feat(app): auto-merge split test suites under one workspace id

CHANGELOG.md

@@ -8,6 +8,18 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ### Added
 
+- **Split test suites merge into one workspace row.** Multiple
+  `coverage-<id>--<suite>.xml` (and `tests-<id>--<suite>.xml`) files with the
+  same base id are now combined automatically: coverage is unioned per source
+  line via `cobertura.Merge` (a line counts as hit if any suite executed it),
+  and JUnit test counts are summed. A workspace exercised by unit +
+  integration jobs — or a language-runtime matrix — reports as one row instead
+  of one row per suite. Single-file usage (`coverage-web.xml`) is unchanged;
+  the merge only kicks in on the `--` (double-dash) sentinel, so existing ids
+  with single dashes (`shared-widget`) or dots (`api.v1`) stay distinct
+  workspaces. Any `coverage.yaml` workspace entry keyed on the base id applies
+  to every same-id artifact. See
+  [Splitting a workspace across test suites](README.md#splitting-a-workspace-across-test-suites).
 - **`cobertura.Merge` — union of parsed reports.** Combines multiple `*Report`
   values into one by unioning classes that share a filename (first-seen order
   preserved) and, within a class, unioning lines by number: `Hits` is the max

README.md

@@ -126,11 +126,16 @@ Both formats are rendered from templates in
 
 | Thing | Convention | Example |
 |---|---|---|
-| Coverage artifact | `coverage-<id>.xml` (Cobertura) | `coverage-thingy.xml` |
-| Test-count artifact | `tests-<id>.xml` (JUnit) | `tests-thingy.xml` |
-| Workspace id | the `<id>` in the filenames; may contain dashes | `shared-widget` |
+| Coverage artifact | `coverage-<id>[--<suite>].xml` (Cobertura) | `coverage-thingy.xml`, `coverage-thingy--unit.xml` |
+| Test-count artifact | `tests-<id>[--<suite>].xml` (JUnit) | `tests-thingy.xml`, `tests-thingy--unit.xml` |
+| Workspace id | the `<id>` in the filenames; may contain single dashes or dots | `shared-widget`, `api.v1` |
 | Input dir | all `coverage-*.xml` + `tests-*.xml` flattened together | `./coverage-artifacts` |
 
+The optional `--<suite>` suffix (double dash) lets one workspace ship coverage
+from multiple test jobs — see
+[Splitting a workspace across test suites](#splitting-a-workspace-across-test-suites)
+below. Double-dash is used so existing dotted or single-dashed ids stay intact.
+
 ## Add it to your workflow
 
 Three moving parts:
@@ -197,6 +202,29 @@ for your detected languages — see [docs/INIT.md](./docs/INIT.md). The Action's
 inputs mirror the [CLI flags](#usage); the full annotated workflow is
 [`examples/coverage.yml`](./examples/coverage.yml).
 
+## Splitting a workspace across test suites
+
+When one workspace is exercised by multiple test jobs (e.g. unit + integration,
+or a language runtime matrix), upload each job's artifacts with the same base id
+and a distinct `--<suite>` suffix:
+
+| Job | Coverage file | Tests file |
+|---|---|---|
+| unit tests | `coverage-web--unit.xml` | `tests-web--unit.xml` |
+| integration tests | `coverage-web--integration.xml` | `tests-web--integration.xml` |
+
+The tool groups them by base id (`web`) and produces one row: coverage is
+**unioned per source line** (a line counts as hit if any suite executed it) and
+test counts are **summed**. A line that's a branch in any suite is a branch in
+the merged row; branch coverage is taken as the max across suites (a safe
+approximation — always `>=` any single suite and `<=` the true union).
+
+A single-file id like `coverage-web.xml` still works exactly as before — the
+merge only kicks in on the `--` sentinel, so ids with single dashes
+(`shared-widget`) or dots (`api.v1`, `api.v2`) remain distinct workspaces. Any
+`coverage.yaml` workspace config keyed on `web` applies to every same-id
+artifact.
+
 ## Optional config
 
