Add bidirectional isolation #30

eemeli · 2023-08-18T15:09:30Z

This fulfills the spec requirements for Handling Bidirectional Text.

The ApplyBidiIsolation AO will in the spec text need more formal descriptions of how the "compatibility" and "none" strategies work, but I thought that would be a bit heavy for the README.

We could also define something like a "best fit" option to match what"s done with locale negotiation, but I left that out as I wasn"t sure exactly what it would look like.

CC @aphillips

aphillips

The locale can only be a hint.

In my view, the message as a whole should have a base paragraph direction. And each placeholder should have a base paragraph direction property.

When formatting to parts, one ends up with a sequence of parts that each have a queryable direction. When formatting to HTML this direction can be used to set the dir attribute on markup, which is how bidi isolation is done in HTML.

When formatting to a string, this algorithm is thus mostly correct in its behavior, since it places isolating controls around placeholders.

Some will note that a placeholder can, itself, contains "parts" that have direction. For example, a date format produces a sequence of "parts" that are date/time fields (such as year or hour or such) or literals. However, I note that these "interior parts" are the responsibility of the formatting function. The important part of bidi isolation in MF2 is that the placeholder "sticks together" (is spanned with isolates).

My concern with this proposal is that it does not start from parts or with having a directionality property. It tries to infer direction from locale (which is not always appropriate and thus shouldn"t form the foundation of directionality support)

eemeli · 2023-08-21T00:20:15Z

@aphillips Do I understand right that you think we should include dir as an explicit option for the formatter in addition to locale, and to include it for each of the formatted parts?

My presumption was that from the locale code we could get the script, from which we could determine the direction. But that"s not always valid then?

aphillips · 2023-08-21T14:10:46Z

@eemeli More like: every formatted part should have a dir property. This can be generated by a formatter (and might, indeed, be inferred from the locale if that is how the formatter figures out the base paragraph direction of e.g. a date string). However, not every placeholder is generated by a formatter. Having a dir property also allows assembly of messages whose character sequences would fool auto. Consider:

Input variables:

{  "item": {
    "id": "978-111887164-5",
    "title": {
        "value": "HTML و CSS: تصميم و إنشاء مواقع الويب!",
        "lang": "ar",
        "dir": "rtl"
    }
}

And message:

{Your purchased the item {$title.value @lang=$title.lang @dir=$title.dir}.}

In this case, you want the title bidi isolated (it"s in Arabic) and you don"t want a direction to be LTR (implied by HTML on the front). Only the metadata associated with the original data value conveys this.

We have a long document discussing this: Strings on the Web: Language and Direction Metadata. It includes a discussion of why language tags (and their extant or implied script subtag) is only a fallback.

Thus dir isn"t really a formatter option. But it might be input about some operand. And it might be something that is overridden on occasion.

eemeli · 2023-09-01T10:57:44Z

After discussion with @aphillips elsewhere, I"ve rebased and updated this PR to include dir as a supported option, and to use that rather than the locale to determine the isolation to use.

This also adds dir: "ltr" | "rtl" | "auto" as a property of the formatted parts that also include locale.

aphillips

Thanks for these changes. Some comments.

aphillips · 2023-09-01T14:40:12Z

README.md

+By default the message's directionality is determined from
+the script corresponding to the first locale,
+but this may be overridden by `dir`.
+Its `"auto"` value corresponds to messages with unknown directionality,
+for which the direction is determined by the first strongly directional character.


I would unpack this so that the hierarchy is clear.

Suggested change

By default the message"s directionality is determined from

the script corresponding to the first locale,

but this may be overridden by `dir`.

Its `"auto"` value corresponds to messages with unknown directionality,

for which the direction is determined by the first strongly directional character.

The base paragraph direction of a message is set by the `dir` property, if it is present.

If there is no `dir` property, the direction is determined by the result of `getTextInfo()` for the `locale` of the message.

