Summary

Grape::Exceptions::Validation is raised once per bad attribute on every failing request; Grape::Exceptions::ValidationArrayErrors wraps them and is raised once per validator. Both are caught internally by Endpoint#run_validators and their backtraces are never surfaced — they point into Grape's own validator stack, not user code, and have no diagnostic value.

Ruby's raise only captures a backtrace if the exception's backtrace ivar is nil. Pre-seeding it via set_backtrace(EMPTY_BACKTRACE) in initialize makes raise skip the C-level make_backtrace call entirely.

Backward compatibility

The outer Grape::Exceptions::ValidationErrors is untouched — its backtrace is preserved for anyone doing rescue_from ValidationErrors or shipping validation failures to error-reporting tools (Sentry, Rollbar, etc.). Only the two inner exceptions — which are private to the validator internals — have their backtrace capture skipped.

Perf

Measured on a synthetic benchmark at stack depth 40:

Savings scale linearly with stack depth since make_backtrace is O(frames). At representative prod stack depth (Rack + middleware chain + Grape + validators, typically 80-150 frames) this is closer to ~1-2 µs per raise.

A request failing 5 attributes across 2-3 validators raises ~7-8 inner exceptions → ~7-15 µs per failing request. Failure paths on public APIs get hammered by bots and misconfigured clients, so at scale this adds up.

Test plan

• Full RSpec suite passes locally (670 validation specs, 0 failures).
• CI green.

🤖 Generated with Claude Code