Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix(parser): allow text output_format for audio family (#18575)

Jack99e1c2eaca5d
diff --git a/internal/ingestion/component/media_dispatch_test.go b/internal/ingestion/component/media_dispatch_test.go
index 2c465616cc..ee98c2bf73 100644
--- a/internal/ingestion/component/media_dispatch_test.go
+++ b/internal/ingestion/component/media_dispatch_test.go
@@ -25,6 +25,7 @@ import (
"ragflow/internal/common"
"ragflow/internal/entity"
modelModule "ragflow/internal/entity/models"
+ "ragflow/internal/ingestion/component/schema"
"ragflow/internal/utility"
"gorm.io/gorm"
@@ -362,9 +363,13 @@ func TestMaybeDispatchAudio_TextCarriesTranscription(t *testing.T) {
}
}
-// TestMaybeDispatchAudio_DefaultOutputFormatJson covers Parser 2.11:
-// the default audio output_format must be "json" (matching Python
-// parser.py:232 and AllowedOutputFormat["audio"]={"json"}).
+// TestMaybeDispatchAudio_DefaultOutputFormatJson covers the
+// maybeDispatchAudio fallback: when an audio setup omits
+// output_format entirely, the dispatch defaults to "json" and wraps
+// the transcription as a JSON item. (The defaultSetups value is
+// "text" to mirror the Python audio setup in parser.py; this test
+// deliberately supplies an empty setup to exercise the fallback
+// inside the dispatch itself.)
func TestMaybeDispatchAudio_DefaultOutputFormatJson(t *testing.T) {
const want = "hello world"
drv := &audioTranscribeDriver{transcription: want}
@@ -373,8 +378,9 @@ func TestMaybeDispatchAudio_DefaultOutputFormatJson(t *testing.T) {
resolveTenantModelByType = func(ctx context.Context, db *gorm.DB, tenantID string, modelType entity.ModelType) (modelModule.ModelDriver, string, *modelModule.APIConfig, int, error) {
return drv, "asr-model", &modelModule.APIConfig{}, 0, nil
}
- setups := defaultSetups()
- // Do NOT set output_format — exercise the default path.
+ // No output_format key — exercise the default path inside
+ // maybeDispatchAudio.
+ setups := map[string]schema.ParserSetup{"audio": {}}
res, dispatched, err := maybeDispatchAudio(
context.Background(),
nil,
diff --git a/internal/ingestion/component/parser.go b/internal/ingestion/component/parser.go
index 5031a5b4be..19c74443f0 100644
--- a/internal/ingestion/component/parser.go
+++ b/internal/ingestion/component/parser.go
@@ -314,7 +314,7 @@ func defaultSetups() map[string]schema.ParserSetup {
"aiff", "au", "midi", "wma", "realaudio", "vqf",
"oggvorbis", "ape",
},
- "output_format": "json",
+ "output_format": "text",
},
"video": {
"suffix": []string{"mp4", "avi", "mkv"},
diff --git a/internal/ingestion/component/parser_dispatch_test.go b/internal/ingestion/component/parser_dispatch_test.go
index 35036258e4..f1f9badb43 100644
--- a/internal/ingestion/component/parser_dispatch_test.go
+++ b/internal/ingestion/component/parser_dispatch_test.go
@@ -305,6 +305,84 @@ func TestDefaultSetups_DOCX_OutputFormatMarkdown(t *testing.T) {
}
}
+// TestResolveOutputFormat_AudioOutputFormats pins the audio-family
+// whitelist against the builtin audio template. The template
+// (ingestion_pipeline_audio.json) and the Python default audio setup
+// (rag/flow/parser/parser.py) both use output_format="text" — an audio
+// transcription is inherently plain text. "text" must therefore pass
+// the whitelist, "json" must stay accepted, and a format outside the
+// whitelist must still be rejected.
+func TestResolveOutputFormat_AudioOutputFormats(t *testing.T) {
+ allowed := schema.ParserParam{}.Defaults().AllowedOutputFormat
+ audioAllowed, ok := allowed["audio"]
+ if !ok {
+ t.Fatal("allowed_output_format: audio key missing")
+ }
+ has := func(want string) bool {
+ for _, v := range audioAllowed {
+ if strings.EqualFold(v, want) {
+ return true
+ }
+ }
+ return false
+ }
+ if !has("text") {
+ t.Errorf("allowed_output_format[audio] = %v, want it to include %q (builtin audio template and Python default use it)", audioAllowed, "text")
+ }
+ if !has("json") {
+ t.Errorf("allowed_output_format[audio] = %v, want it to include %q", audioAllowed, "json")
+ }
+
+ cases := []struct {
+ name string
+ format string
+ wantErr bool
+ }{
+ {name: "text accepted", format: "text"},
+ {name: "json accepted", format: "json"},
+ {name: "html rejected", format: "html", wantErr: true},
+ }
+ for _, tc := range cases {
+ t.Run(tc.name, func(t *testing.T) {
+ setups := map[string]schema.ParserSetup{"audio": {"output_format": tc.format}}
+ got, err := resolveOutputFormat("audio", setups, allowed)
+ if tc.wantErr {
+ if err == nil {
+ t.Fatalf("want error for audio output_format=%q, got %q", tc.format, got)
+ }
+ return
+ }
+ if err != nil {
+ t.Fatalf("audio output_format=%q: unexpected error: %v", tc.format, err)
+ }
+ if got != tc.format {
+ t.Errorf("got %q, want %q", got, tc.format)
+ }
+ })
+ }
+}
+
+// TestDefaultSetups_Audio_OutputFormatText pins the audio default to
+// "text", matching the Python default setup
+// (rag/flow/parser/parser.py audio block) and the builtin audio
+// template. The default feeds both the whitelist gate and the ASR
+// dispatch, so it must not drift to a format the audio pipeline does
+// not produce.
+func TestDefaultSetups_Audio_OutputFormatText(t *testing.T) {
+ setups := defaultSetups()
+ audio, ok := setups["audio"]
+ if !ok {
+ t.Fatal("defaultSetups: audio key missing")
+ }
+ got, ok := audio["output_format"].(string)
+ if !ok {
+ t.Fatal("defaultSetups: audio.output_format missing or not a string")
+ }
+ if got != "text" {
+ t.Errorf("audio.output_format = %q, want %q", got, "text")
+ }
+}
+
func TestConfigureParserFromSetups_UsesPythonFamilySetup(t *testing.T) {
setups := defaultSetups()
got := &captureSetupConfigurer{}
diff --git a/internal/ingestion/component/schema/parser.go b/internal/ingestion/component/schema/parser.go
index d47ed57df9..5f6e91d1fc 100644
--- a/internal/ingestion/component/schema/parser.go
+++ b/internal/ingestion/component/schema/parser.go
@@ -85,7 +85,7 @@ func (ParserParam) Defaults() ParserParam {
"markdown": {"text", "json"},
"text&code": {"text", "json"},
"html": {"text", "json"},
- "audio": {"json"},
+ "audio": {"text", "json"},
"video": {},
"epub": {"text", "json"},
"json": {"json"},