If the direction cannot be determined, the value of `dir` is `"auto"` and directionality of the

message will be determined at runtime using the directionality of characters in the message

(known as "first-strong" detection).

aphillips · 2023-09-01T14:44:52Z

README.md

 Custom user-defined message formatting and selection functions may defined by the `functions` option.
 These allow for any data types to be handled by custom functions.
 Such functions may be referenced within messages,
 and then called with the resolved values of their arguments and options.

 ```ts
 interface MessageFormatOptions {
+  bidiIsolation?: 'compatibility' | 'none';


compatibility sounds like it exists for legacy reasons rather than wanting to be the default options. Perhaps give it a less redolent name, like unicode or controls? Would formatToParts have a different option (for when the caller wants to use markup)?

The "compatibility" name is explicitly mentioned in the MF2 spec: https://github.com/unicode-org/message-format-wg/blob/main/spec/formatting.md#handling-bidirectional-text

If it should be called something else, it would be best to re-name it there so that the same name will end up in other implementations as well.

The bidi isolation strategy is set in the constructor, so each needs to work for all formatting targets.

I agree that we should figure out the naming there.

README.md

aphillips · 2023-09-01T14:54:12Z

README.md

 1. Let `parts` be the result of calling `mv.toParts()`.
 1. If the call fails or `parts` is not an array:
-   1. Set `parts` to be `[{ type: 'fallback', locale: 'und', source: mv.source }]`.
+   1. Set `parts` to be `[{ type: "fallback", source: mv.source }]`.
+   1. Set `dir` to be `"auto"`.


Same comment as above.

aphillips · 2023-09-01T15:01:44Z

README.md

+1. Let `bidi` be the `{ start: string, end: string }` result of calling
+   `ApplyBidiIsolation(bidiIsolation, msgDir, dir)`.
+1. If `bidi.start` is not an empty string:
+   1. Append `{ type: 'bidiIsolation', value: bidi.start }` to `res`.
 1. For each `part` or `parts`:
   1. Append `part` to `res`.
+1. If `bidi.end` is not an empty string:
+   1. Append `{ type: 'bidiIsolation', value: bidi.end }` to `res`.


This is the discussion we were having on the parts spec in MF2. This design makes it hard for developers who are using the parts to build HTML using HTML"s dir attribute, because instead of each part just having a direction, the developer getting a bidiIsolation part has to peek ahead to figure out what the text is that it applies to. They then need to decide whether to generate separate bidi markup (such as a span element) or try to apply dir to a known inline element (which is why they got parts in the first place--to be able to decorate the string with em, strong, bdi 😉, or some CSS class.

This algorithm works for compatibility mode, since it just has the bidi control in it. Is your intention that none conveys what I just described above? If so, then the name might be wrong?

Perhaps make the names:

compatibility => controls
none => inline or maybe attributes

The bidiIsolation parts are in addition to the dir that"s on the formatted parts. So a developer building HTML could just ignore the bidiIsolation parts and build e.g. <span dir> elements as necessary from the dir attributes.

On the other hand, in this particular case the bidiIsolation parts also act as start/end markers in case the value formats to more than one top-level part (Note: this is different from e.g. number parts containing sub-parts), so that a single multi-part value could be completely encompassed in an appropriate <span dir> if necessary.

Regarding the "compatibility" name, see my comment here. Regarding "none", I still think it"s right to say that using it no bidi isolation is provided. Given that the dir values are still included in the parts output, it is of course possible for a consumer to construct its own bidi isolation from the provided information, but Intl.MessageFormat isn"t doing it in that case.

Okay, that works for me, although I still want to think about the naming.

aphillips · 2023-09-01T15:03:47Z

README.md

  locale: 'und';
+  dir: 'auto'


Both locale and dir can be known from the call to MF. Any reason to suppress that information?

