Can the list of used languages in a chunk always be derived from the content?

[ftomassetti]

In the serialization chunk we list the used languages. Now, I am wondering if the list can always be derived from the metapointers used in the chunk or there are case when this does not happen. And this leads me to a second question about its meaning.

For example:

• In the chunk I serialize a node of Concept C, and Concept C implements an interface I from another language. Interface I does not define features, or it defines features for which the node has empty values, and so they are omitted. There will be no reference to I or its language in the chunk.
• In the chunk I serialize a node of Concept C, which has containments or references that can hold values of type D. If they are empty type D is not mentioned and it could be coming from another language

So it seems to me that these unmentioned languages are necessary to properly work with these nodes.

So I wonder if we should interpret the list of used languages:

• As a sort of cache that can be calculated by looking at the chunk
• As a carefully calculated list of languages which specify which languages are necessary to work with the nodes in the chunk, even if it is not obvious at a first glance

[enikao]

At the moment we specified it as

elements MUST contain all language/version referred to by any Meta-pointer in the processed document.

So only what's inside the chunk.

Your proposed alternative would indeed be useful. I could construct edge cases where this leads to problems. For example if we added several new empty super-interfaces from not-yet-mentioned languages to instantiated concepts -- the content of the chunk would not change, but the list of used languages. However, we said in #213 that we consider a language (identified by [key, version]) fixed -- so if we changed the language without changing the version number, we're in unspecified territory.