Refactor the taxonomy script for testing classification

[kwryankrattiger]

I re-worked this a bit so I can use this script as an imported module for testing.

import upload_gitlab_failure_logs as taxonomy

with read("my-test-trace.txt", "r") as trace:
    print(taxonomy.classify(trace))

[mvandenburgh]

These changes are good with me, but they will conflict with @AlmightyYakob's PR #433.

[kwryankrattiger]

np, it is a pretty easy thing to fix-up. I also added the script I used to do the testing.

Usage is something like

python validate_taxonomy.py $(ls trace*)

[alalazo]

Following #440 (comment), why can't we:

Suggested change

	return job_error_class
	return matching_patterns, job_error_class

?

[alalazo]

And here do something like:

Suggested change

	job_error_class = classify(job_trace)
	matches, job_error_class = classify(job_trace)
	job_input_data["matches"] = sorted(matches)

?¹

Footnotes

1. Obviously, "matches" is just a placeholder, so any better name for the field works for me ↩

[kwryankrattiger]

I think one good reason is not filtering to a single match can lead to mislabeled classifications. The ordering below is meant to identify the "best' error match.

For example, file_not_found is only an error if there are no other errors since it is possible for uploading artifacts to not find a file, but not be a relevant error.

[alalazo]

I don't mean to hide taxonomy, that's another field (as you can see I am returning both). I'd like to know all the matches. Then I can try to query things like: how many fetch errors also match a network error? Or stuff like that...

[mvandenburgh]

Sorry, thought I approved this already. This will need to be rebased to resolve conflicts but the changes LGTM