For fallback, we can"t know without introspection what the source direction is, and we should not presume that the contents match the locale. In other words, we do not know the locale or direction of a fallback part.

Again, if source is externally provided and isn"t the pattern string, I can agree about the direction. The locale, though, really is the locale of the formatter, I think. Otherwise the caller won"t know if some locale defaulting happened. If the source is just an identifier, its actual language is probably "programmer-ese" and not any particular natural language. But the language of the words in the source don"t have to be in the locale"s language for that to be the correct locale for the message.

README.md

Co-authored-by: Addison Phillips <[email protected]>

aphillips

Changes look good. My comments about naming remain, but they are minor and I could live with what"s here.

eemeli · 2023-12-04T12:45:27Z

I"ve added an upstream issue about the "compatibility" name: unicode-org/message-format-wg#549

eemeli requested review from stasm, catamorphism, ryzokuken and dminor August 18, 2023 15:09

aphillips reviewed Aug 18, 2023

View reviewed changes

Add bidirectional isolation

d682b26

eemeli force-pushed the bidi branch from 33921e2 to d682b26 Compare September 1, 2023 10:55

eemeli requested a review from aphillips September 1, 2023 10:57

aphillips suggested changes Sep 1, 2023

View reviewed changes

Apply suggestions from code review

2899e08

Co-authored-by: Addison Phillips <[email protected]>

eemeli mentioned this pull request Sep 17, 2023

Apply recent Intl.MessageFormat spec updates messageformat/messageformat#410

Merged

eemeli added 2 commits December 2, 2023 16:38

Merge branch "main" into bidi

8ec36a2

Add missing bidiIsolation parts to MessagePart

297aba3

aphillips approved these changes Dec 2, 2023

View reviewed changes

eemeli mentioned this pull request Dec 4, 2023

Consider renaming the "compatibility" bidi isolation strategy unicode-org/message-format-wg#549

Closed

eemeli merged commit f70da0f into main Dec 4, 2023

eemeli deleted the bidi branch December 4, 2023 12:45

eemeli mentioned this pull request Dec 18, 2023

Wrap MessageFunction source & locales args in an object, adding dir #38

Merged

eemeli mentioned this pull request Mar 29, 2024

This proposal is stuck #49

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add bidirectional isolation #30

Add bidirectional isolation #30

eemeli commented Aug 18, 2023

aphillips left a comment

eemeli commented Aug 21, 2023 •

edited

Loading

aphillips commented Aug 21, 2023 •

edited

Loading

eemeli commented Sep 1, 2023

aphillips left a comment

aphillips Sep 1, 2023

aphillips Sep 1, 2023

eemeli Sep 1, 2023

aphillips Sep 1, 2023

aphillips Sep 1, 2023

aphillips Sep 1, 2023

eemeli Sep 1, 2023

aphillips Sep 1, 2023

aphillips Sep 1, 2023

eemeli Sep 1, 2023

aphillips Sep 1, 2023

aphillips left a comment

eemeli commented Dec 4, 2023

-By default the message"s directionality is determined from
-the script corresponding to the first locale,
-but this may be overridden by `dir`.
-Its `"auto"` value corresponds to messages with unknown directionality,
-for which the direction is determined by the first strongly directional character.
+The base paragraph direction of a message is set by the `dir` property, if it is present.
+If there is no `dir` property, the direction is determined by the result of `getTextInfo()` for the `locale` of the message.
+If the direction cannot be determined, the value of `dir` is `"auto"` and directionality of the
+message will be determined at runtime using the directionality of characters in the message
+(known as "first-strong" detection).

Add bidirectional isolation #30

Add bidirectional isolation #30

Conversation

eemeli commented Aug 18, 2023

aphillips left a comment

Choose a reason for hiding this comment

eemeli commented Aug 21, 2023 • edited Loading

aphillips commented Aug 21, 2023 • edited Loading

eemeli commented Sep 1, 2023

aphillips left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aphillips left a comment

Choose a reason for hiding this comment

eemeli commented Dec 4, 2023

eemeli commented Aug 21, 2023 •

edited

Loading

aphillips commented Aug 21, 2023 •

edited

Loading