Recover hardcoded host in Go HTTP-client routes (v101)

internal/cachecov/coverage_test.go

@@ -132,6 +132,7 @@ var versionCoverage = map[int][]string{
 	98: {"TestKtComplexity_ScalingLoopDepth_ConstantRangeDiscounted", "TestKtComplexity_ScalingLoopDepth_VariableRangeNotDiscounted", "TestKtComplexity_ScalingLoopDepth_ConstantIteratorReceiverDiscounted", "TestKtComplexity_ScalingLoopDepth_VariableIteratorReceiverNotDiscounted", "TestKtComplexity_ScalingLoopDepth_ConstantOuterScalingInner", "TestKtComplexity_ScalingLoopDepth_ConstantInnerScalingOuter", "TestKtComplexity_ScalingLoopDepth_InfiniteLoopDiscounted", "TestKtComplexity_ScalingLoopDepth_ConditionalWhileNotDiscounted", "TestKtComplexity_ScalingLoopDepth_AbsentWithoutLoops", "TestKtComplexity_LoopCountAndCyclomaticUnchangedByBounding", "TestGolden"}, // Kotlin bounded-loop discounting: scaling_loop_depth joins the Go/Python/TS convention
 	99: {"TestKtComplexity_CallsInScalingLoop_ConstantExcluded", "TestKtComplexity_CallsInScalingLoop_InfiniteLoopCallsRetained", "TestKtComplexity_CallsInScalingLoop_ScalingRetained", "TestKtComplexity_CallsInScalingLoop_MixedLoops", "TestKtComplexity_CallsInScalingLoop_AbsentWithoutLoopCalls", "TestExtract_CallsInScalingLoop_InfiniteLoopCallsRetained", "TestExtract_CallsInScalingLoop_PresentButEmptyWhenAllBounded", "TestExtract_CallsInScalingLoop_AbsentWithoutLoopCalls", "TestPyComplexity_CallsInScalingLoop_InfiniteLoopCallsRetained", "TestPyComplexity_CallsInScalingLoop_PresentButEmptyWhenAllBounded", "TestPyComplexity_CallsInScalingLoop_AbsentWithoutLoopCalls", "TestTsComplexity_CallsInScalingLoop_InfiniteLoopCallsRetained", "TestTsComplexity_CallsInScalingLoop_PresentButEmptyWhenAllBounded", "TestTsComplexity_CallsInScalingLoop_AbsentWithoutLoopCalls", "TestGolden"}, // calls_in_scaling_loop = repeated (not merely scaling) loops, emitted even when empty
 	100: {"TestExtractTestRefs_InPackageCallResolvesToProductionSymbol", "TestExtractTestRefs_ExternalTestPackageResolvesThroughImport", "TestExtractTestRefs_SkipsBuiltinsAndEmitsNoSymbols", "TestExtractTestRefs_IgnoresFilesItDoesNotOwn", "TestExtractTestRefs_ReferenceFreeFileYieldsNoFact", "TestDefaultTestGlobsCoverGoAndStayIgnored", "TestGolden"}, // Go test_ref facts: both gates, so a function called only from its _test.go is no longer high-confidence dead
+	101: {"TestGoHTTPClient_PackageConstAbsoluteBaseURL_TaggedExternalWithHost", "TestGoHTTPClient_PackageScopedConstFromSiblingFile_TaggedExternal", "TestGoHTTPClient_StructFieldAbsoluteBaseURL_TaggedExternalNoHostWhenAmbiguous", "TestGoHTTPClient_ConfigInjectedBaseURL_NotTaggedExternal", "TestGoHTTPClient_RelativeLiteralBaseURL_NotTaggedExternal", "TestGoHTTPClient_MixedAbsoluteAndRelativeBindings_NotTaggedExternal", "TestGoHTTPClient_RouteNameUnchangedByExternalTagging", "TestComputeLinks_ExternalClientStillMatchesLoadedServer", "TestGolden"}, // Go base-URL host recovery: third-party client calls tagged external, hardcoded internal host still resolves
 }
 
