2 files changed, 50 insertions, 0 deletions
diff --git a/internal/extract/extract.go b/internal/extract/extract.go
index aa3ae93..9768db1 100644
--- a/internal/extract/extract.go
+++ b/internal/extract/extract.go
@@ -7,12 +7,18 @@ package extract
 import (
 	"context"
 	"errors"
+	"fmt"
 	"os"
 	"path/filepath"
 	"strings"
 	"time"
 )
 
+// Version is the version of the text this package extracts. Bump it
+// whenever a change could make any format's text differ, so every keyword
+// cache built from the old text is discarded (Fingerprint).
+const Version = 1
+
 var (
 	// ErrUnsupported is returned when the format carries no text krino
 	// knows how to extract.
@@ -111,6 +117,26 @@ func newWithPath(path string) *Extractor {
 	return &Extractor{tools: tools, Timeout: 30 * time.Second}
 }
 
+// Fingerprint identifies what text this Extractor would produce: Version,
+// and each external tool found with its path, size and modification time.
+// A keyword cache built under another fingerprint is discarded, so
+// installing, removing or upgrading a tool invalidates it.
+func (e *Extractor) Fingerprint() string {
+	var b strings.Builder
+	fmt.Fprintf(&b, "v%d", Version)
+	for _, name := range toolNames {
+		p := e.tools[name]
+		if p == "" {
+			continue
+		}
+		fmt.Fprintf(&b, " %s=%s", name, p)
+		if fi, err := os.Stat(p); err == nil {
+			fmt.Fprintf(&b, ":%d:%d", fi.Size(), fi.ModTime().UnixNano())
+		}
+	}
+	return b.String()
+}
+
 // Tools reports every external tool Extractor knows about, in a fixed
 // order, with the path it was found at or "" if it was not found.
 func (e *Extractor) Tools() []Tool {
diff --git a/internal/extract/tools_test.go b/internal/extract/tools_test.go
index 4a03f1b..b481e8a 100644
--- a/internal/extract/tools_test.go
+++ b/internal/extract/tools_test.go
@@ -237,3 +237,27 @@ func TestMaxReadCapsToolOutput(t *testing.T) {
 		t.Fatalf("got %v, want ErrTooLarge (max-read 1024 should have capped a 200000-byte tool output)", err)
 	}
 }
+
+// TestFingerprintFollowsTools: the fingerprint changes when a tool is
+// installed or replaced, and not otherwise.
+func TestFingerprintFollowsTools(t *testing.T) {
+	bin := t.TempDir()
+	none := newWithPath(bin).Fingerprint()
+	if none != newWithPath(bin).Fingerprint() {
+		t.Fatal("fingerprint not stable")
+	}
+	tool := filepath.Join(bin, "pdftotext")
+	if err := os.WriteFile(tool, []byte("#!/bin/sh\n"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	installed := newWithPath(bin).Fingerprint()
+	if installed == none {
+		t.Error("installing pdftotext did not change the fingerprint")
+	}
+	if err := os.WriteFile(tool, []byte("#!/bin/sh\nexit 0\n"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if newWithPath(bin).Fingerprint() == installed {
+		t.Error("replacing pdftotext did not change the fingerprint")
+	}
+}