Browse the repository at this point in the history

* test update

* fix array of objects unmarshaling

* update tests

* update tests

* update tests

* chore: revert cosmetic changes to deepobject.go

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

3 people authored

Browse the repository at this point in the history

Browse the repository at this point in the history

Browse the repository at this point in the history

Browse the repository at this point in the history

* fix: improve email validation using net/mail package

* fix: improve email validation using net/mail package

* fix: update test expectations for email validation behavior change

Invalid emails no longer persist their value after failed UnmarshalJSON
validation, since the new net/mail-based validation sets the value only
after successful parsing.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

3 people authored

Browse the repository at this point in the history

* feat(fix): bump gin version

* feat: bump gin to v1.10.1

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

3 people authored

Browse the repository at this point in the history

Browse the repository at this point in the history

…api-codegen#97) (oapi-codegen#98)

When an OpenAPI spec uses type: string, format: byte, the generated Go
code produces *[]byte fields. Previously the runtime treated []byte as
a generic []uint8 slice -- splitting on commas and parsing individual
integers -- instead of base64-encoding/decoding it as a single value.

This commit adds Type and Format string fields (matching the OpenAPI
spec type/format) to the Options structs for parameter binding and
styling functions. When Format is "byte" and the destination is
[]byte, the runtime base64-encodes (styling) or base64-decodes
(binding) the value as a single string.

New WithOptions functions and options structs:
- BindStringToObjectWithOptions + BindStringToObjectOptions
- StyleParamWithOptions + StyleParamOptions
- BindQueryParameterWithOptions + BindQueryParameterOptions

Extended existing options struct:
- BindStyledParameterOptions: added Type and Format fields

Existing functions delegate to WithOptions with zero-value options,
preserving backward compatibility for all current callers.

Encoding uses base64.StdEncoding (standard alphabet, padded) per
OpenAPI 3.0 / RFC 4648 Section 4. Decoding is lenient: it inspects
padding and URL-safe characters to select the correct decoder,
avoiding the silent corruption that occurs when RawStdEncoding
accepts padded input.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

Browse the repository at this point in the history

…) (oapi-codegen#96)

OpenAPI 3.0's deepObject style has undefined behavior for arrays.
Two conventions exist in the wild:

- Indexed: p[vals][0]=a&p[vals][1]=b (oapi-codegen's current format)
- Non-indexed: p[vals]=a&p[vals]=b (Swagger UI / Rails convention)

Swagger UI generates the non-indexed format, which previously failed
with "[field] has multiple values". This change makes UnmarshalDeepObject
accept both formats by detecting repeated query parameter keys and
expanding them into synthetic indexed entries (e.g. [vals][0], [vals][1])
before feeding them into the existing tree-construction and assignment
logic.

Marshaling (MarshalDeepObject) intentionally remains unchanged and
continues to emit the indexed format. The indexed format is
unambiguous, already consumed correctly by all known implementations,
and is what oapi-codegen's own generated clients expect. Changing it
to the non-indexed format would be a breaking change for consumers
that rely on the current wire format, with no practical benefit since
both formats are now accepted on the unmarshaling side.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

…gen#92)

Add BindRawQueryParameter which operates on the raw (undecoded) query
string instead of pre-parsed url.Values. For form/explode=false
parameters, this splits on literal commas before URL-decoding each part,
correctly preserving %2C as a literal comma in values.

Also adds findRawQueryParam internal helper, comprehensive tests, and
round-trip (serialize/deserialize) tests.

Deprecates BindQueryParameter, which cannot distinguish delimiter commas
from literal commas because url.Values pre-decodes %2C.

Fixes oapi-codegen#91

Note: oapi-codegen/oapi-codegen will need to update its generated code
to call BindRawQueryParameter (passing r.URL.RawQuery) once this change
is released.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

… (oapi-codegen#93)

BindQueryParameter treated all structs as key-value objects in the
non-exploded form path, causing types.Date and time.Time to fail with
"property/values need to be pairs". Scalar struct types that implement
Binder or encoding.TextUnmarshaler are now bound directly via their
interface methods instead of being routed to bindSplitPartsToDestinationStruct.