 func TestCacheVersionCoverage(t *testing.T) {

internal/engine/cache.go

@@ -296,7 +296,19 @@ import (
 // spelled exactly as one from production and inherits goBuiltins filtering: a package-level
 // `min` shadowing the Go 1.21 builtin is still not credited, from either side. The file set
 // reaching the extractors changes, so cached snapshots must re-extract.
-const cacheVersion = "v100"
+// v101: the Go HTTP-client extractor recovers the host that `baseURL + "/path"`
+// concatenation discards. It resolves the base identifier through a package-scoped
+// string-literal index (const/var/assignment/composite-field bindings) and, when
+// every binding is an absolute http(s) URL, tags the route external=true (plus a
+// host prop when the bindings agree on one host). Before this, a service calling
+// only third-party APIs — golf, calling ZeptoMail and MailerLite via base-URL
+// concats — accumulated phantom "unresolved internal edges" and was misclassified
+// coverage_gap. A config-injected base (options.BaseURL, no literal binding) stays
+// an internal client route. Route facts gain external/host props, so cached Go
+// snapshots must re-extract. (The paired linker change — matching a client route
+// before bucketing it external, so a hardcoded internal host still resolves — runs
+// post-extraction and needs no cache bump; it is covered by TestGolden here.)
+const cacheVersion = "v101"
 
 // extractorCache holds per-extractor facts keyed by a content hash of the files
 // the extractor depends on. It is loaded from disk at the start of a snapshot and

internal/engine/golden_test.go

@@ -58,6 +58,11 @@ var fixtures = []fixture{
 	{name: "multirepo", subRepos: []string{"repoA", "repoB"}},
 	{name: "php_multirepo", subRepos: []string{"provider", "consumer"}},
 	{name: "go_grpc_multirepo", subRepos: []string{"server", "client"}},
+	// A Go backend plus a Go client that calls it and two third-party APIs. Pins
+	// GAP-LK-02 (v101): a `baseURL + "/path"` concat to a hardcoded host is tagged
+	// external, a hardcoded INTERNAL host still resolves to its loaded repo, and a
+	// config-injected base URL stays an unresolved internal edge.
+	{name: "go_httpclient_multirepo", subRepos: []string{"api", "consumer"}},
 	{name: "py_grpc_multirepo", subRepos: []string{"server", "client"}},
 	// Two different-language repos sharing only nested type names. The linker must
 	// draw no shared_symbols edge between them; see GAP-LK-03.

