Align Go ingestion boundaries with Python (#16647)

Moves doc_id blob resolution into Parser, tightens chunker/tokenizer to
Python output_format semantics, updates extractor list handling, and
fixes real-template integration tests.

internal/parser/parser/doc_parser.go

@@ -0,0 +1,68 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"fmt"
+
+	officeOxide "github.com/yfedoseev/office_oxide/go"
+)
+
+type DOCParser struct {
+	libType string
+}
+
+func NewDOCParser(libType string) (*DOCParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &DOCParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported DOC library type: %s", libType)
+	}
+}
+
+func (p *DOCParser) String() string {
+	return "DOCParser"
+}
+
+// ParseWithResult captures the office_oxide PlainText output for
+// the DOC family. Python parser.py routes .doc through tika;
+// the Go side uses office_oxide which supports DOC via PlainText.
+// OutputFormat="text" — the python side falls back to text for
+// legacy DOC files since structured extraction is unreliable.
+func (p *DOCParser) ParseWithResult(filename string, data []byte) ParseResult {
+	doc, err := officeOxide.OpenFromBytes(data, "doc")
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("doc open: %w", err)}
+	}
+	defer doc.Close()
+
+	text, err := doc.PlainText()
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("doc plain-text: %w", err)}
+	}
+
+	return ParseResult{
+		OutputFormat: "text",
+		File:         map[string]any{"name": filename, "format": "doc"},
+		Text:         text,
+	}
+}

internal/parser/parser/docx_parser.go

@@ -0,0 +1,69 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"fmt"
+
+	officeOxide "github.com/yfedoseev/office_oxide/go"
+)
+
+type DOCXParser struct {
+	libType string
+}
+
+func NewDOCXParser(libType string) (*DOCXParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &DOCXParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported DOCX library type: %s", libType)
+	}
+}
+
+// ParseWithResult captures the office_oxide ToMarkdown output
+// instead of discarding it. Returns OutputFormat="markdown" with
+// the rendered Markdown on the matching key. Mirrors the python
+// parser.py:Docx branch which uses rag.app.naive.Docx to render
+// markdown; the Go side delegates to office_oxide for now and
+// will switch to a native Markdown renderer in a follow-up.
+func (p *DOCXParser) ParseWithResult(filename string, data []byte) ParseResult {
+	doc, err := officeOxide.OpenFromBytes(data, "docx")
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("docx open: %w", err)}
+	}
+	defer doc.Close()
+
+	md, err := doc.ToMarkdown()
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("docx to-markdown: %w", err)}
+	}
+
+	return ParseResult{
+		OutputFormat: "markdown",
+		File:         map[string]any{"name": filename, "format": "docx"},
+		Markdown:     md,
+	}
+}
+
+func (p *DOCXParser) String() string {
+	return "DOCXParser"
+}

