AutoFill: Fix two heuristic issues with nrm.org.

AutoFill: Fix two heuristic issues with nrm.org.
* Parse state before zip, as they may share one label.
* Try to parse the current TD for text when parsing table labels.

BUG=61442
TEST=FormManagerTest.LabelsInferredFromTablTDInterveningElements,
     AddressFieldTest.ParseStateAndZipOneLabel

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=64761

[James Hawkins, 2010-11-02 01:58:53]

Whoever gets it first...

[dhollowa, 2010-11-02 14:47:51]

LGTM.

[Ilya Sherman, 2010-11-02 15:35:08]

LGTM, too

http://codereview.chromium.org/4249002/diff/1/2
File chrome/browser/autofill/address_field.cc (right):

http://codereview.chromium.org/4249002/diff/1/2#newcode344
chrome/browser/autofill/address_field.cc:344: if (address_field->state_)
nit: Do we not want "&& !address_field->state_->IsEmpty()" here, as we had in
the previous logic?

[James Hawkins, 2010-11-02 17:32:13]

http://codereview.chromium.org/4249002/diff/1/2
File chrome/browser/autofill/address_field.cc (right):

http://codereview.chromium.org/4249002/diff/1/2#newcode344
chrome/browser/autofill/address_field.cc:344: if (address_field->state_)
On 2010/11/02 15:35:08, Ilya Sherman wrote:
> nit: Do we not want "&& !address_field->state_->IsEmpty()" here, as we had in
> the previous logic?

No. I don't know why it was written in the first place, but it's wrong, because
the type of a field shouldn't depend on the emptiness of the value of the field.