Spec test requires different functions to compose with each other

[catamorphism]

Right now the spec tests include a test for the :time and :datetime functions:

.input {$dt :time style=medium} {{{$dt :datetime dateStyle=long}}}

The expected output is: "January 2, 2006 at 3:04:06 PM".

To realize this expected output, the implementation would have to uphold the following properties in this specific instance:

• :time returns an object that encodes the option style=medium as well as the formatter (:time) used to construct it
• :datetime can inspect this object, check that it was constructed by :time, and treat the value of the style option as the value of its own timeStyle option

(:time can't just be treated as an alias to :datetime, since their options have different names.)

These requirements go beyond what #686 added to the spec, because they require different functions to compose with each other. @aphillips commented to that effect on #686: 'For now, we should basically say something like "functions can decide what operands to support" and "functions can decide what functions or function options to support".'

I don't think this test should be in the spec test suite at this time (perhaps there should be an appendix to the spec test suite to show what implementations can do if they choose to support more composability between functions), because it requires the implementation to support some pretty complex logic for which functions can compose with each other and how, without providing the machinery in the spec to guide the implementation of that logic. If the spec said something like, ":time has to return something that can be inspected to get the options and the name of the formatter used to construct it, and :datetime has to accept things created with :time or :date and use their options in combination with its own options", then it would be fine to require all implementations to produce the expected output here.

#645 could address this, but then there would also need to be specific language in the registry spec about the requirements for composability between :date/:time/:datetime.

[macchiati]

I agree with you about removing this until (and unless) we have built up the spec and registry to support it.

…

On Wed, Mar 13, 2024, 11:47 Tim Chevalier ***@***.***> wrote: Right now the spec tests include a test <https://github.com/unicode-org/message-format-wg/blob/main/test/test-functions.json#L77> for the :time and :datetime functions: .input {$dt :time style=medium} {{{$dt :datetime dateStyle=long}}} The expected output is: "January 2, 2006 at 3:04:06 PM". To realize this expected output, the implementation would have to uphold the following properties in this specific instance: - :time returns an object that encodes the option style=medium as well as the formatter (:time) used to construct it - :datetime can inspect this object, check that it was constructed by :time, and treat the value of the style option as the value of its own timeStyle option (:time can't just be treated as an alias to :datetime, since their options have different names.) These requirements go beyond what #686 <#686> added to the spec, because they require different functions to compose with each other. @aphillips <https://github.com/aphillips> commented <https://github.com/unicode-org/message-format-wg/pull/686/files#r1504951171> to that effect on #686 <#686>: 'For now, we should basically say something like "functions can decide what operands to support" and "functions can decide what functions or function options to support".' I don't think this test should be in the spec test suite at this time (perhaps there should be an appendix to the spec test suite to show what implementations can do if they choose to support more composability between functions), because it requires the implementation to support some pretty complex logic for which functions can compose with each other and how, *without* providing the machinery in the spec to guide the implementation of that logic. If the spec said something like, ":time has to return something that can be inspected to get the options and the name of the formatter used to construct it, and :datetime has to accept things created with :time or :date and use their options in combination with its own options", then it would be fine to require all implementations to produce the expected output here. #645 <#645> could address this, but then there would also need to be specific language in the registry spec about the requirements for composability between :date/:time /:datetime. — Reply to this email directly, view it on GitHub <#726>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACJLEMDWLVSHO647N6LRXHLYYCNNFAVCNFSM6AAAAABEUXCVI6VHI2DSMVQWIX3LMV43ASLTON2WKOZSGE4DINRXGYZDMOI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[aphillips]

The expected output of this test is dead wrong, in my opinion. The meta-point is also correct: we shouldn't have a test like this. Let's just remove it.