 - **`.coverageignore`** — gitignore syntax, matched against repo-root-relative

docs/README.md

@@ -9,10 +9,15 @@ Each project produces two files in a shared directory:
 
 | File | Format | Purpose |
 |---|---|---|
-| `coverage-<id>.xml` | Cobertura | line & branch coverage |
-| `tests-<id>.xml` | JUnit | test count (renders `—` if absent) |
-
-`<id>` is any workspace id you choose (may contain dashes).
+| `coverage-<id>[--<suite>].xml` | Cobertura | line & branch coverage |
+| `tests-<id>[--<suite>].xml` | JUnit | test count (renders `—` if absent) |
+
+`<id>` is any workspace id you choose (may contain single dashes or dots). The
+optional `--<suite>` suffix (double dash) lets a workspace split across
+multiple test jobs (unit + integration, or a matrix) contribute one merged row
+— coverage is unioned per line, test counts are summed. See
+[Splitting a workspace across test suites](../README.md#splitting-a-workspace-across-test-suites)
+in the main README.
 
 ## Per-language how-to guides
 

internal/app/app.go

@@ -204,31 +204,44 @@ func aggregateCoverage(opts Options, cfg *config.Config, matcher ignore.Matcher)
 	}
 	sort.Strings(files)
 
-	var workspaces []*workspaceAgg
-	excluded := 0
-
+	// Group files by workspace id — multiple coverage-<id>--<suite>.xml files
+	// with the same id are unioned via cobertura.Merge so a workspace split
+	// across test suites renders as one row.
+	var order []string
+	byID := map[string][]*cobertura.Report{}
 	for _, file := range files {
 		id := artifactID(file, "coverage-")
 		if id == "" {
 			continue
 		}
-		wsCfg, configured := cfg.Workspaces[id]
-		if !configured && opts.Verbose {
-			fmt.Fprintf(opts.Stderr, "warning: workspace %q has no config entry; matching with raw filenames\n", id)
-		}
-
 		report, err := cobertura.ParseFile(file)
 		if err != nil {
 			fmt.Fprintf(opts.Stderr, "warning: skipping %q: %v\n", file, err)
 			continue
 		}
+		if _, seen := byID[id]; !seen {
+			order = append(order, id)
+		}
+		byID[id] = append(byID[id], report)
+	}
+
+	var workspaces []*workspaceAgg
+	excluded := 0
+
+	for _, id := range order {
+		wsCfg, configured := cfg.Workspaces[id]
+		if !configured && opts.Verbose {
+			fmt.Fprintf(opts.Stderr, "warning: workspace %q has no config entry; matching with raw filenames\n", id)
+		}
+
+		merged := cobertura.Merge(byID[id])
 
 		ws := &workspaceAgg{
 			id:          id,
 			displayName: cfg.DisplayName(id),
 			folders:     map[string]*folderAgg{},
 		}
-		for _, class := range report.Classes {
+		for _, class := range merged.Classes {
 			rel := stripPrefix(class.Filename, wsCfg.StripPrefix)
 			full := wsCfg.Prefix + rel
 			if matcher.Match(full) {
@@ -281,6 +294,8 @@ func attachTests(opts Options, workspaces []*workspaceAgg) {
 		byID[w.id] = w
 	}
 
+	// Multiple tests-<id>--<suite>.xml files with the same id sum their test
+	// counts, matching the coverage-side merge for split suites.
 	for _, file := range files {
 		id := artifactID(file, "tests-")
 		if id == "" {
@@ -292,7 +307,7 @@ func attachTests(opts Options, workspaces []*workspaceAgg) {
 			continue
 		}
 		if w, ok := byID[id]; ok {
-			w.tests = report.Tests
+			w.tests += report.Tests
 			w.hasTests = true
 		}
 	}
@@ -451,13 +466,23 @@ func resolveFormat(opts Options) (string, error) {
 // --- helpers ---
 
