mirror of
https://github.com/hanzoai/gateway.git
synced 2026-11-02 15:19:24 +00:00
zeekayandHanzo Dev
53350699b6
Every import in this module now resolves to github.com/hanzoai/gateway/v2.
The 34 github.com/krakend/* + github.com/luraproject/lura modules that
supplied the engine and its components are vendored under internal/ and
their 91 packages rewritten onto our own path.
Why fold rather than mirror into 34 hanzoai/* repos:
- It removes the exposure outright. 34 rows leave 34 rows of go.sum to
reason about; zero modules leave nothing. `go list -m all` and go.sum
now contain no krakend/lura entry at all.
- Cross-references cannot go wrong. 28 of the 34 import lura (97 files
inside lura self-reference it). Mirroring means rewriting every one of
those to our lura, and a single miss silently links a SECOND copy of
lura into the binary. With one module that is structurally impossible.
- It creates no duplicates. These are lura-coupled gateway glue with no
consumer outside this binary; publishing 34 repos would be 34 new
things to maintain for zero second caller. internal/ makes that
guarantee mechanical — Go forbids importing it from outside this
module, so none of it can become a competing public package. Verified
no repo outside gateway imports any hanzoai/gateway/v2 subpackage.
Layout mirrors provenance so an upstream re-sync stays a diff:
luraproject/lura/v2 -> internal/lura (the engine, 20 pkgs)
krakend/krakend-<x> -> internal/plugin/<x> (28 components)
krakend/<x> -> internal/pkg/<x> (6 standalone libs)
This commit is PURELY MECHANICAL: 196 of the 197 vendored files are
byte-identical to their upstream original once the import rewrite is
applied. The single exception is bloomfilter's subpackage, which upstream
names `krakend` — the one package name that could not survive as-is; it
becomes internal/pkg/bloomfilter/register. Hand edits land separately in
the next commit so they are readable on their own.
Per `go mod vendor` semantics, dependencies' _test.go files are not
vendored. They test upstream code at a pinned version that we do not
modify, several cannot run at all from a module distribution (they need
TLS certs and compiled .so plugins upstream generates but never ships),
and one is a 100k-request stress test that exhausts the ephemeral port
range and fails every package running beside it. Fold integrity is proven
by the byte-diff above instead, which is the stronger check.
Preserved deliberately — each of these is behaviour, not branding:
- the `github.com/devopsfaith/krakend*` extra_config namespace strings
(35 of them). These are config-surface keys, not import paths;
renaming one silently stops that component finding its config. The
jose validator namespace backs `auth/validator` on 212 endpoints.
- core.KrakendVersion / KrakendUserAgent / the X-KRAKEND consts, whose
VALUES rebrand.go overwrites — that seam is unchanged.
- the KRAKEND_ koanf env prefix (a hardcoded upstream const).
- every vendored module's LICENSE (34 files, Apache-2.0/MIT).
The Makefile/CI -X paths move with the engine. An -X naming a symbol that
does not exist is silently ignored by the linker, so a stale path here
would have shipped an "undefined" version in the backend User-Agent and
X-Api-Version with a perfectly green build. Verified live:
`gateway version` reports 2.13.0, not undefined.
Verification (GOWORK=off CGO_ENABLED=0, both tag modes):
go build ./... OK go build -tags legacy ./... OK
go vet ./... clean go vet -tags legacy ./... clean
go test ./... 6/6 ok go test -tags legacy 5/5 ok
tests/ integration harness 47 fixtures pass
gateway check -c configs/{hanzo,lux}/gateway.json -> Syntax OK!
go list -m all | grep -iE 'krakend|lura' -> empty
go mod verify -> all modules verified
Co-authored-by: Hanzo Dev <[email protected]>
65 lines
2.0 KiB
Go
65 lines
2.0 KiB
Go
package jose
|
|
|
|
import (
|
|
"github.com/hanzoai/gateway/v2/internal/lura/config"
|
|
"github.com/hanzoai/gateway/v2/internal/lura/logging"
|
|
)
|
|
|
|
// Rejecter defines the interface for the components responsible for rejecting tokens.
|
|
type Rejecter interface {
|
|
Reject(map[string]interface{}) bool
|
|
}
|
|
|
|
// RejecterFunc is an adapter to use functions as rejecters
|
|
type RejecterFunc func(map[string]interface{}) bool
|
|
|
|
// Reject calls r(v)
|
|
func (r RejecterFunc) Reject(v map[string]interface{}) bool { return r(v) }
|
|
|
|
// FixedRejecter is a rejecter that always returns the same bool response
|
|
type FixedRejecter bool
|
|
|
|
// Reject returns f
|
|
func (f FixedRejecter) Reject(_ map[string]interface{}) bool { return bool(f) }
|
|
|
|
// RejecterFactory is a builder for rejecters
|
|
type RejecterFactory interface {
|
|
New(logging.Logger, *config.EndpointConfig) Rejecter
|
|
}
|
|
|
|
// RejecterFactoryFunc is an adapter to use a function as rejecter factory
|
|
type RejecterFactoryFunc func(logging.Logger, *config.EndpointConfig) Rejecter
|
|
|
|
// New calls f(l, cfg)
|
|
func (f RejecterFactoryFunc) New(l logging.Logger, cfg *config.EndpointConfig) Rejecter {
|
|
return f(l, cfg)
|
|
}
|
|
|
|
// NopRejecterFactory is a factory returning rejecters accepting all the tokens
|
|
type NopRejecterFactory struct{}
|
|
|
|
// New returns a fixed rejecter that accepts all the tokens
|
|
func (NopRejecterFactory) New(_ logging.Logger, _ *config.EndpointConfig) Rejecter {
|
|
return FixedRejecter(false)
|
|
}
|
|
|
|
// ChainedRejecterFactory returns rejecters chaining every rejecter contained in tne collection
|
|
type ChainedRejecterFactory []RejecterFactory
|
|
|
|
// New returns a chainned rejected that evaluates all the rejecters until v is rejected or the chain
|
|
// is finished
|
|
func (c ChainedRejecterFactory) New(l logging.Logger, cfg *config.EndpointConfig) Rejecter {
|
|
rejecters := []Rejecter{}
|
|
for _, rf := range c {
|
|
rejecters = append(rejecters, rf.New(l, cfg))
|
|
}
|
|
return RejecterFunc(func(v map[string]interface{}) bool {
|
|
for _, r := range rejecters {
|
|
if r.Reject(v) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
})
|
|
}
|