Discrepancies to `drun`

[ggreif]

So far I have found a few aspects where drun differs from ic-ref:

• former rejects hyphenated endpoints (e.g. memory-footprint)
• former rejects IC custom sections (i.e. prefixed with icp:) that are > 64kB

I am asking the oracle here, to clarify what the expected behaviour is, and shall write follow-ups.

[nomeata]

With endpoints you mean canister method names? The spec has no restrictions, as long as its UTF8-encoded text, and even stuff like the empty string should work. I had always meant to test hat with ic-ref-test, but never got around to do it; I think there is an issue on dfinity/ic-ref about it.

I also don’t see a reason for restricting the size of custom sections.

[ggreif]

Yes, endpoint = method name. Thanks for the explanation!

[mraszyk]

Methods with names such as canister_query get-counter-# are allowed by the latest replica.

[mraszyk]

The limit on custom section name length (incl. icp:public or icp:private prefix) is 100000 as of now (see here).

[mraszyk]

More generally, there are wasmtime-specific limits that are not reflected in the spec.