[ericproulx]

Summary

Rack::Head was the first middleware in each endpoint's stack, so it only stripped bodies from responses that went through one. Two kinds didn't:

• a Rack app mounted with mount: the router calls it directly (config.app || build_stack), so a HEAD request got the app's body back, and raised Rack::Lint::LintError: Response body was given for HEAD request under lint!;
• a middleware inserted with insert 0 (the README's CustomOverwriter example), which sat outside Rack::Head.

The router's own 404 had to strip its body itself (#2958) for the same reason.

This puts a single Rack::Head around the router in Grape::API::Instance, inside Rack::Lint so Lint checks the response that is served, and drops the router's 404 special case. A request that cascades through several routes goes through one Rack::Head instead of one per route tried.

This came up while looking at where Rack::TempfileReaper should go in #3041: the same config.app || build_stack gap means a per-endpoint middleware never runs for a mounted Rack app.

Contract change

An endpoint's stack now starts with Grape::Middleware::Error, so:

• insert_before Rack::Head / insert_after Rack::Head raise No such middleware to insert before: Rack::Head when the API compiles;
• a middleware inserted by index lands one place further in: insert 1 now goes inside Grape::Middleware::Error instead of between it and Rack::Head;
• a middleware inserted with insert 0 sees the full body of a HEAD response, as one added with use always has.

Covered in UPGRADING.

Test plan

• New specs: HEAD to a mounted Rack app, and HEAD answered by an insert 0 middleware, both come back without a body. Both fail against master's lib.
• bundle exec rake: RuboCop clean, 3158 examples, 0 failures.
• gemfiles/rack_2_2.gemfile: 3162 examples, 0 failures.
• CI green.

🤖 Generated with Claude Code

[github-actions]

Danger Report

No issues found.

View run