Sum/Histogram/Gauge.data_points annotated as Sequence but actual type is generator

[aabmass]

This applies to all of the MetricsData data types Sum, Histogram and Gauge. The type annotation is Sequence:

opentelemetry-python/opentelemetry-sdk/src/opentelemetry/sdk/metrics/_internal/point.py

Line 68 in 778c4b1

data_points: Sequence[NumberDataPoint]

But the implementation assigns an Iterable (specifically a generator which is not a valid Sequence):

opentelemetry-python/opentelemetry-sdk/src/opentelemetry/sdk/metrics/_internal/metric_reader_storage.py

Line 160 in 778c4b1

data_points=view_instrument_match.collect(

What is the expected behavior?
Type annotated as Sequence should work with len(), subscription, etc.. A more subtle bug the generator causes is that data_points can only be iterated over once, which will consume the generator.

What is the actual behavior?

TypeError: object of type 'generator' has no len()

Additional context
Possible fixes

I'd recommend adding calls to list() or tuple() to convert the generator to a sequence or alter _ViewInstrumentMatch.collect() to return a sequence itself. I think this is appropriate as the generators are actually holding SDK locks which can remain locked if the generator is not consumed.

[lancetarn]

This seems like an approachable first issue for me, would give me some introduction to the metrics implementation and force me to get the project up and running. Do you mind if I give it a shot?

[ocelotl]

Dumb question, can't we just change Sequence to Iterable? (would that be a breaking change?)

I am more concerned with this statement, @aabmass:

I think this is appropriate as the generators are actually holding SDK locks which can remain locked if the generator is not consumed.

Can this produce hard-to-find bugs? Is there a way to test if this actually happens?