feat(embedding): allof embedding structs impl#6
Merged
zdunecki merged 45 commits intolivesessionfrom Mar 2, 2026
Merged
Conversation
…oapi-codegen#2197) This is a prototype implementation of a future versions of oapi-codegen. It's almost a full rewrite, heavily inspired by previous code, and lessons learned. - much more flexibility in configuring generated code - move from kin-openapi to libopenapi to support 3.1 and 3.2 specs - webhook support - callback support - incompatible codegen changes to aggregate types (allOf, anyOf, oneOf) - many existing codegen bugs around schemas fixed
oapi-codegen#2198) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
* Add output filtering Fixes oapi-codegen#2200 Add output filtering like in V2, and support fetching specs via HTTP. * Fix lint issue --------- Co-authored-by: Marcin Romaszewicz <mromaszewicz@nvidia.com>
Co-authored-by: Marcin Romaszewicz <mromaszewicz@nvidia.com>
Thanks to Speakeasy and LivePeer for their support over the years!
This reverts commit 17ba42e.
This reverts commit db8abb8.
…security] (oapi-codegen#2198)" This reverts commit 98e5d1a.
…bopenapi (oapi-codegen#2197)" This reverts commit 5206145.
As a way to keep updates to the CI pipelines more straightforward, we can extract this out to a shared, versioned (and updated by Renovate) Action. We can make sure to keep our binary builds going, but only for the current versions of Go - other build failures will be caught by our shared Action.
…urity] (oapi-codegen#2207) Co-authored-by: renovate[bot] <29139614+renovate[bot]@users.noreply.github.com>
Co-authored-by: Yuichiro Tsuji <yui.tsuji@mbk-wellness.co.jp>
Signed-off-by: Richard Kosegi <richard.kosegi@gmail.com> Co-authored-by: Marcin Romaszewicz <mromaszewicz@users.noreply.github.com>
…api-codegen#1438) Co-authored-by: Ula <ulad@users.noreply.github.com> Co-authored-by: Marcin Romaszewicz <mromaszewicz@users.noreply.github.com>
…ers are exist in Iris strict server. (oapi-codegen#1411) Co-authored-by: Marcin Romaszewicz <mromaszewicz@users.noreply.github.com>
…gnature (oapi-codegen#1419) * hotfix: - Bump fiber version to 2.52.0 - Adopted middleware template for fiber to handle new GetReqHeader() method, that has changed signature in fiber 2.50.0 (https://github.com/gofiber/fiber/releases/tag/v2.50.0) * Use latest fiber, and fix go deps The latest Fiber requires Go 1.24, therefore, we have to increase the version in the modules which use it. This is constrained to tests and examples, so it doesn't affect the main repo. Go 1.24 can't compile the version of golang.org/x/tools which we were using, so update that as well. * fix: use valueList[0] for fiber header IsPassThrough case The merge with upstream/main resolved a conflict in the fiber middleware template's header IsPassThrough handler by taking upstream's version, which still used the old single-string `value` variable. This is incorrect because fiber 2.50.0+ changed GetReqHeaders() to return map[string][]string. Fix by using valueList[0] to match the rest of the header handling block. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Tidy up modules This seems to be about the minimal set of changes to have everything build and test cleanly. * fix: add Go 1.24 version guards to Makefiles The internal/test and examples modules now require Go 1.24+ in their go.mod files, but their Makefiles lacked version guards, causing CI failures on Go 1.22 and 1.23. Add execute-if-go-124 guards matching the pattern used by other Go 1.24+ modules. Also bump the strict-server/stdhttp Makefile guard from 1.22 to 1.24. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Alexey Boltunov <apboltunov@mts.ru> Co-authored-by: Marcin Romaszewicz <marcinr@gmail.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fixed duplicate type names * add typename dedup functions * Fixup: use full-word suffixes and add regression test for issue oapi-codegen#200 - Rename auto-dedup suffixes to use full words: Parameter, Response, RequestBody (instead of Param, Resp, ReqBody) - Add internal/test/issues/issue-200/ with spec, config, generated code, and a compile-time regression test that instantiates every expected type Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Gate duplicate type name resolution behind output-options config flag Add ResolveTypeNameCollisions bool to OutputOptions and the JSON schema. When false (the default), the codegen errors on duplicate type names as before. When true, FixDuplicateTypeNames auto-renames colliding types. Also cleans up ComponentType: removes unused constants, improves doc. 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>
