Spec text format by rossberg · Pull Request #385 · WebAssembly/gc

rossberg · 2023-06-12T19:49:21Z

Also, to sync up, implement dependent name spaces for fields in interpreter.

tlively

Does this contain the abbreviation allowing fields to be combined, e.g. (field i32 i64)? I didn't see that when reading through this PR.

tlively · 2023-06-13T20:02:37Z

+   If inline declarations are given, their types must be *syntactically* equal to the types from the indexed definition;
+   possible type :ref:`substitutions <notation-subst>` from other definitions that might make them equal are not taken into account.
+   This is to simplify syntactic pre-processing in the presence of recursive types.


Can you give an example of what this allows and disallows? I don't grok the implications yet.

For example, if you have

(type $s1 (struct)) (type $s2 (struct)) (type $t (func (param (ref $s1)))) (func $f (param (ref $s2)))

then it would not pick up $t but introduce a new $t2 that is using $t2. This is so that desugaring does not depend on type canonicalisation. This choice doesn't affect typability, and the difference is not observable for well-formed modules, unless you were to inspect the desugared syntax itself to discover the additional type entries.

Edit: Actually, type definitions added by desugaring are "observable" if you are sneaky and predict their raw type index. For example, this is technically valid, and always has been:

(func $f (param i32)) (func $g (type 0))

You could abuse this to observe that the above example indeed introduces an additional type as well.

Makes sense, thanks!

tlively · 2023-06-13T20:05:07Z

+where :math:`x` is the smallest existing :ref:`type index <syntax-typeidx>` whose definition in the current module is the :ref:`final <syntax-final>` :ref:`function type <syntax-functype>` :math:`[t_1^\ast] \toF [t_2^\ast]`.
+If no such index exists, then a new :ref:`recursive type <text-rectype>` of the form


Does this capture the constraint that only types in singleton rec groups are eligible to be found? IIRC, that's how we decided this should work.

Ah, you are right, that's missing. Fixed and made more precise.

tlively · 2023-06-13T20:19:29Z

+
+  (func (param (ref 0)) (result i32) (struct.get 0 $x (local.get 0)))
+  (func (param (ref $t1)) (result f32) (struct.get 1 $x (local.get 0)))
+  (func (param (ref $t2)) (result i64) (struct.get $t2 $x (local.get 0)))


It would be good to have tests for all combinations of #names and raw numbers for the type index and the field index if we don't have that already. Did you decide to do anything to anticipate having instructions with field indices but no type indices?

I added a simple test for each combination. (To be honest, we could probably use more tests for the various struct/array instructions in general.)

In terms of anticipation, no, what do you have in mind?

I had one idea for future-proofing the syntax described in the second paragraph here: #361 (comment)

Can you elaborate on that idea? I'm not really understanding why the syntax struct.get t (field t i) is more future-proof than struct.get t i. What does repeating the type twice buy us?

In the future if we have some struct.get_unannotated <fieldidx> instruction that has no type annotation, and if the text syntax for fieldidx contains a type index along with the symbolic field name, then we can still unambiguously figure out how to resolve the symbolic field name to an index. For the current instructions, which already contain type indices, the extra type index in the fieldidx syntax is redundant, so we can define abbreviations to make the repetition unnecessary.

Why couldn't we just use the text syntax struct.get_unannotated tidx? $x in that case? (Allowing omitting the type index only when the field index is numeric.)

That would work, too! If you'd be happy with that, then it sounds good to me.

rossberg

Comments addressed, PTAL.

rossberg · 2023-06-14T09:27:38Z

+where :math:`x` is the smallest existing :ref:`type index <syntax-typeidx>` whose definition in the current module is the :ref:`final <syntax-final>` :ref:`function type <syntax-functype>` :math:`[t_1^\ast] \toF [t_2^\ast]`.
+If no such index exists, then a new :ref:`recursive type <text-rectype>` of the form


Ah, you are right, that's missing. Fixed and made more precise.

rossberg · 2023-06-14T09:36:35Z

+
+  (func (param (ref 0)) (result i32) (struct.get 0 $x (local.get 0)))
+  (func (param (ref $t1)) (result f32) (struct.get 1 $x (local.get 0)))
+  (func (param (ref $t2)) (result i64) (struct.get $t2 $x (local.get 0)))


I added a simple test for each combination. (To be honest, we could probably use more tests for the various struct/array instructions in general.)

In terms of anticipation, no, what do you have in mind?

tlively

LGTM % my previous comment about the (field i32 i64 ...) abbreviation.

rossberg · 2023-06-14T17:02:13Z

LGTM % my previous comment about the (field i32 i64 ...) abbreviation.

Oops, sorry, had missed that comment. Indeed, added now.

rossberg added 2 commits June 12, 2023 16:45

Spec text format

9d50b9f

Implement dependent field names

f4d7591

rossberg requested a review from tlively June 12, 2023 19:49

rossberg changed the base branch from main to spec.binary June 12, 2023 19:49

rossberg mentioned this pull request Jun 12, 2023

Spec authoring tracking issue #376

Open

53 tasks

Base automatically changed from spec.binary to main June 13, 2023 09:09

Merge branch 'main' into spec.text

0995179

tlively reviewed Jun 13, 2023

View reviewed changes

Comments

0ed6679

rossberg commented Jun 14, 2023

View reviewed changes

tlively approved these changes Jun 14, 2023

View reviewed changes

Fields abbreviation

09e0055

rossberg merged commit 5c128c9 into main Jun 14, 2023

rossberg deleted the spec.text branch June 14, 2023 17:03

rossberg mentioned this pull request Jul 11, 2023

Text format: field names #361

Closed

		where :math:`x` is the smallest existing :ref:`type index <syntax-typeidx>` whose definition in the current module is the :ref:`final <syntax-final>` :ref:`function type <syntax-functype>` :math:`[t_1^\ast] \toF [t_2^\ast]`.
		If no such index exists, then a new :ref:`recursive type <text-rectype>` of the form

Conversation

rossberg commented Jun 12, 2023

Uh oh!

tlively left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rossberg Jun 14, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rossberg left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tlively left a comment

Choose a reason for hiding this comment

Uh oh!

rossberg commented Jun 14, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rossberg Jun 14, 2023 •

edited

Loading