Instantiate validators at definition time

[ericproulx]

Instantiate validators at definition time and fix thread safety

Summary

Validators are now instantiated at definition time (in ParamsScope and ContractScope) rather than at request time via ValidatorFactory. This eliminates per-request object allocation overhead and allows expensive setup (option parsing, converter building) to happen once.

Because ParamsScope instances are now shared across all concurrent requests, the two instance variables previously mutated at request time (@index and @params_meeting_dependency) have been moved into a per-request ParamScopeTracker stored in Fiber[], eliminating a long-standing thread-safety race.

I18n translation is handled through a shared Grape::Util::Translation module. Validators that need interpolation parameters (e.g. LengthValidator, SameAsValidator) store a Hash { key: :length, min: 2, max: 5 } as their message. Exceptions::Base#translate_message dispatches on type — Symbol, Hash, Proc, or String — so translation with the correct locale always happens at error-raise time.

Changes

Core architecture

• ParamsScope#validate and ContractScope now store validator instances instead of option hashes in namespace_stackable[:validations]
• ParamsScope#coerce_type receives validations.extract!(:coerce, :coerce_with, :coerce_message) instead of the full hash
• Endpoint#run_validators takes a request: keyword arg, reads validators directly from inheritable_setting.route[:saved_validations], and short-circuits with return if validators.blank?
• Removed Endpoint#validations enumerator method
• Removed ValidatorFactory

Thread safety (ParamScopeTracker)

• @index and @params_meeting_dependency removed from ParamsScope
• Per-request ParamScopeTracker stored in Fiber[FIBER_KEY], set/restored around each request in Endpoint#run_validators
• AttributesIterator#store_indices writes current + ancestor indices; ParamsScope#full_name and #qualifying_params read from it

Shared translation module (Grape::Util::Translation)

• Extracted translate (with fallback locale logic) into Grape::Util::Translation, included by both Exceptions::Base and Validators::Base
• Exceptions::Base#translate_message now supports Hash messages: { key: :symbol, **interpolation_params } for deferred translation with interpolation

Validator base (Validators::Base)

• fail_fast? is now an explicit public method
• Moved validate!, message, options_key? to private
• Added private helpers: hash_like?, option_value, scrub, translate_message
• message accepts a block for lazy default generation

Validator optimizations — eagerly computed in initialize

• AllowBlankValidator: caches @value and @exception_message
• CoerceValidator: resolves type and builds converter at definition time; removes type, converter, valid_type?
• DefaultValidator: pre-builds @default_call lambda
• ExceptValuesValidator / ValuesValidator: validates proc arity; pre-builds call lambda
• LengthValidator: stores a Hash message for deferred i18n translation with interpolation
• RegexpValidator / SameAsValidator / PresenceValidator: cache exception messages
• AllOrNoneOfValidator, AtLeastOneOfValidator, MutuallyExclusiveValidator, ExactlyOneOfValidator: cache exception messages
• ContractScopeValidator: no longer inherits from Base; standalone class with (schema:) constructor

params: argument simplification

• Grape::Exceptions::Validation.new(params: [@scope.full_name(attr_name)]) simplified to params: @scope.full_name(attr_name) throughout, since Validation now coerces to array internally

Specs

• Replaced shared let_it_be(:app) with per-describe let(:app) blocks
• Added i18n tests for LengthValidator and SameAsValidator confirming request-time locale is used regardless of definition-time locale
• Added zh-CN translations for length and same_as messages

Test plan

• Run full test suite (bundle exec rspec)
• Verify no regressions in validation behavior
• Benchmark request throughput to confirm reduced per-request allocations

[github-actions]

Danger Report

No issues found.

View run

[ericproulx]

Missing UPGRADING notes. Working on it

[dblock]

Is there a tradeoff with things like @exception_message = message(:all_or_none) being always allocated? can they become class variables?

[ericproulx]

Is there a tradeoff with things like @exception_message = message(:all_or_none) being always allocated? can they become class variables?
message involves @option so it can't be a class variable.