Fullphysio / brevo_api / 35141909443

99%

push

github

Assert the generated layer instead of smoke-testing it

The generated decode tier checked only that toJson() returned a Map and the
mock tier asserted nothing at all, so a model reading the wrong key or a
resource dropping a query parameter passed both. Decoding now round-trips:
toJson() must reproduce the payload it decoded, and no known field may go
missing. Each mock case reads WireMock's request journal back and asserts the
method, path and exact set of query keys it received, and every declared
query parameter is passed so the tier exercises query encoding across all 291
operations rather than the twelve whose mappings happen to match on one.

One deadline now covers a whole attempt rather than its response headers
alone; package:http applies no read timeout, so a server that answered and
then stalled mid-body could hang a caller indefinitely. A body that is not
valid UTF-8 raises BrevoDecodeException instead of letting a FormatException
escape the typed hierarchy, and a path parameter that is neither a string nor
a number is rejected rather than stringified into a request addressing the
wrong resource.

The resolver stops on a schema or parameter naming a type it does not know,
which is what its own documentation always claimed; an absent type stays
"any". The path-template normaliser and the HTTP-method set had been copied
into four tools that must agree on the same operation keys, and now come from
tool/operation_key.dart. Generated output is byte-identical throughout.

CI runs with contents: read, skips the Coveralls upload on fork pull requests
whose token cannot post, and gates publishing on a pub.dev environment so a
tag alone cannot release. CLAUDE.md described a generated layout that never
existed, and the README overstated which parts of the retry policy the
captured fixtures pin.

8 of 8 new or added lines in 2 files covered. (100.0%)

401 of 407 relevant lines covered (98.53%)

3.29 hits per line