 // artifactID strips a leading prefix and a trailing ".xml" from a path's base
-// name. The id itself may contain dashes.
+// name. The id itself may contain single dashes and dots. A double-dash "--"
+// in the stripped remainder separates the id from an optional suite suffix,
+// so split test suites can upload sibling artifacts (coverage-web--unit.xml,
+// coverage-web--integration.xml) that get merged into one workspace row. The
+// suffix itself is discarded — it only exists so multiple files can share the
+// same id in one input directory. Double-dash is chosen so existing dotted ids
+// like "api.v1" and dashed ids like "shared-widget" remain intact.
 func artifactID(path, prefix string) string {
 	base := filepath.Base(path)
 	if !strings.HasPrefix(base, prefix) || !strings.HasSuffix(base, ".xml") {
 		return ""
 	}
-	return base[len(prefix) : len(base)-len(".xml")]
+	rest := base[len(prefix) : len(base)-len(".xml")]
+	if sep := strings.Index(rest, "--"); sep >= 0 {
+		return rest[:sep]
+	}
+	return rest
 }
 
 func stripPrefix(filename, prefix string) string {

internal/app/app_test.go

@@ -263,3 +263,107 @@ func TestRunUnknownFormatErrors(t *testing.T) {
 		t.Fatal("expected error for unknown format")
 	}
 }
+
+// Same workspace id split across two coverage artifacts (unit + integration
+// suites) merges into one row with per-line hits unioned.
+func TestRunMergesSplitSuitesUnderOneWorkspace(t *testing.T) {
+	dir := t.TempDir()
+	// Same file, three lines. Unit hits lines 1 and 2; integration hits lines 2
+	// and 3. Merged: 3/3 covered.
+	write(t, dir, "coverage-web--unit.xml", coverageDoc(map[string][]int{"src/a.ts": {1, 1, 0}}))
+	write(t, dir, "coverage-web--integration.xml", coverageDoc(map[string][]int{"src/a.ts": {0, 1, 1}}))
+	// And the corresponding split test artifacts — counts should sum.
+	write(t, dir, "tests-web--unit.xml", `<testsuites tests="3"/>`)
+	write(t, dir, "tests-web--integration.xml", `<testsuites tests="5"/>`)
+
+	out, err := run(t, Options{Input: dir})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if !strings.Contains(out, "| web | 8 | 3 / 3 | 100.0% |") {
+		t.Errorf("expected merged web row (8 tests, 3/3 lines), got:\n%s", out)
+	}
+	if strings.Count(out, "| web ") != 1 {
+		t.Errorf("expected exactly one web row, got:\n%s", out)
+	}
+}
+
+// A dashed id (single dashes, no "--") is untouched — split-suite parsing only
+// kicks in on the double-dash sentinel.
+func TestRunPreservesDashedIDs(t *testing.T) {
+	dir := t.TempDir()
+	write(t, dir, "coverage-shared-widget.xml", coverageDoc(map[string][]int{"src/w.ts": {1, 1}}))
+	write(t, dir, "tests-shared-widget.xml", `<testsuites tests="4"/>`)
+
+	out, err := run(t, Options{Input: dir})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if !strings.Contains(out, "| shared-widget | 4 | 2 / 2 | 100.0% |") {
+		t.Errorf("dashed id should not be split at the dash, got:\n%s", out)
+	}
+}
+
+// Dotted ids (e.g. versioned service names like "api.v1") must stay distinct
+// after this feature landed — regression guard for
+// https://github.com/aanantaco/coverage/pull/23#discussion_r3849572610.
+func TestRunPreservesDottedIDsAsDistinctWorkspaces(t *testing.T) {
+	dir := t.TempDir()
+	write(t, dir, "coverage-api.v1.xml", coverageDoc(map[string][]int{"src/v1.go": {1, 1}}))
+	write(t, dir, "coverage-api.v2.xml", coverageDoc(map[string][]int{"src/v2.go": {1, 0, 0}}))
+	// Only v1 gets a display name in config — used to prove config lookup still
+	// resolves by the full dotted id, not a truncated prefix.
+	cfgPath := filepath.Join(dir, "coverage.yaml")
+	write(t, dir, "coverage.yaml", `
+workspaces:
+  "api.v1":
+    display_name: API v1
+`)
+
+	out, err := run(t, Options{
+		Input: dir, ConfigPath: cfgPath, ConfigSet: true,
+	})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	if !strings.Contains(out, "| API v1 | — | 2 / 2 | 100.0% |") {
+		t.Errorf("api.v1 row missing (config lookup should hit the full dotted id), got:\n%s", out)
+	}
+	if !strings.Contains(out, "| api.v2 | — | 1 / 3 | 33.3% |") {
+		t.Errorf("api.v2 row missing or merged into api.v1, got:\n%s", out)
+	}
+}
+
+// A workspaces config entry keyed on the base id applies to every same-id
+// artifact regardless of suffix.
+func TestRunSplitSuitesShareConfig(t *testing.T) {
+	dir := t.TempDir()
+	write(t, dir, "coverage-worker--unit.xml", coverageDoc(map[string][]int{
+		"github.com/acme/repo/services/worker/logic/a.go": {1, 1},
+	}))
+	write(t, dir, "coverage-worker--integration.xml", coverageDoc(map[string][]int{
+		"github.com/acme/repo/services/worker/logic/b.go": {1},
+	}))
+	cfgPath := filepath.Join(dir, "coverage.yaml")
+	write(t, dir, "coverage.yaml", `
+workspaces:
+  worker:
+    strip_prefix: github.com/acme/repo/services/worker/
+`)
+
+	out, err := run(t, Options{
+		Input: dir, ConfigPath: cfgPath, ConfigSet: true,
+	})
+	if err != nil {
+		t.Fatalf("Run: %v", err)
+	}
+	// 2 lines from unit + 1 line from integration, all covered.
+	if !strings.Contains(out, "| worker | — | 3 / 3 | 100.0% |") {
+		t.Errorf("split-suite config bridging failed, got:\n%s", out)
+	}
+	// Folder grouping should reflect the stripped rel path, proving strip_prefix
+	// was applied to both merged files.
+	if !strings.Contains(out, "└ logic") {
+		t.Errorf("expected stripped folder grouping for merged files:\n%s", out)
+	}
+}

