allow narrowing of inherited types

[mr-c]

No description provided.

[codecov]

Codecov Report

Merging #535 (4020d0f) into main (c6cc2e5) will increase coverage by 0.26%.
The diff coverage is 100.00%.

@@ Coverage Diff @@ ## main #535 +/- ## ========================================== + Coverage 81.01% 81.27% +0.26%  ========================================== Files 19 19 Lines 3597 3631 +34 Branches 991 1008 +17 ========================================== + Hits 2914 2951 +37  + Misses 451 449 -2  + Partials 232 231 -1 

Impacted Files	Coverage Δ
schema_salad/avro/schema.py	67.24% <100.00%> (+4.50%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c6cc2e5...4020d0f. Read the comment docs.

[tetron]

If I understand correctly, the goal is to be able to re-specify the type of a record field with a narrower range of allowed types than the the field in record it inherits from.

This seems reasonable, since the inherited type will still be a valid shape of the parent type.

Implementation comments:

• Should recognize parent field type of "Any" (with reminder that "Any" is any non null)
• Should say something about this in the documentation https://www.commonwl.org/v1.2/SchemaSalad.html#Inheritance_and_specialization -- is this a schema salad 1.2 feature?
• Implementation covers arrays, but not records and enums. Is that a stretch goal or intentionally excluded? If we support narrowing for those types, should it be based on explicit inheritance, or duck typing (I would say duck typing, but I wanted to raise the issue).

[tetron]

I'm also curious, what is the specific motivation is to add this now?

[tetron]

See previous comments

[mr-c]

I'm also curious, what is the specific motivation is to add this now?

It would help with common-workflow-language/cwltool#1665

[mr-c]

explicit inheritance, or duck typing

is_subtype() is only triggered if the existing field has inherited_from set; so it is always an explicit inheritance check, in a way. However I'm currently comparing enums & records via duck typing.

[mr-c]

• is this a schema salad 1.2 feature?

Good point, I've added documentation about that.