Fix: when save dataset parser_config, warn user only new files take effects (#21416)

web/src/locales/en.ts

@@ -772,6 +772,8 @@ Example: A 1 KB message with 1024-dim embedding uses ~9 KB. The 5 MB default lim
       parsingMode: 'Parsing mode',
       manualSetup: 'Custom ingestion pipeline',
       builtIn: 'Built-in parsing template',
+      parserSettingsChangeWarning:
+        'Updated parser settings apply only to newly uploaded files. To apply these settings to existing files, update their parser settings and re-parse them.',
       noConfigChunkerHint:
         'The built-in {{name}} parser handles chunking automatically, so no additional configuration is required.',
       titleDescription:

web/src/locales/zh.ts

@@ -708,6 +708,8 @@ export default {
       parsingMode: '解析方式',
       manualSetup: '自定义Ingestion pipeline',
       builtIn: '内置解析模板',
+      parserSettingsChangeWarning:
+        '更新后的解析设置仅对新上传的文件生效。如需对现有文件生效，请修改这些文件的解析设置并重新解析。',
       noConfigChunkerHint:
         '内置的 {{name}} 解析器会自动处理分块，因此无需额外配置。',
       titleDescription: '在这里更新知识库配置，特别是大语言模型和提示词。',

web/src/pages/dataset/setting/index.tsx

@@ -9,6 +9,7 @@ import {
   CardTitle,
 } from '@/components/ui/card';
 import Divider from '@/components/ui/divider';
+import { Modal } from '@/components/ui/modal/modal';
 import {
   Form,
   FormField,
@@ -24,6 +25,7 @@ import {
   usePipelineOperatorNodes,
   useResetParserConfigOnPipelineChange,
 } from '@/hooks/use-pipeline-operator';
+import { buildParserConfigFromNodes } from '@/utils/pipeline-operator';
 import { zodResolver } from '@hookform/resolvers/zod';
 import { isEqual } from 'lodash';
 import { useCallback, useEffect, useState } from 'react';
@@ -42,6 +44,7 @@ import {
   usePipelineDataList,
   useSaveDatasetSetting,
 } from './hooks';
+import { normalizeParserConfigForComparison } from './parser-config';
 import { useRevealSubmitErrors } from './use-reveal-submit-errors';
 
 export default function DatasetSetting() {
@@ -146,9 +149,39 @@ export default function DatasetSetting() {
 
   const handleSubmit = useCallback(
     async (data: z.infer<typeof formSchema>) => {
-      await handleSave(data);
+      const savedValues = form.formState.defaultValues;
+      const operatorDefaults = buildParserConfigFromNodes(operatorNodes);
+      const [savedParserConfig, parserConfig] = [
+        savedValues?.parser_config,
+        data.parser_config,
+      ].map((config) =>
+        normalizeParserConfigForComparison(config, operatorDefaults, t),
+      );
+      const parserSettingsChanged =
+        data.parse_type !== savedValues?.parse_type ||
+        (data.parser_id || '') !== (savedValues?.parser_id || '') ||
+        (data.pipeline_id || '') !== (savedValues?.pipeline_id || '') ||
+        !isEqual(parserConfig, savedParserConfig);
+
+      if (knowledgeDetails.chunk_count > 0 && parserSettingsChanged) {
+        await new Promise<void>((resolve) => {
+          Modal.warning({
+            title: t('common.warn'),
+            content: t('knowledgeConfiguration.parserSettingsChangeWarning'),
+            okText: t('common.ok'),
+            closable: false,
+            showCancel: false,
+            onOk: resolve,
+          });
+        });
+      }
+
+      const result = await handleSave(data);
+      if (result.code === 0) {
+        form.reset(data);
+      }
     },
-    [handleSave],
+    [form, handleSave, knowledgeDetails.chunk_count, operatorNodes, t],
   );
 
   const { errors } = useFormState({
@@ -294,6 +327,7 @@ export default function DatasetSetting() {
                   loading={
                     datasetSettingLoading ||
                     saveLoading ||
+                    form.formState.isSubmitting ||
                     (operatorNodesLoading && operatorNodes.length === 0)
                   }
                 >

web/src/pages/dataset/setting/parser-config.ts

@@ -0,0 +1,85 @@
+import { Languages } from '@/components/cross-language-form-field';
+import { ParseDocumentType } from '@/components/layout-recognize-form-field';
+import { Operator } from '@/constants/agent';
+import { FileType } from '@/constants/file';
+import { normalizeParserFormValues } from '@/pages/agent/form/parser-form/utils';
+import { transformApiResponseToForm } from '@/pages/agent/form/title-chunker-form/hook';
+import {
+  getOperatorType,
+  normalizeOperatorForm,
+  transformExtractorConfigToForm,
+  transformFormConfigToApi,
+} from '@/utils/pipeline-operator';
+import type { TFunction } from 'i18next';
+
+export function normalizeParserConfigForComparison(
+  config: Record<string, any> | undefined,
+  operatorDefaults: Record<string, any>,
+  t: TFunction,
+) {
+  const entries = Object.entries({ ...operatorDefaults, ...config })
+    .filter(([operatorId]) => operatorId.includes(':'))
+    .map(([operatorId, values]) => {
+      const operatorType = getOperatorType(operatorId);
+      let normalized = normalizeOperatorForm(operatorId, values);
+
+      // Mirror initialization performed inside the operator forms, including
+      // defaults that their mounted controls write back without user edits.
+      if (operatorType === Operator.Parser) {
+        normalized = normalizeParserFormValues(normalized);
+        normalized.setups = normalized.setups.map(
+          (setup: Record<string, any>) => {
+            const next = { ...setup };
+            if (
+              setup.fileFormat === FileType.PDF &&
+              setup.parse_method &&
+              ![
+                ParseDocumentType.DeepDOC,
+                ParseDocumentType.PlainText,
+                ParseDocumentType.TCADPParser,
+              ].includes(setup.parse_method) &&
+              !setup.lang
+            ) {
+              next.lang = Languages[0];
+            }
+            if (
+              [
+                FileType.PDF,
+                FileType.Spreadsheet,
+                FileType.PowerPoint,
+              ].includes(setup.fileFormat) &&
+              setup.parse_method === ParseDocumentType.TCADPParser
+            ) {
+              next.table_result_type ||= '1';
+              next.markdown_image_response_type ||= '1';
+            }
+            return next;
+          },
+        );
+      } else if (operatorType === Operator.Extractor) {
+        normalized = transformExtractorConfigToForm(normalized);
+        for (const feature of ['keywords', 'questions', 'summary']) {
+          normalized[feature] = {
+            ...normalized[feature],
+            system_prompt:
+              normalized[feature]?.system_prompt ||
+              t(`flow.prompts.system.${feature}`),
+          };
+        }
+      } else if (
+        operatorType === Operator.TitleChunker ||
+        operatorType === Operator.ManualChunker
+      ) {
+        normalized = transformApiResponseToForm(normalized);
+        if (operatorType === Operator.ManualChunker) {
+          normalized.method = 'group';
+        }
+      }
+
+      return [operatorId, transformFormConfigToApi(operatorType, normalized)];
+    });
+
+  // Registered optional controls can add undefined properties. Compare the
+  // serialized settings sent to the API, where those properties are omitted.
+  return JSON.parse(JSON.stringify(Object.fromEntries(entries)));
+}