[BUG] The fallback test only validates that the fallback URL is reached

[GingerGraham]

The fallback test only validates that the fallback URL is reached, but doesn't verify the specific retry attempt count. Since wget now has 4 retry attempts while curl has 3, the test mock that succeeds on attempt 3 (MOCK_CURL_CALLS > 2 or MOCK_WGET_CALLS > 2) will succeed on wget's IPv4 fallback attempt rather than testing the final default-settings fallback attempt.

Consider either:

1. Adjusting the test to fail the first 3 attempts (≤3 instead of ≤2) to properly test wget's 4th and final retry, OR
2. Documenting that this test validates fallback behavior but not the complete retry sequence

This would ensure the test thoroughly validates the retry logic for both tools.

Originally posted by @Copilot in #87 (comment)

[GingerGraham]

Plan: Fix Issue #88 — Fallback Test Retry Coverage

TL;DR

The test_download_release_fallback_on_primary_failure test has two problems: (1) the wget mock threshold is wrong (≤2 should be ≤3 since wget has 4 attempts vs curl's 3), and (2) the wget code path is never actually exercised because curl is checked first in download_release and both tools are co-mocked in the same test. Fix by splitting into separate curl and wget tests, forcing the wget code path via a command override, correcting the wget threshold, and adding inline comments to document the asymmetry.

No production code changes — install behavior is unchanged.

---

Retry Chain Reference

Tool	Attempt 1	Attempt 2	Attempt 3	Attempt 4
curl	IPv4 + primary URL	default + primary URL	default + fallback URL	—
wget	IPv4 + primary URL	default + primary URL	IPv4 + fallback URL	default + fallback URL

curl threshold: fail when MOCK_CURL_CALLS -le 2, succeed on call 3 ✓ (already correct)
wget threshold: fail when MOCK_WGET_CALLS -le 3, succeed on call 4 ✗ (currently ≤2, wrong)

---

Steps

Phase 1 — Fix existing test (tests/test_install.sh ~line 1286)

1. Rename test_download_release_fallback_on_primary_failure → test_download_release_curl_fallback_on_primary_failure to make it explicitly curl-specific.
2. Remove the wget mock from that test (dead code; curl always wins when both are available).
3. Add call-count assertion: assert output contains CURL_CALL_3 to prove the 3rd attempt (fallback URL) was reached, and that it contains the fallback URL string.
4. Update function call at bottom of file to use the new name.

Phase 2 — Add wget-specific test (tests/test_install.sh, after the curl test)

5. Add test_download_release_wget_fallback_on_primary_failure test.
6. Force wget code path by overriding command in the test script to return false for curl:

   command() { case "$1 $2" in "-v curl") return 1;; *) builtin command "$@";; esac; }
   

7. Mock wget to fail on MOCK_WGET_CALLS -le 3, succeed on call 4, creating the fake tarball.
8. Assert: output contains "falling back to tag archive", "archive/refs/tags/0.10.0", and WGET_CALL_4 (proving the 4th attempt was reached).
9. Add function call at bottom of file for the new test.

Phase 3 — Documentation in install.sh

10. Add inline comment block above the elif command -v wget section in download_release (around line 564) explaining:
    • curl uses 3 attempts: IPv4+primary → default+primary → default+fallback
    • wget uses 4 attempts: IPv4+primary → default+primary → IPv4+fallback → default+fallback
    • The extra wget step mirrors curl's IPv4-first preference for the fallback URL path

---

Relevant Files

• tests/test_install.sh — lines ~1286-1359: the three download_release tests; function call registrations at bottom
• install.sh — lines ~564-578: wget block in download_release; add comment above elif command -v wget

---

Verification

1. Run tests/run_tests.sh — all existing tests must pass
2. Confirm new test test_download_release_wget_fallback_on_primary_failure appears in test output and passes
3. Confirm renamed test test_download_release_curl_fallback_on_primary_failure passes
4. Manually verify tests/run_tests.sh output shows no regressions
5. ShellCheck: shellcheck --severity=warning --external-sources tests/test_install.sh — no new warnings

---

Decisions

• No production code changes: curl (3 attempts) and wget (4 attempts) behavior preserved exactly
• No retry count normalization: making them symmetric would change install experience (excluded per user requirement)
• Blended approach adopted: test fix + inline documentation (per user preference)
• Scope: only tests/test_install.sh and a comment in install.sh; no other files

---

Key Technical Constraint

Overriding command inside the run_install_test_script heredoc requires using builtin command as the fallback to avoid infinite recursion. The heredoc runs in a bash subshell so the override is isolated to that test.

[GingerGraham]

Released with v2.5.0