Coerce through dry-types' non-raising call

[ericproulx]

Summary

DryTypeCoercer#call coerced with @coercer[val] and rescued Dry::Types::CoercionError for a value it could not coerce. The raise is what cost: dry-types' CoercionError.handle re-raises with backtrace: exception.backtrace of the error underneath (the ArgumentError from Integer(), Float(), Date.parse, …), which builds that whole backtrace as Strings. At request depth a rejected value cost about 30 µs, where an accepted one costs a few hundred nanoseconds.

It is paid by:

• every types: [Integer, String] param given a string, since each coercer tried before the one that accepts it fails first;
• every 400 for a mistyped value (?id=abc for an Integer);
• invalid dates, numbers and booleans.

dry-types' documented block form of Type#call ("When a block is passed, call will never throw an exception on failed coercion, instead it will call the block") reports a failure by calling the block, so the coercer now passes one that answers InvalidValue. The rescue stays for anything that raises regardless.

A rejected Integer at a request's stack depth: 31.2 → 4.4 µs.

Benchmarks

Median of interleaved subprocess rounds against master, Ruby 4.0.6 (7 rounds without JIT, 5 with YJIT):

request	no JIT	YJIT
GET with types: [Integer, String] given abc, plus an Array[Integer] with coerce_with	+114.7%	+195.1%
400 for ?id=abc on requires :id, type: Integer	+49.1%	+95.6%
GET with valid query params (control)	−1.3% (noise)
POST with a valid JSON body (control)	+1.1% (noise)

Behaviour

Byte-identical to master over a 2,214-case coercer matrix, calling Types.build_coercer directly: 27 types (every primitive, Numeric, Grape::API::Boolean, TrueClass/FalseClass, Hash, Array, Array[T], Set[T], multiple-type lists including nested collections), each coerced and strict, × 41 inputs (valid, malformed, blank, nil, arrays, hashes, symbols, invalid bytes). 1,640 of those cases are rejections, so the changed path is the one exercised. The request-level coercion and validation matrices are unchanged too.

Test plan

• Full RSpec suite passes locally.
• RuboCop clean.
• Mutation-checked: a failed coercion answering nil fails 37 specs; answering the raw value fails 29. Reverting to the raising call changes nothing observable, which is the point; the benchmark pins that.
• CI green.

🤖 Generated with Claude Code

[github-actions]

Danger Report

Errors

• One of the lines below found in CHANGELOG.md doesn't match the expected format. Please make it look like the other lines, pay attention to version numbers, periods, spaces and date formats.

Markdowns

* [#XXXX](https://github.com/ruby-grape/grape/pull/XXXX): Coerce through dry-types' non-raising call so a rejected value no longer builds a backtrace - [@ericproulx](https://github.com/ericproulx).
does not include a pull request link

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.