fix: resolve retry logic inconsistencies in github_api_tools

[eren-karakus0]

Summary

This PR fixes 5 related bugs in gittensor/utils/github_api_tools.py, all involving retry logic, type annotations, and log message accuracy.

Changes

1. Off-by-one in get_pull_request_file_changes() retry guard (lines 278, 288)

Bug: The condition attempt < max_attempts is always True inside a range(max_attempts) loop, causing an unnecessary time.sleep() after the last failed attempt before the function returns. This adds up to 20 seconds of wasted delay on total failure.

Fix: Changed to attempt < (max_attempts - 1) to match the pattern used in execute_graphql_query() and get_github_graphql_query().

2. Uncapped exponential backoff in execute_graphql_query() exception handler (line 352)

Bug: The HTTP status code retry path correctly caps backoff at 30s with min(5 * (2**attempt), 30), but the RequestException handler computes backoff_delay = 5 * (2**attempt) without a cap. At attempt 6, this produces a 320-second (5+ minute) sleep.

Fix: Added min(..., 30) cap to match the status code path and other retry functions.

3. Incorrect docstring default value in execute_graphql_query() (line 316)

Bug: Docstring says max_attempts: Maximum retry attempts (default 6) but the actual function signature is max_attempts: int = 8.

Fix: Updated docstring to (default 8).

4. Wrong type annotation in load_miners_prs() (line 595)

Bug: prs: Dict = pr_data.get('nodes', []) — the nodes key returns a JSON array (Python list), not a dict.

Fix: Changed annotation to prs: List.

5. Misleading log message in should_skip_merged_pr() (line 498)

Bug: The condition merged_dt < lookback_date_filter filters PRs merged before the lookback window, but the log says "merged within {PR_LOOKBACK_DAYS} day lookback window", implying the opposite.

Fix: Changed to "merged before {PR_LOOKBACK_DAYS}-day lookback window".

Test Evidence

Updated and added tests in tests/utils/test_github_api_tools.py:

• Updated TestFileChangesRetryLogic.test_gives_up_after_three_attempts: sleep count assertion corrected from 3 to 2 (no sleep on last attempt)
• Updated TestFileChangesRetryLogic.test_exponential_backoff_timing: expected delays corrected from [5, 10, 20] to [5, 10]
• Added TestExecuteGraphQLQueryRetryLogic class (3 tests):
  • test_exception_backoff_capped_at_30s — verifies backoff cap in exception handler
  • test_status_code_backoff_capped_at_30s — verifies backoff cap in status code handler
  • test_successful_request_no_retry — verifies no retry on success
• Added TestShouldSkipMergedPr class (2 tests):
  • test_log_message_says_before_for_old_pr — verifies corrected log wording
  • test_recent_pr_not_skipped_by_lookback — verifies recent PRs pass lookback check

All existing tests continue to pass. Ruff linting passes with no warnings.

$ ruff check gittensor/utils/github_api_tools.py tests/utils/test_github_api_tools.py
All checks passed!

[anderdc]

I commented on what I like, everything else, reset/revert it

[eren-karakus0]

Done, reverted all changes except the two you approved:

1. execute_graphql_query backoff cap (min(5 * (2**attempt), 30))
2. prs: List type annotation fix

Kept the corresponding tests for these two changes, reverted everything else including the get_pull_request_file_changes retry boundary change and max_attempts docstring.

[anderdc]

sorry took a while to get back to this, resolve conflicts then it'll be good to merge thanks

[eren-karakus0]

Rebased on latest test branch - conflicts resolved. Ready to merge.

[anderdc]

hmm another conflict, resolve so I can get this in today

I was doing reviews on other PRs so that caused a conflict here

[eren-karakus0]

Rebased on latest test branch - conflicts resolved. Ready to merge.

[anderdc]

LGTM