Bug 8100 – REGRESSION (NativeTextField): Right-aligned and RTL text jitters in new text fields

• Component: Forms
• Status: VERIFIED FIXED

Summary: glyphs' horizontal positions slightly change as more text is entered into a right-aligned or RTL text field.

To reproduce:
1) Open the attached test case.
2) In the first text field, type lower-case "x"s. Watch the glyphs' positions as you add more "x"s.
3) In the second text field, type (or paste) Alephs. Watch the position of the last-typed Aleph as you type the next one.
4) From the Debug menu, choose Use ATSU For All Text, reload and repeat (in step 3, watch all Alephs).

I think these are artifacts of the "rounding hack". They can be observed in Mail too (and in other editable or dynamic HTML).

The right-aligned case is somewhat artificial, but the RTL case is a real problem.

Created attachment 7415 [details]
Test case

Created attachment 7416 [details]
Test case

Oops

Created attachment 7443 [details]
Apply rounding on the left in RTL runs

This patch addresses the RTL part of this bug. The right-aligned-LTR case (and the symmetric left-aligned-LTR case) are arguably non-issues. I think that in order to solve them, WebTextRenderer will need to know if the run is going anchored on the right or on the left, and use that instead of the rtl flag.

This patch affects the pixel results of all tests that include RTL test in Lucida Grande (and maybe other fractional-width fonts). Updated results will need to be generated and landed.

[Darin Adler]

Comment on attachment 7443 [details]
Apply rounding on the left in RTL runs

OK, r=me

[Darin Adler]

(In reply to comment #3)
> This patch addresses the RTL part of this bug. The right-aligned-LTR case (and
> the symmetric left-aligned-LTR case) are arguably non-issues. I think that in
> order to solve them, WebTextRenderer will need to know if the run is going
> anchored on the right or on the left, and use that instead of the rtl flag.

Seems like the jittering *is* an issue in the right-aligned-LTR and left-aligned-RTL cases, although a much smaller one. Maybe we should have a second bug report about those.

[Maciej Stachowiak]

These are all text field regressions so they should all be P1.