fix: Speed-up readxml by caching key lookup instead of using try/except

[kratsg]

Pull Request Description

This resolves #1690 and was brought up by the discussion in #1687.

Fundamentally, when accessing a path that does not exist in uproot via f.__getitem__, it would then trigger a user-friendly levenshtein distance calculation to recommend similarly-named keys. However, in the case of pyhf, it was assumed that f[name] would be fast to return or at least fast to raise an exception, but this is a bad assumption.

Instead, the set of keys in the file (without cycle numbers) is cached along with the file itself, and we will rely on a key lookup in the set instead.

This only occurs in HistFactory when the XML has <Sample/> with HistoPath attributes and a suboptimal computation path in pyhf is then used, because we need to build fullname in order to figure out the right path for the histogram.

Checklist Before Requesting Reviewer

• Tests are passing
• "WIP" removed from the title of the pull request
• Selected an Assignee for the PR to be responsible for the log summary

Before Merging

For the PR Assignees:

• Summarize commit messages into a comprehensive review of the PR

* Refactor the readxml logic
* Cache the set of keys in a file and use that to look up if a histogram
exists or not
   - Avoids issue of accessing a path that does not exist in uproot via
     f.__getitem__, triggering a user-friendly levenshtein distance
     calculation to recommend similarly-named keys.
* Add tests and test files for key name lookups

[kratsg]

@gollumben - this has the fix.

[codecov]

Codecov Report

Merging #1691 (cde983b) into master (c493c7c) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1691      +/-   ##
==========================================
- Coverage   98.05%   98.05%   -0.01%     
==========================================
  Files          64       64              
  Lines        4213     4212       -1     
  Branches      585      587       +2     
==========================================
- Hits         4131     4130       -1     
  Misses         49       49              
  Partials       33       33              

Flag	Coverage Δ
contrib	25.30% <0.00%> (+<0.01%)	⬆️
doctest	61.06% <0.00%> (+0.01%)	⬆️
unittests	96.36% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/pyhf/readxml.py	94.37% <100.00%> (-0.04%)	⬇️

---

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 c493c7c...cde983b. Read the comment docs.

[matthewfeickert]

LGTM. 👍 Thanks @kratsg!

[matthewfeickert]

Thanks for the coverage boost. LGTM now.

[matthewfeickert]

@gollumben If you are willing to install a development version of pyhf from GitHub your Issue described in Discussion #1687 should be fixed now. Please let us know if you encounter any other problems. 👍

[gollumben]

Hi @kratsg, @matthewfeickert and @lukasheinrich, thank you very much for your help and the very fast turn around time! It takes 3 minutes to convert the workspace now instead of the previous 30 h (~600x speed increase), which is amazing!

[kratsg]

Related to scikit-hep/uproot5#595 as well.