fix(operators): reset retry budget per subscription to fix retry+repeat

[github-actions]

🤖 This PR was created by Repo Assist, an automated AI assistant.

Summary

Fixes retry(n) interacting incorrectly with repeat() (issue #712).

Root Cause

The retry generator (range(n) or infinite()) was created once at observable creation time and shared across all subscriptions:

# Before — generator created once, shared across resubscriptions
gen = range(retry_count)
return reactivex.catch_with_iterable(source for _ in gen)

catch_with_iterable stores iter(sources) at construction, so the iterator is shared. When repeat() resubscribes to retry(n), the iterator may already be fully consumed, causing catch_with_iterable to immediately call observer.on_completed() without subscribing to the source. This triggered repeat() again in a tight loop that produced no values after the nth subscription.

Fix

Move the generator creation inside the subscribe closure so each subscription gets a fresh retry budget:

def subscribe(observer, scheduler_=None):
    # Fresh generator on every subscription
    gen = infinite() if retry_count is None else range(retry_count)
    return reactivex.catch_with_iterable(source for _ in gen).subscribe(
        observer, scheduler=scheduler_
    )
return Observable(subscribe)

This matches the expected ReactiveX semantics: retry(n) means "on error, resubscribe up to n times per subscription attempt". Each invocation from repeat() is a new subscription and should start with a full retry allowance.

Changes

• reactivex/operators/_retry.py: wrap logic in subscribe closure
• tests/test_observable/test_retry.py: add regression test for retry(n) + repeat()
• changes.md: add changelog entry; also resolve lingering merge-conflict markers

Testing

All 10 tests pass (pytest tests/test_observable/test_retry.py). The new test test_retry_with_count_combined_with_repeat reproduces the exact scenario from issue #712 and verifies the fix.

Closes #712

Note

🔒 Integrity filter blocked 67 items

The following items were blocked because they don't meet the GitHub integrity level.

• #749 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #748 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #742 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #736 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #711 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #700 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #698 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• #634 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Missing tap operator #750 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Python 3.14 Support #737 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• AsyncIOScheduler.now returns a datetime in 1970. #734 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Offering help with Unit Tests #730 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Schedulers using different now() implementation breaks schedule_absolute() compatibility. #724 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Lack of concat_map / concat_all #723 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Concurrent Execution Not Working as Expected with RxPY ThreadPoolScheduler #713 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• switch_map() operator not listed in documentation #710 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• ... and 51 more items

To allow these resources, lower min-integrity in your GitHub frontmatter:

tools:
  github:
    min-integrity: approved  # merged | approved | unapproved | none

Generated by Repo Assist · ● 3.3M · ◷

[coveralls]

No base build to compare — repo-assist/fix-issue-712-retry-resubscription-2b9941042da6d10e into master