internal/engine/testdata/golden/go_httpclient_multirepo.facts.jsonl

@@ -0,0 +1,21 @@
+{"kind":"dependency","name":". -\u003e context","file":"consumer/client.go","line":4,"repo":"consumer","props":{"language":"go","source":"stdlib"},"relations":[{"kind":"imports","target":"context"}]}
+{"kind":"dependency","name":". -\u003e github.com/gorilla/mux","file":"api/server.go","line":6,"repo":"api","props":{"language":"go","source":"external"},"relations":[{"kind":"imports","target":"github.com/gorilla/mux"}]}
+{"kind":"dependency","name":". -\u003e net/http","file":"api/server.go","line":4,"repo":"api","props":{"language":"go","source":"stdlib"},"relations":[{"kind":"imports","target":"net/http"}]}
+{"kind":"dependency","name":". -\u003e net/http","file":"consumer/client.go","line":5,"repo":"consumer","props":{"language":"go","source":"stdlib"},"relations":[{"kind":"imports","target":"net/http"}]}
+{"kind":"dependency","name":"consumer -\u003e api","repo":"consumer","props":{"confidence":"probable","endpoint_count":1,"endpoints":["GET /v1/things/{id}"],"synthetic":"crossrepo","type":"cross_repo","via":["http-client"]}}
+{"kind":"module","name":".","file":"api/.","repo":"api","props":{"language":"go","modulePath":"example.com/api","package":"main"}}
+{"kind":"module","name":".","file":"consumer/.","repo":"consumer","props":{"language":"go","modulePath":"example.com/consumer","package":"main"}}
+{"kind":"route","name":"/v1/internal","file":"consumer/client.go","line":44,"repo":"consumer","props":{"api":"client","framework":"net-http","language":"go","method":"GET","role":"client","source":"go-http-client","target_hint":"","unmatched_by_server":true,"unmatched_reason":"path_unknown"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"route","name":"/v1/things/{id}","file":"api/server.go","line":13,"repo":"api","props":{"framework":"gorilla/mux","handler":"getThing","language":"go","method":"GET"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"route","name":"/v1/things/{id}","file":"consumer/client.go","line":42,"repo":"consumer","props":{"api":"client","external":true,"framework":"net-http","host":"api:8080","language":"go","method":"GET","role":"client","source":"go-http-client","target_hint":""},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"route","name":"/v1/widgets","file":"consumer/client.go","line":41,"repo":"consumer","props":{"api":"client","external":true,"framework":"net-http","host":"api.example.com","language":"go","method":"GET","role":"client","source":"go-http-client","target_hint":""},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"route","name":"/v3/messages","file":"consumer/client.go","line":43,"repo":"consumer","props":{"api":"client","external":true,"framework":"net-http","language":"go","method":"POST","role":"client","source":"go-http-client","target_hint":""},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"service","name":"api","repo":"api","props":{"synthetic":"crossrepo"}}
+{"kind":"service","name":"consumer","repo":"consumer","props":{"edge_coverage":[{"detected":4,"edge_type":"http_client","external":2,"resolved":1,"unresolved":1}],"synthetic":"crossrepo"},"relations":[{"kind":"depends_on","target":"api"}]}
+{"kind":"symbol","name":"..Client","file":"consumer/client.go","line":35,"repo":"consumer","props":{"exported":true,"language":"go","symbol_kind":"struct"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..Client.run","file":"consumer/client.go","line":40,"repo":"consumer","props":{"cyclomatic":1,"exported":false,"language":"go","receiver":"Client","symbol_kind":"method"},"relations":[{"kind":"calls","target":"net/http.Get"},{"kind":"calls","target":"net/http.NewRequest"},{"kind":"calls","target":"net/http.NewRequestWithContext"},{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..NewZepto","file":"consumer/client.go","line":20,"repo":"consumer","props":{"cyclomatic":2,"exported":true,"language":"go","symbol_kind":"function"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..Options","file":"consumer/client.go","line":33,"repo":"consumer","props":{"exported":true,"language":"go","symbol_kind":"struct"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..Zepto","file":"consumer/client.go","line":18,"repo":"consumer","props":{"exported":true,"language":"go","symbol_kind":"struct"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..getThing","file":"api/server.go","line":17,"repo":"api","props":{"cyclomatic":1,"exported":false,"language":"go","symbol_kind":"function"},"relations":[{"kind":"declares","target":"."}]}
+{"kind":"symbol","name":"..main","file":"api/server.go","line":11,"repo":"api","props":{"cyclomatic":1,"exported":false,"language":"go","symbol_kind":"function"},"relations":[{"kind":"calls","target":"github.com/gorilla/mux.NewRouter"},{"kind":"calls","target":"github.com/gorilla/mux.Router.HandleFunc"},{"kind":"calls","target":"net/http.ListenAndServe"},{"kind":"declares","target":"."}]}

internal/engine/testdata/repos/go_httpclient_multirepo/api/go.mod

@@ -0,0 +1,3 @@
+module example.com/api
+
+go 1.25

internal/engine/testdata/repos/go_httpclient_multirepo/api/server.go

@@ -0,0 +1,17 @@
+package main
+
+import (
+	"net/http"
+
+	"github.com/gorilla/mux"
+)
+
+// The api service serves one route; the consumer's hardcoded internal-host call
+// must resolve to it even though that call is also tagged external.
+func main() {
+	r := mux.NewRouter()
+	r.HandleFunc("/v1/things/{id}", getThing).Methods("GET")
+	http.ListenAndServe(":8080", r)
+}
+
+func getThing(w http.ResponseWriter, req *http.Request) {}

