Declutter jwt_spec.rb

[anakinj]

Description

First pass at making spec/jwt/jwt_spec.rb about one thing: the JWT.encode / JWT.decode public surface. 951 lines and 130 examples down to 785 and 118, with no coverage lost.

Move the custom algorithm contract out. ~125 lines describing what a class must implement to be usable as an algorithm — that is the JWT::JWA::SigningAlgorithm contract, not the encode/decode surface. Now spec/jwt/jwa/signing_algorithm_spec.rb, mirroring the lib layout. The block depended only on the shared payload let, so this is a pure move: the suite reports the same 1246 example descriptions before and after, the only diff being the group header.

Trim the duplicated iat semantics. The verify_iat block re-tested a float iat, an integer iat, one second of clock drift, and a leeway covering that drift — all four already covered against the claim itself in claims/issued_at_spec.rb. The only thing this file can say that the claim spec cannot is that the decode options reach the verifier, so that is what it now asserts, plus the default of leaving iat unverified, which nothing covered before.

Drop an unused let in the issuer claim context.

Dismantle the data fixture. A 29 entry hash mixing symbol and string keys, holding an HMAC secret, hardcoded tokens and key fixtures all together, plus six '' placeholders that existed only so before(:each) could overwrite them by mutating a memoized let. Each value now lives where it is used: the HMAC and RSA loops iterate a map of algorithm to expected token instead of indexing the fixture by the loop variable, the ECDSA and PSS loops build their token in a let, and the four malformed token strings are inline at their single use each.

Two values were not doing what they appeared to. A keyfinder example resolved its key by using the decoded payload as an index into the shared hash, and the multi-key context passed the hardcoded HS256 token string as an HMAC secret. Both now say what they mean.

What did not move, and why

I had planned to pull the x5c block into x5c_key_finder_spec.rb, but it doubles out JWT::X5cKeyFinder entirely — it tests that JWT.decode routes the x5c option to the finder, which is squarely the decode surface. Moving it into a spec whose subject it mocks would have made things worse.

The same reasoning applies more broadly than I first expected: nearly everything left goes through JWT.encode / JWT.decode and legitimately belongs here. The remaining clutter is internal — chiefly the 29-entry data fixture hash with mixed symbol and string keys, six '' placeholders that exist only to be overwritten by before(:each), and the mutation of a memoized let. That is worth a follow-up, but it is a different change from this one.

Verification

• 644 examples, 0 failures. RuboCop clean.
• The data removal verified the same way: identical example descriptions across the whole suite before and after, so nothing was added, removed or renamed.
• The move slice verified by diffing the full --dry-run --format doc output: identical example descriptions, 1246 before and after.

Checklist

Before the PR can be merged be sure the following are checked:

• There are tests for the fix or feature added/changed
• A description of the changes and a reference to the PR has been added to CHANGELOG.md — skipped, spec-only with no effect on the published gem