I have added comments explaining this.

[regis, 2016-08-04 22:18:06]

Description was changed from

==========
Introduce getter p on double in the VM to control fractional precision.
Add test.
==========

to

==========
Introduce getter p on double in the VM to control fractional precision.
Add test.

R=asiva@google.com, fschneider@google.com

Committed:
https://github.com/dart-lang/sdk/commit/fbbedf4f184f3815afc057e41a3404858d005e10
==========

[regis, 2016-08-04 22:18:07]

Committed patchset #3 (id:40001) manually as
fbbedf4f184f3815afc057e41a3404858d005e10 (presubmit successful).

[floitsch, 2016-08-25 12:57:39]

floitsch@google.com changed reviewers:
+ floitsch@google.com

[floitsch, 2016-08-25 12:57:39]

When you change dart core libraries or the language please talk to our team
first and CC us on all the changes.

When you introduce experiments to the language or libraries please guard them
behind a flag.

What is the deprecation strategy for this feature?
How do you intend to remove it again without affecting users that accidentally
discovered the member?

[kevmoo, 2016-08-25 17:56:20]

On 2016/08/25 12:57:39, floitsch wrote:
> When you change dart core libraries or the language please talk to our team
> first and CC us on all the changes.
> 
> When you introduce experiments to the language or libraries please guard them
> behind a flag.
> 
> What is the deprecation strategy for this feature?
> How do you intend to remove it again without affecting users that accidentally
> discovered the member?

What he said...