Define algorithms for dictionaries with lists as default values

[anssiko]

Via @domenic https://github.com/webmachinelearning/webnn/pull/204/files#r703625252

You need to have a normative algorithm which checks if input["scales"] exists, and if not, use the defaults. I.e. something like

Let scales be the list « 1.0, 1.0, 1.0, 1.0 ».
If input["scales"] exists, then set scales to input["scales"].

(and maybe also validate that input["scales"] is of length 4, or whatever you need?)

This seems difficult since right now your spec has no algorithms (i.e. no method steps) for its operations, just a non-normative description of the inputs and return value.

We should make this a global update to all affected dictionaries. We probably want to define a reusable algorithm we reference from all these places.

Given this issue is specific to dictionaries with list defaults, for "method steps" convention in general, I opened #210

[inexorabletash]

This is mostly taken care of by 0fe9e5c and prior changes, albeit with slightly different wording than suggested by @domenic

Suggested:

Let scales be the list « 1.0, 1.0, 1.0, 1.0 ».
If input["scales"] exists, then set scales to input["scales"].

Current:

If options.padding does not exist, set it to « 0, 0, 0, 0 ».

Differences:

• Literal list syntax « 0, 0, 0, 0 » is used without "the list" prefix (linked to the Infra definition) - I missed this!
• Dictionary member syntax - see Use web spec best practices #450 (comment) for some discussion
• Unpacking to a local variable vs. mutating the dictionary. Either is fine; while the latter looks like it would modify the passed JS object but the WebIDL processing model means that doesn't happen. In most cases in the spec an options dict is passed verbatim to "... an implementation-defined platform operator for the ..., given options." (see Improve "underlying platform" references in spec #462 for some thoughts)

I'll put up a PR for the "the list" wording change, after that perhaps this can be closed?

[zolkis]

I'll put up a PR for the "the list" wording change, after that perhaps this can be closed?

I like the most compact formulation, which simplifies reading of a long algorithm:

If options.padding does not exist, set it to the list « 0, 0, 0, 0 ».

But I get the logic in Domenic's suggestion, especially when used in a shorter algorithm. However, when there are repetitions, IMHO the more compact version is far more readable.

[inexorabletash]

layerNormalization() still has axes that have "assumed to be [1,2,3]" but nothing in the algorithm. I'll file a PR for that.

[inexorabletash]

Testing: @webmachinelearning/triage folks should review and consider closing this after the above PR merges

[inexorabletash]

I claim this can be closed now, so I'm gonna do it. New bugs if we missed any cleanup, please!