Stop building a backtrace in the default rescue handler

[ericproulx]

Summary

Middleware::Error#default_rescue_handler renders any exception matched by rescue_from :all, or by rescue_from SomeError given without a block, a common way to turn unexpected errors into 500s. It passed backtrace: exception.backtrace to the error payload. Exception#backtrace builds the whole backtrace as Strings, which at a request's stack depth is the dearest part of rendering the error, and the result was then discarded unless the API had asked for rescue_from ..., backtrace: true.

#resolved_backtrace already falls back to original_exception.backtrace when a backtrace is wanted, and the handler passes original_exception, so the eager read goes. What an API that asked for the backtrace renders is unchanged.

Benchmarks

Median of 7 interleaved subprocess rounds against master, Ruby 4.0.6, no JIT:

response	delta
500 from rescue_from :all (no block)	+92.8%
500 from a rescue_from :all block calling error! (control, does not use the default handler)	+0.6% (noise)

Not changed

Formatter builds its payload the same way in two places, for an InvalidFormatter and for a custom parser raising a non-Grape error, but a spec (formatter_spec.rb:509) pins the backtrace of the payload it throws, and neither path is hot: Grape's own parsers turn a parse failure into InvalidMessageBody, which does not go through that branch (a malformed JSON body measured +0.1% with both changed). They are left as they are.

Behaviour

Byte-identical to master over a 30-case matrix of rescue_from :all APIs, with and without backtrace: true and original_exception: true, in JSON, txt and XML, for raised Ruby and Grape exceptions and malformed bodies: 18 of the responses render a backtrace, identical once paths are normalized. The 228-case error matrix from #2929 is unchanged too.

Test plan

• Full RSpec suite passes locally.
• RuboCop clean.
• Mutation-checked: also dropping original_exception: from the handler, so no backtrace is available when one is asked for, fails 5 specs.
• CI green.

🤖 Generated with Claude Code

[github-actions]

Danger Report

No issues found.

View run

[ericproulx]

Closing in favour of #2936, which combines #2922–#2934 into one PR, re-benchmarked as a whole against master. The write-up here (behaviour matrix, mutation results) still describes this part of the change.