repro: introduce --continue-on-error

[skshetry]

Fixes #8647 and #7292.

TODO:

• Tests
• How to get rid of an extra newline at the end?

This PR might change the execution order.

graph TD
6-->2
7-->2
3-->1
4-->3
5-->3
2-->1

Loading

If you have a graph like this, the execution order would have been: [4,5,3,6,7,2,1].
It's a valid topological ordering.

But with this PR, we'll do order based on degree, so, zero-degree nodes 4,5,6,7 will get processed first, then 3 and 2 will be processed and then 1. i.e. [4,5,6,7,3,2,1].
Which is still a valid topological ordering.

I think the new logic makes it simpler for implementing --continue-on-error, and allows easier parallelization in the future.

[codecov]

Codecov Report

Patch coverage: 69.62% and project coverage change: -0.07 ⚠️

Comparison is base (63de16c) 90.53% compared to head (1189a68) 90.47%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9705      +/-   ##
==========================================
- Coverage   90.53%   90.47%   -0.07%     
==========================================
  Files         480      480              
  Lines       36421    36480      +59     
  Branches     5232     5243      +11     
==========================================
+ Hits        32975    33005      +30     
- Misses       2855     2880      +25     
- Partials      591      595       +4     

Impacted Files	Coverage Δ
tests/unit/command/test_repro.py	100.00% <ø> (ø)
dvc/repo/reproduce.py	82.26% <68.83%> (-16.50%)	⬇️
dvc/commands/repro.py	100.00% <100.00%> (ø)
dvc/exceptions.py	93.70% <100.00%> (-0.09%)	⬇️

... and 2 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[dberenbaum]

@skshetry Do you need QA? Product-wise, LGTM except for the long flag name, but I don't have any better ideas for it.

[skshetry]

@skshetry Do you need QA? Product-wise, LGTM except for the long flag name, but I don't have any better ideas for it.

yeah, make has -i/--ignore-errors and --keep-going/-k, which I am leaning towards.

I am still not sure about this PR. I do like the simplicity this execution order provides in terms of implementation, I prefer the existing execution order that tries to execute a subtree before moving on to the next one.

[skshetry]

Closing in favour of #9712.