Sort attributes by name, not by source

[lmolkova]

How attributes are added to semconv is an implementation detail, rendered markdown should not depend on it.

This change sorts attributes in final md table alphabetically.

This would also reduce diffs in semconv and help us review uninteresting changes when hierarchy refactorings cause massive changes because of attributes moving in the table.

[Oberon00]

I would prefer either using a sensible order in the YAML (alphabetical might or might not be the best, there might be closely related attributes that are best put together or where there is another logical order), or one could argue that this should instead be a lint (error if input is not correctly sorted)

[lmolkova]

I would prefer either using a sensible order in the YAML (alphabetical might or might not be the best, there might be closely related attributes that are best put together or where there is another logical order), or one could argue that this should instead be a lint (error if input is not correctly sorted)

why though? what's the benefit to someone reading the semconv?

For authors, there is always a possibility of rendering individual tables per group of attributes. Having a predictable order also helps to review things.

As a reader, I want to be able to skim through the table and find my attribute in certain place, not knowing where author of semconv wanted to put it.

[Oberon00]

Having a predictable order also helps to review things.

Definitely. I was under the impression that currently the order is the same as in the YAML? Do we have some random order?

[lmolkova]

Definitely. I was under the impression that currently the order is the same as in the YAML? Do we have some random order?

the order is random since:

1. afaik nobody puts any meaning into order of attributes in yaml.
2. even if we did, extends are implementation detail, but once we're adding one, we no longer control the relative order for defined and extended attributes.

either way, the order author had in mind is not obvious to readers, who end up with de-facto randomly ordered attributes
PTAL here (or anywhere else)

https://github.com/open-telemetry/semantic-conventions/blob/main/docs/http/http-spans.md#common-attributes

[lmolkova]

One more observation: since we have a lot of namespaces now, most attributes are going to be referenced from somewhere and namespaces provide all the reasonable grouping.

Later on, we can consider adding sorting by requirement level and whatnot, but current random thing hurts right now during reviews.

[jsuereth]

I'm in favor of moving this direction. The simplicity gained is a big win.