Test Coverage Gaps: Request Layer, API Methods, and Beyond

[SeanTAllen]

The JSON converter and utility tests from the original analysis are now implemented. This discussion covers what's still untested.

HTTP request actors

None of the request actors have any test coverage:

• JsonRequester — GET/POST/PATCH with JSON response parsing. Untested: correct URL construction, header setting (auth, content-type), response body parsing, HTTP error status handling.
• LinkedJsonRequester — Follows pagination links. Untested: Link header parsing into follow-up requests, redirect handling (301, 307), terminal page behavior.
• NoContentRequester — DELETE/PUT expecting 204. Untested: success on 204, failure on other status codes, response body ignored.
• CheckRequester — Status-based boolean (204=true, 404=false). Untested: both branches, unexpected status codes.

These are actors that do real HTTP, so testing them requires either mock HTTP servers or an abstraction boundary that can be substituted in tests.

High-level API methods

No methods on GitHub, Repository, Issue, Gist, PullRequest, or GistComment are tested. This includes:

• GitHub: get_repo(), get_gist(), create_gist(), get_user_gists(), get_public_gists(), search_issues(), get_org_repos()
• Repository: create_label(), delete_label(), create_release(), get_commit(), get_issue(), get_issues(), get_pull_request()
• Issue: create_comment(), get_comments()
• PullRequest: get_files()
• Gist: update_gist(), delete_gist(), get_revision(), fork(), get_forks(), get_commits(), star(), unstar(), is_starred(), create_comment(), get_comments()
• GistComment: update(), delete()

These methods construct URLs, set query parameters, and wire up result receivers. Testing them would verify correct URL construction and parameter passing without necessarily hitting the network — if the request layer were abstractable.

Result receivers

ResultReceiver[A], PaginatedResultReceiver[A], SearchResultReceiver[A], DeletedResultReceiver, and BoolResultReceiver are all untested. These handle the callback from request actors to user code.

Authentication

Credentials class and bearer token header construction are untested. Worth verifying the Authorization: Bearer <token> header is set correctly.

Search and pagination

SearchResults[A] pagination (prev/next page navigation) and SearchIssues request building are untested. The pagination link extraction is tested, but the pagination behavior (making follow-up requests) is not.

SSL

SSLContextFactory has no tests. Low priority since it's thin wrapper code, but worth noting.

Architectural barrier

The biggest barrier to testing the remaining code is that the request actors (JsonRequester, etc.) make real HTTP calls with no substitution point. Testing the high-level API methods and result receivers likely requires either:

1. Introducing a request abstraction that can be replaced with a test double (similar to how corral uses VCSBuilder)
2. Standing up a local mock HTTP server in tests

The JSON converter tests cover the data layer thoroughly. Everything above it — request construction, HTTP execution, response routing, and the public API surface — is currently untested.

[SeanTAllen]

Result receiver tests are done in #99. All five receiver types (DeletedResultReceiver, BoolResultReceiver, ResultReceiver[A], PaginatedResultReceiver[A], SearchResultReceiver[A]) now have coverage for success, failure, and converter-error paths — 14 tests total.

[SeanTAllen]

"High Level API" tests are done in #100

[SeanTAllen]

Completed by #101.

[SeanTAllen]

The remaining gap — search and pagination behavior tests — has been implemented in #102.

What's covered:

• 4 unit tests verifying the converter layer correctly parses Link headers and exposes pagination state (present vs absent) for both SearchResults and PaginatedList
• 4 mock HTTP server tests verifying the full pagination chain: fetch page → extract Link header → follow link → fetch next page → verify results

This covers both next_page() and prev_page() navigation for both SearchResults and PaginatedList, which were the last untested areas identified in this discussion.