fix(ngModelController): always use last commited view value in validators

[Narretz]

...ed

This fixes issues where a parser calls $setViewValue. This is a common
strategy for manipulating the $viewValue while the user is entering
data into an input field.

When the $viewValue was changed inside the parser, the new viewValue
would be committed and used for validation. The original parser
however would run last and pass the original (outdated) viewValue on
to the validators, which could cause false positives, e.g. for
minlength.

This is a regression since 1.3.0-rc.1, caused by 6046e14#diff-c244afd8def7f268b16ee91a0341c4b2L2055

[googlebot]

CLAs look good, thanks!

[Narretz]

@shahata Could you take a look at this?

[shahata]

@Narretz I'm not sure about this. Validators get both modelValue and viewValue as parameters, so I could argue that even if modelValue didn't change, it should still be called again since the viewValue is not the same.

Also, I'm not sure I understand why a parser would call $setViewValue. I would do something like ctrl.$viewValue = value; ctrl.$render();. Can you show an example of where this behavior is causing trouble?

[Narretz]

One problem is in this issue: #10126
Enter a, in the input. , gets stripped out, but the minlength validator says true which is not correct.

The minlength validator will validate with a viewValue that is out of date, because parser -> (setViewValue -> parser -> validate) -> validate and because $$parseAndValidate caches the $$lastCommittedViewValue.

Setting $viewValue directly does not fix this.

[shahata]

Okay, I get it now, but let's say I have a parser that just removes the last , (without touching $viewValue and $setViewValue) and I also have maxlength="1" - wouldn't your implementation do the wrong thing when typing a and then adding a , (since a and a, have the same model value, validators will not run and won't detect that maxlength is now invalid)

I think that simply passing ctrl.$$lastCommittedViewValue instead of viewValue to $$runValidators will solve the issue in #10126 without causing the problem I described. WDYT?

[Narretz]

You are absolutely right, this breaks validation if a parser only truncates the model value. I even had the last committed view value solution, but wanted to skip the extra validation. I'll fix this up.

[Narretz]

I updated the PR with your suggestion

[shahata]

LGTM