Alexey Knyazev

- [x] There're slightly different usage semantics of `ans_decoder_`: - `ProcessInteriorEdges` and `DecodeAttributeSeams` have local declarations of `ans_decoder_` of undefined type `AnsDecoder`. There're no such declarations in other relevant places....

We'll include this fix in the next spec release.

glTF doesn't have alignment requirements for image buffer views since PNG and JPEG formats do not pad/align their data. We may recommend aligning KTX buffer views to facilitate zero-copy processing...

Note that KTX2 needs 8-byte alignment because some of its header fields are 64-bit.

@zeux Only the KTX2 header may potentially benefit from an 8-byte alignment because both BasisU codecs (BasisLZ and UASTC/zstd) are very tightly packed and require per-bit reads anyway, so I...

> I would suggest 64 bits personally. In JS, all numbers are doubles (IEEE float-64), which means that the largest precisely representable integer value is 253, unless more sophisticated techniques...

Allowing `number | string` JSON type for all integers is incredibly error-prone: native exporters may easily write int64 values directly and then JS-based importers would be unable to recover the...

Enforcing JSON strings for BigInt values would certainly be an improvement, assuming that the spec would normatively reference JS BigInt parsing rules. On the other hand, there's no reason to...

To move this further, the PR needs JSON schemas for the extension objects.

> This is an invalid glTF. You are accessing a value with the JSON pointer of an extension, which does not exist in the glTF. This is a bit more...