There should be an escape hatch to allow unknown import attributes

[gibson042]

Now that #131 has established a requirement for implementations to reject unknown attributes, deferring the upgrade concerns of https://github.com/tc39/proposal-import-assertions/issues/21#issuecomment-551987820 and https://github.com/tc39/proposal-import-assertions/issues/56#issuecomment-629698719 is more hazardous. As suggested in https://github.com/tc39/proposal-import-assertions/pull/131#issuecomment-1453846367 , I think it would be valuable to add an escape hatch allowing authors to explicitly indicate that particular import attributes should be ignored when they are unknown to an implementation. It would be logically similar to the ! criticality flag of the IETF timestamp suffixes draft, but with reversed polarity. As a strawperson, @nicolo-ribaudo suggested syntax like with { type: "css", layer?: "utilities" } or with { type: "css", "layer?": "utilities" }.

And a final point noted at https://github.com/tc39/proposal-import-assertions/pull/131#issuecomment-1458296576 :

an alternative design could involve consideration of not just the attribute name but also the value, such that a new implementation might want to reject import "./image.png" with { type: "image", "as?": "not-yet-shipped-format" } while an old implementation that doesn't even know about as would treat the import like with { type: "image" }. But that kind of variance doesn't seem helpful in either the name-only or the name-and-value design anyway. It's worth noting that neither design is fully general with respect to fallback, but I'm just striving for a bridge to something better while the range of valid values for any particular attribute has not been subject to many changes and thus that lack of generality isn't detrimental.

[ljharb]

I think that it's very important there not be an escape hatch - giving the ecosystem carte blanche to throw whatever they like in here risks fragmenting the language.

[gibson042]

The proposal already allows each implementation to define its own list of supported attributes: https://tc39.es/proposal-import-assertions/#sec-hostgetsupportedimportassertions

This suggestion wouldn't change that. It would, however, allow a way for new assertions to be added without requiring fragmentation of source files.

[juner]

@gibson042 Maybe the link has changed to this? https://tc39.es/proposal-import-attributes/#sec-hostgetsupportedimportattributes