As discussed offline, I realized while reviewing this patch that we do need to have the Locale on the TextStyle, because a single Text widget can have multiple parts of its text with different locales, e.g. "The English word foo is bar in Japanese and baz in Chinese" would have three locales.

The TextStyle should probably default to null locale and then Text can fill it in from the ambient Locale.

The TextStyle should probably default to null locale and then Text can fill it in from the ambient Locale.

If you create a RichText with TextSpans whose TextStyles have non-null locales, the RichText will carry the ambient locale and the TextSpan styles will carry along their locales via the ui.ParagraphBuilder.

In this case the engine uses the text style's locale.

I've verified that this works. Will add tests when tests can use fonts.

    final String character = '骨';
    final TextStyle style = Theme.of(context).textTheme.display3;
    return new Scaffold(
      body: new Container(
        padding: const EdgeInsets.all(48.0),
        alignment: Alignment.center,
        child: new RichText(
          text: new TextSpan(
            children: <TextSpan>[
              new TextSpan(text: character, style: style.copyWith(locale: const Locale('ja'))),
              new TextSpan(text: character, style: style.copyWith(locale: const Locale('zh'))),
            ],
          ),
        ),
      ),
    );

Can you add the test anyway today (as a golden test) and then add a TODO to enable it, with a link to the relevant font bug? That way when we implement it we'll have a better chance to also update the test.

We should also add a test that checks that when the text style has a locale and the text has a locale, the text style one isn't replaced.

I've added golden image tests that will run and succeed with the Ahem font. When #17700 is fixed, the golden files will need to be updated

This issue covers updating the golden images : #18317