internal/parser/parser/html_parser.go

@@ -0,0 +1,185 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+
+	"golang.org/x/net/html"
+)
+
+const (
+	Official string = "official"
+)
+
+type HTMLParser struct {
+	libType string
+}
+
+func NewHTMLParser(libType string) (*HTMLParser, error) {
+	switch libType {
+	case Official:
+		return &HTMLParser{
+			libType: Official,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported HTML library type: %s", libType)
+	}
+}
+
+func (p *HTMLParser) String() string {
+	return "HTMLParser"
+}
+
+// ParseWithResult emits one item per block-level HTML element
+// (headings, paragraphs, lists, pre blocks). The walker is a
+// pure-Go replacement for the previous `fmt.Printf` debug output:
+// it descends the html.Parse tree, joins the leaf text of each
+// block-level element, and emits the python-compatible
+// `{text, doc_type_kwd:"text"}` shape.
+//
+// Phase 2.5 (Slice 1) of port-rag-flow-pipeline-to-go.md makes
+// HTMLParser a ParseResultProducer so the dispatch seam routes
+// the html family through the structured path. Inline formatting
+// (bold / links / images) is intentionally NOT surfaced as a
+// separate ck_type — the python HtmlParser collapses inline
+// formatting into the parent block's text.
+func (p *HTMLParser) ParseWithResult(filename string, data []byte) ParseResult {
+	if p.libType != Official {
+		return ParseResult{Err: fmt.Errorf("unsupported HTML library type: %s", p.libType)}
+	}
+	doc, err := html.Parse(bytes.NewReader(data))
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("html parse: %w", err)}
+	}
+	var items []map[string]any
+	walkHTMLBlocks(doc, &items)
+	if items == nil {
+		items = []map[string]any{{"text": "", "doc_type_kwd": "text"}}
+	}
+	return ParseResult{
+		OutputFormat: "json",
+		File: map[string]any{
+			"name":     filename,
+			"encoding": "utf-8",
+		},
+		JSON: items,
+	}
+}
+
+// walkHTMLBlocks emits one normalized item per block-level
+// descendant of root. Inline elements (b, i, a, span, …) are
+// collapsed into the parent's text via leafText. <script> and
+// <style> blocks are skipped entirely so they don't pollute the
+// downstream chunker input.
+func walkHTMLBlocks(root *html.Node, out *[]map[string]any) {
+	for child := root.FirstChild; child != nil; child = child.NextSibling {
+		if child.Type != html.ElementNode {
+			continue
+		}
+		tag := child.Data
+		switch tag {
+		case "script", "style", "noscript":
+			// Skip executable / stylistic blocks entirely.
+			continue
+		case "html", "head", "body":
+			// Wrapper elements: descend into their children.
+			walkHTMLBlocks(child, out)
+			continue
+		}
+		text := htmlLeafText(child)
+		if strings.TrimSpace(text) == "" {
+			continue
+		}
+		*out = append(*out, map[string]any{
+			"text":         strings.TrimSpace(text),
+			"doc_type_kwd": "text",
+			"ck_type":      htmlTagToCkType(tag),
+		})
+	}
+}
+
+// htmlTagToCkType maps HTML block tags to the python `ck_type`
+// vocabulary used downstream by TitleChunker and similar
+// components. Tags not in the map fall back to "text".
+func htmlTagToCkType(tag string) string {
+	switch tag {
+	case "h1", "h2", "h3", "h4", "h5", "h6":
+		return "heading"
+	case "p":
+		return "paragraph"
+	case "ul", "ol", "li":
+		return "list"
+	case "pre", "code":
+		return "code"
+	case "table", "tr", "td", "th":
+		return "table"
+	case "blockquote":
+		return "quote"
+	case "img":
+		return "image"
+	}
+	return "text"
+}
+
+// htmlLeafText joins the visible text of an HTML node and its
+// descendants. <script>/<style> subtrees are skipped (mirrors
+// the python html.parser behaviour). The output preserves
+// whitespace runs so headings like "<h1>Hello   world</h1>"
+// round-trip with their spacing intact.
+func htmlLeafText(n *html.Node) string {
+	var b strings.Builder
+	walkHTMLLeaf(n, &b)
+	return b.String()
+}
+
+func walkHTMLLeaf(n *html.Node, b *strings.Builder) {
+	switch n.Type {
+	case html.TextNode:
+		b.WriteString(n.Data)
+	case html.ElementNode:
+		if n.Data == "script" || n.Data == "style" {
+			return
+		}
+		// Add a line break between block children so headings,
+		// paragraphs, and list items don't run together.
+		switch n.Data {
+		case "h1", "h2", "h3", "h4", "h5", "h6", "p", "li", "pre",
+			"tr", "blockquote":
+			if b.Len() > 0 && !strings.HasSuffix(b.String(), "\n") {
+				b.WriteString("\n")
+			}
+		}
+		for child := n.FirstChild; child != nil; child = child.NextSibling {
+			walkHTMLLeaf(child, b)
+		}
+		if isBlockTag(n.Data) && b.Len() > 0 && !strings.HasSuffix(b.String(), "\n") {
+			b.WriteString("\n")
+		}
+	}
+}
+
+func isBlockTag(tag string) bool {
+	switch tag {
+	case "h1", "h2", "h3", "h4", "h5", "h6", "p", "li", "pre",
+		"tr", "blockquote", "div", "section", "article", "header", "footer":
+		return true
+	}
+	return false
+}