internal/engine/testdata/repos/go_httpclient_multirepo/consumer/client.go

@@ -0,0 +1,45 @@
+package main
+
+import (
+	"context"
+	"net/http"
+)
+
+// extBase is a third-party host with no loaded server: tagged external, host known.
+const extBase = "https://api.example.com"
+
+// internalBase is a hardcoded INTERNAL host that IS a loaded repo. The call to it
+// must resolve to the api service AND be tagged external — the linker attempts the
+// server match before falling back to the external bucket (GAP-LK-02, v101).
+const internalBase = "http://api:8080"
+
+// Zepto mirrors golf's region-switch idiom: a struct field bound to several
+// absolute literals that disagree on host — external, but no single host.
+type Zepto struct{ baseURL string }
+
+func NewZepto(region string) *Zepto {
+	var baseURL string
+	switch region {
+	case "eu":
+		baseURL = "https://api.zeptomail.eu/v1.1"
+	default:
+		baseURL = "https://api.zeptomail.com/v1.1"
+	}
+	return &Zepto{baseURL: baseURL}
+}
+
+// Options is injected from config; BaseURL has no string-literal binding, so the
+// call below stays an internal client route and remains an unresolved edge.
+type Options struct{ BaseURL string }
+
+type Client struct {
+	zepto   *Zepto
+	options Options
+}
+
+func (c *Client) run(ctx context.Context) {
+	http.Get(extBase + "/v1/widgets")
+	http.NewRequestWithContext(ctx, "GET", internalBase+"/v1/things/{id}", nil)
+	http.NewRequest("POST", c.zepto.baseURL+"/v3/messages", nil)
+	http.Get(c.options.BaseURL + "/v1/internal")
+}

internal/engine/testdata/repos/go_httpclient_multirepo/consumer/go.mod

@@ -0,0 +1,3 @@
+module example.com/consumer
+
+go 1.25

internal/extractors/goextractor/go.go

@@ -162,12 +162,17 @@ func (e *GoExtractor) extractPackage(fset *token.FileSet, pkgDir string, pp *par
 	// resolves at its call sites.
 	pkgVarClients := collectPackageVarClients(pp.parsedFiles, grpcStubs)
 
+	// Package-scoped string-literal bindings (const/var/assign/field) so a
+	// `baseURL + "/path"` client call can recover the host the concat discards,
+	// even when the base URL is declared in a sibling file of the same package.
+	baseURLLits := collectBaseURLLiterals(pp.parsedFiles)
+
 	for _, relFile := range pp.relFiles {
 		f, ok := pp.fileMap[relFile]
 		if !ok {
 			continue
 		}
-		result = append(result, e.extractFile(fset, f, relFile, pkgDir, modulePath, fieldTypes, pkgNames, grpcStubs, pkgVarClients)...)
+		result = append(result, e.extractFile(fset, f, relFile, pkgDir, modulePath, fieldTypes, pkgNames, grpcStubs, pkgVarClients, baseURLLits)...)
 	}
 
 	moduleFact := facts.Fact{
@@ -189,7 +194,7 @@ func (e *GoExtractor) extractPackage(fset *token.FileSet, pkgDir string, pp *par
 	return result
 }
 
-func (e *GoExtractor) extractFile(fset *token.FileSet, f *ast.File, relFile, pkgDir, modulePath string, fieldTypes map[string]string, pkgNames map[string]string, grpcStubs *goGRPCStubIndex, pkgVarClients map[string]string) []facts.Fact {
+func (e *GoExtractor) extractFile(fset *token.FileSet, f *ast.File, relFile, pkgDir, modulePath string, fieldTypes map[string]string, pkgNames map[string]string, grpcStubs *goGRPCStubIndex, pkgVarClients map[string]string, baseURLLits map[string][]string) []facts.Fact {
 	var result []facts.Fact
 
 	// Build per-file import alias map for call resolution.
@@ -240,7 +245,7 @@ func (e *GoExtractor) extractFile(fset *token.FileSet, f *ast.File, relFile, pkg
 	result = append(result, extractRoutes(fset, f, relFile, pkgDir)...)
 
 	// Extract outbound HTTP-client calls
-	result = append(result, extractHTTPClientFacts(fset, f, relFile, pkgDir)...)
+	result = append(result, extractHTTPClientFacts(fset, f, relFile, pkgDir, baseURLLits)...)
 
 	// Extract outbound gRPC-client calls
 	result = append(result, extractGRPCClientFacts(fset, f, relFile, pkgDir, modulePath, fileImports, fieldTypes, pkgVarClients, grpcStubs)...)