…egen#2212) * Fiber deps in their own modules Move the fiber dependences to their own modules, to contain Go 1.24 to the minimal subset of our code. Update make rules to exclude Go 1.24 tests on Go < 1.24 Update Readme with minimum Go version for each router. * Update Fiber in test issue Contain Fiber to its own module in test for Issue 1469
We only need to run lint on a single version of Go, as the output isn't Go runtime dependent, however, we need to be careful to run it in an environment where the Go runtime isn't newer than what the golangci-lint executable was built with, because it panics on some kind of version check in that case. Resolves: oapi-codegen#2214
…api-codegen#2211) * support specifying package alias * update test * fix local package * simplify
* fix: escape quoted media type directives fixes oapi-codegen#1529 * squashme: expanded tests to cover iris, fiber & echo tests * squashme: added makefile to fiber test directory, fixing root level tests * squashme: actually fixed the generate issues with fiber tests * squashme: skip fiber test for go versions <1.24 * squashme: make actually uses the execute-if-go-124 function
…2220) Some real-world OpenAPI specs (e.g. Keycloak) reuse the same path parameter more than once in a single URI, such as: /clients/{client-uuid}/roles/{role-name}/composites/clients/{client-uuid} Previously this caused a "has 4 positional parameters, but spec has 3 declared" error because SortParamsByPath compared raw URI placeholder count against unique declared parameters. Fix this by deduplicating the path parameters extracted from the URI (preserving first-occurrence order) before matching them against the spec declared parameters. This is the right level to fix the issue rather than scattering dedup logic across template helpers. Fixes oapi-codegen#1574 Supersedes oapi-codegen#2175 Co-authored-by: Junior Rantila <junior.rantila@gmail.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Response and RequestBodies need to include any additional types into model generation. This is an ancient bug, from the very first days of oapi-codegen. Extend the issue-200 spec with inline objects containing additionalProperties inside components/responses and components/requestBodies. The test confirms the codegen produces named types (Bar_Pagination, Bar_Metadata) for these inline schemas. Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…api-codegen#2225) * test: add regression test for issue oapi-codegen#2190 Add a minimal reproduction for invalid generated code when reusing response components. The generated VisitGetTestResponse method attempts []byte(response) on a struct type, which does not compile. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: generate correct type for $ref text responses in strict server (oapi-codegen#2190) When a response is defined as a $ref to a component response with text/plain content type (and no headers), the strict-server template generated a struct-embedding type whose Visit method tried to call []byte(response) on a struct, which failed to compile. The root cause was the revert of PR oapi-codegen#1132 (commit 891a067), which had originally fixed this by making all text responses string types. That PR was reverted because it broke text responses with headers (oapi-codegen#1676), which require the struct form with a Body field. The fix extends the existing multipart special case in Branch 1A of the strict-interface template to also cover text responses without headers. This generates a named type alias (e.g. `type GetTest401TextResponse UnauthorizedTextResponse`) instead of a struct embedding, so []byte(response) compiles and works correctly. Text responses with headers continue to go through Branch 1C (struct with Body + Headers fields), so oapi-codegen#1676 is unaffected — confirmed by verifying no diff in internal/test/issues/issue-1676/ping.gen.go. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…en#2222) Fixes oapi-codegen#2116 Strict server handlers unconditionally decoded the request body, causing an EOF error when no body was sent for operations with optional request bodies (requestBody.required defaults to false in OpenAPI 3.0). For optional JSON bodies, the generated handlers now treat io.EOF from the decoder as "no body provided" and leave request.Body as nil rather than returning an error. For optional text bodies, an empty read is similarly treated as absent. Required bodies retain the original unconditional error handling. All five strict server templates are updated: stdhttp (used by chi and gorilla), echo, gin, fiber, and iris. Two new test operations (RequiredJSONBody, RequiredTextBody) with required: true are added to the strict server test schema to verify both code paths are generated correctly. Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…-codegen#2223) * feat: add configurable type mapping for OpenAPI primitive types Supersedes oapi-codegen#428, since the original author's branch is gone and his PR only exists in the ref log. Added a TypeMapping configuration that lets users override the default OpenAPI type/format to Go type mappings. User-specified mappings are merged on top of built-in defaults, so only overrides need to be specified. For example, mapping string/date-time to a civil.DateTime or changing the default integer type to int64. Changes: - New TypeMapping, FormatMapping, SimpleTypeSpec types with Merge() and Resolve() methods (pkg/codegen/typemapping.go) - TypeMapping field added to Configuration struct - Merged type mapping stored on globalState, initialized in Generate() - Replaced hardcoded switch statements in oapiSchemaToGoType() with map-based lookups via globalState.typeMapping - Updated configuration-schema.json with type-mapping property and reusable $defs for format-mapping and simple-type-spec - Unit tests for merge, resolve, and default mapping completeness Co-Authored-By: natsukagami <natsukagami@gmail.com> Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Move TypeMapping to OutputOptions * Add example for type-mapping * Address PR feedback Move type-mapping examples to output-options --------- Co-authored-by: natsukagami <natsukagami@gmail.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…api-codegen#2172) As noted in oapi-codegen#2171, this is a workaround for yaml/go-yaml#76.
Otherwise the sentence ends abruptly.
* feat: add Valid() method to generated enum types Generate a Valid() bool method on each enum type that returns true when the receiver matches one of the defined enum constants and false otherwise. This lets callers validate enum values at runtime with a simple method call instead of hand-writing switch statements. This is default-on because it only adds a new method to an already generated type -- existing code that does not call Valid() is completely unaffected, so this should be very unlikely to break anything. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Reduce complexity in enum validation It turns out we don't need a sorted map of names, since we have the same thing already present on the template context in a different way. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Gaiaz Iusipov <g.iusipov@gmail.com>
Don't automatically review each PR, users can request a code review via tagging @greptileai Exclude generated files from code reviews. It's pointless.
This fixes a regression from 59fb0e8.
As a fix similar to c4ba545, we had cases where the global `prefer-skip-optional-pointer` could lead to not nil checking when we should have been. The previous behaviour would lead to incorrectly converting i.e. `user_ids=nil` to `user_ids[]=`, which is not correct. To do this, we can add the same helper functions that we have elsewhere.
…egen#2232) (oapi-codegen#2235) When x-oapi-codegen-extra-tags was placed on the schema inside a parameter definition (rather than on the parameter itself), the extension was silently ignored. This happened because GenerateParamsTypes only read parameter-level extensions, not schema-level ones. Merge extensions from both the parameter and its schema when building the Params struct, with parameter-level extensions taking precedence. Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
As a follow-up to oapi-codegen#2185, we can make sure we generate a Nullable type if required to do so.
Fixes oapi-codegen#2091 The `omitempty` JSON tag was not being added to optional nullable fields. The condition `!p.Nullable && shouldOmitEmpty` explicitly prevented any nullable field from receiving `omitempty`, even when the field was optional (not required). This contradicted the documented behavior. The fix removes the `!p.Nullable` guard so nullable fields follow the same `omitempty` rules as non-nullable fields. The special-case exception for the `nullable-type` output option is no longer needed since the logic is now uniform. Note: `x-go-type-skip-optional-pointer: true` on a nullable field suppresses the pointer (generating `string` instead of `*string`) but still correctly receives `omitempty` when the field is optional. Whether skip-optional-pointer should be allowed to suppress the pointer on nullable fields is a separate concern. Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
So it's clearer when the PR is for `examples/` or `internal/test/`, similar to regular PRs.
So they don't get pulled into Release Drafter releases.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
15 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.