Also adds Binder implementation to types.Date so it self-identifies as a
scalar binding target.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

…ct (oapi-codegen#86)

Previously, the Binder interface was only checked within the struct case
of the type switch, causing primitive types (strings, ints, etc.) that
implement Binder to bypass their custom binding logic. This fix moves
the Binder check to the beginning of BindStringToObject, ensuring all
types implementing Binder use their custom Bind method, regardless of
their underlying type. The redundant check in the struct case has been
removed.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Viktor Stanchev <viktor@anchorlabs.com>

3 people authored

Browse the repository at this point in the history

Browse the repository at this point in the history

* feat: add support for echo v5

* fix: move echo v5 strict types to own module as peer directory

Move strictmiddleware/echo/v5/ to strictmiddleware/echo-v5/ with its
own go.mod requiring Go 1.25, so the root module stays at Go 1.20.

This avoids forcing all runtime users to upgrade to Go 1.25 just
because echo v5 support was added. The echo-v5 package is a separate
Go module with only the echo v5 dependency.

Also changes package name from `v5` to `echo` to follow Go convention
(matching how github.com/labstack/echo/v5 uses `package echo`).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* chore/update CI for nested modules

The echo-v5 strict middleware requires Go 1.25+, which is ahead of the
rest of the runtime module (Go 1.20). To support this as a separate
child module, the CI and build tooling needed to become multi-module
aware.

- Replace four separate workflow files (ci.yml, lint.yml, tidy.yml,
  generate.yml) with a single ci.yml that calls the shared reusable
  workflow from oapi-codegen/actions v0.5.0
- Update the root Makefile to iterate child modules via
  git ls-files '**/*go.mod', matching the pattern used by oapi-codegen
- Bump golangci-lint from v1.55.2 to v2.10.1 and update flags for v2
- Add tidy-ci target for the reusable workflow
- Add strictmiddleware/echo-v5/Makefile with a Go version guard that
  skips gracefully when running on Go < 1.25

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

3 people authored

Browse the repository at this point in the history

* feat: improve parameter handling

Add AllowReserved support for OpenAPI parameter serialization

Add the AllowReserved field to StyleParamOptions,
BindStyledParameterOptions, and BindQueryParameterOptions. When set to
true for query parameters, RFC 3986 reserved characters are left
unencoded in the serialized output, per the OpenAPI 3.x allowReserved
specification.

Changes:
- Add AllowReserved bool to all parameter option structs
- Implement escapeQueryAllowReserved for custom escaping that preserves
  reserved characters while still encoding spaces and control characters
- Thread allowReserved through internal helpers (stylePrimitive,
  styleSlice, styleStruct, styleMap, processFieldDict)
- Fix styleStruct recursive call to use StyleParamWithOptions instead
  of StyleParamWithLocation, which was silently dropping options

Partially resolves oapi-codegen/oapi-codegen#1342
Prerequisite for oapi-codegen/oapi-codegen#2183

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* feat: encode parameter names in styled output for RFC 3986 compliance

Add escapeParameterName helper that percent-encodes parameter names
in the prefix/separator construction of styleSlice, stylePrimitive,
and processFieldDict. This ensures characters like [] in parameter
names (e.g. user_ids[]) are properly encoded as %5B%5D in query
strings, per RFC 3986.

Note: MarshalDeepObject handles its own serialization independently
and does not yet encode parameter names in bracket notation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

…odegen#99) (oapi-codegen#100)

BindStyledParameterWithOptions passed the raw styled value directly to
BindStringToObject for primitive types, without first removing the style
prefix. This caused label (leading ".") and matrix (";param=") styles to
fail with parse errors for non-struct, non-slice destinations.

Route primitive values through splitStyledParameter — the same function
already used by the struct and slice paths — so the prefix is stripped
before binding.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Browse the repository at this point in the history

…codegen#68)

* Fix: deepObject return without assign on !required

* Add tests for deepObject required/optional query parameter binding

Exercise the required and optional code paths for deepObject-style
query parameters to verify no unintended side effects from PR oapi-codegen#68.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

3 people authored

Browse the repository at this point in the history

Browse the repository at this point in the history