llms.txt

@@ -9,12 +9,17 @@
 
 Each project/test job produces two files in one shared directory:
 
-- `coverage-<id>.xml` — Cobertura coverage
-- `tests-<id>.xml` — JUnit test results (optional; the Tests column shows `—` if absent)
-
-`<id>` is any workspace id you choose (it may contain dashes). Then, in one
-aggregation job, download every `coverage-*` / `tests-*` artifact into a single
-directory (`merge-multiple: true`) and run the tool once:
+- `coverage-<id>[--<suite>].xml` — Cobertura coverage
+- `tests-<id>[--<suite>].xml` — JUnit test results (optional; the Tests column shows `—` if absent)
+
+`<id>` is any workspace id you choose (it may contain single dashes or dots).
+The optional `--<suite>` suffix (double dash) lets a workspace split across
+multiple test jobs (unit + integration, matrix) contribute one merged row:
+coverage is unioned per source line, test counts are summed. Single-file usage
+(`coverage-web.xml`) is unchanged, and ids with single dashes (`shared-widget`)
+or dots (`api.v1`) stay distinct. Then, in one aggregation job, download every
+`coverage-*` / `tests-*` artifact into a single directory
+(`merge-multiple: true`) and run the tool once:
 
     coverage --input ./cov --output "$GITHUB_STEP_SUMMARY"      # Markdown
     coverage --input ./cov --output report.html                # HTML (auto-detected)