Introduce sanity checks for IDs and fix overriding its type

[malarzm]

Fixes #857 (and by the way fixes #1178), closes #1183 and closes #819.

[alcaeus]

Technically, this could happen with any generator strategy, including custom. It probably would be best if the generator itself could handle this check, including on whether to ignore it, throw an exception or quietly change the value.

[malarzm]

Partially I agree but since AUTO strategy is the default one its misuse is the most problematic (see #819 or tests I needed to fix in this PR like this one), I think we can safely assume that if somebody is changing strategy he is clever enough to pass correct id (or will have his own checks in the type).

[alcaeus]

To be honest, I don't know why we have the generatorOptions type at all. In the end, it does exactly the same as the field mapping but it just introduces complexity and mistakes along the way. I'm not sure about BC here, but IMO the best option would be to drop generatorOptions.type, always use type from the field mapping and setting it to id or custom_id depending on the generator strategy only if no type was given. As it is now, writing a generator that generates a MongoId still requires specifying a generatorOptions.type mapping because the field mapping of id will be overwritten (which, incidentally, needed a trip through the debugger to figure out why it wouldn't work when I first tried it without generatorOptions.type).

Either way we do this, we need to write some documentation about this feature.

[malarzm]

I'm not sure about BC here, but IMO the best option would be to drop generatorOptions.type

Maybe trigger deprecation?

writing a generator that generates a MongoId

What is the point of such generator?

Either way we do this, we need to write some documentation about this feature.

I'll create a ticket for this later

[alcaeus]

What is the point of such generator?

The use case is a bit special: we needed predictable identifiers to trigger upserts but wanted to keep MongoId values. So we assigned a specific time identifier (the first 4 bytes of MongoId) and added a hash based on document settings. Weird, but it works ;)

[malarzm]

@alcaeus I've commited a change we discussed on IRC :)

[alcaeus]

@malarzm Looks good. This way we ensure using MongoId with AUTO strategy but don't overly complicate the mapping. I've also created #1197 to track documentation changes.