internal/parser/parser/markdown_parser.go

@@ -0,0 +1,167 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+
+	"github.com/gomarkdown/markdown/ast"
+	"github.com/gomarkdown/markdown/parser"
+)
+
+const (
+	GoMarkdown = "go_markdown"
+)
+
+type MarkdownParser struct {
+	libType string
+}
+
+func NewMarkdownParser(libType string) (*MarkdownParser, error) {
+	switch libType {
+	case GoMarkdown:
+		return &MarkdownParser{
+			libType: GoMarkdown,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported Markdown library type: %s", libType)
+	}
+}
+
+// ParseWithResult implements ParseResultProducer (plan §6.5) and
+// returns a structured markdown payload that mirrors the Python
+// parser's `output_format == "json"` shape. Each top-level block
+// emits one item with `text` + `doc_type_kwd: "text"`. The legacy
+// debug-print path has been removed; callers consume ParseResult directly.
+func (p *MarkdownParser) ParseWithResult(filename string, data []byte) ParseResult {
+	if p.libType != GoMarkdown {
+		return ParseResult{Err: fmt.Errorf("unsupported Markdown library type: %s", p.libType)}
+	}
+	doc := markdownNew().Parse(data)
+
+	var items []map[string]any
+	walkMarkdownBlocks(doc, &items)
+	if items == nil {
+		// No blocks emitted — surface a single empty item so the
+		// downstream chunker sees a non-nil JSON slice (mirrors the
+		// Python contract of always producing at least one item).
+		items = []map[string]any{{"text": "", "doc_type_kwd": "text"}}
+	}
+	return ParseResult{
+		OutputFormat: "json",
+		File: map[string]any{
+			"name": filename,
+		},
+		JSON: items,
+	}
+}
+
+func (p *MarkdownParser) String() string {
+	return "MarkdownParser"
+}
+
+// markdownNew is a thin constructor so the extension set is owned
+// in one place (both Parse and ParseWithResult consume it).
+func markdownNew() *parser.Parser {
+	extensions := parser.CommonExtensions | parser.AutoHeadingIDs | parser.NoEmptyLineBeforeBlock
+	return parser.NewWithExtensions(extensions)
+}
+
+// walkMarkdownBlocks emits one normalized item per top-level block.
+// Headings (LeafBlock / H*) are emitted with the heading text so a
+// downstream title chunker can find them; paragraphs (Paragraph)
+// emit the leaf-node text. The walker is intentionally a
+// best-effort pass — full TOC / outline handling lands with the
+// deepdoc/parser port — but it satisfies the per-block "emit
+// text+doc_type_kwd" contract enough for a Phase-1a migration
+// test to verify wire-shape parity.
+func walkMarkdownBlocks(doc ast.Node, out *[]map[string]any) {
+	for _, child := range doc.GetChildren() {
+		switch n := child.(type) {
+		case *ast.Heading:
+			*out = append(*out, map[string]any{
+				"text":         headingText(n),
+				"doc_type_kwd": "text",
+				"ck_type":      "heading",
+			})
+		case *ast.Paragraph:
+			*out = append(*out, map[string]any{
+				"text":         leafText(n),
+				"doc_type_kwd": "text",
+				"ck_type":      "text",
+			})
+		case *ast.List:
+			*out = append(*out, map[string]any{
+				"text":         leafText(n),
+				"doc_type_kwd": "text",
+				"ck_type":      "list",
+			})
+		case *ast.CodeBlock:
+			*out = append(*out, map[string]any{
+				"text":         leafText(n),
+				"doc_type_kwd": "text",
+				"ck_type":      "code",
+			})
+		default:
+			// Block types we don't yet normalize (HTML, tables,
+			// images, definitions) are best-effort: emit the leaf
+			// text without a ck_type so downstream components can
+			// still treat them as text chunks.
+			txt := leafText(n)
+			if strings.TrimSpace(txt) != "" {
+				*out = append(*out, map[string]any{
+					"text":         txt,
+					"doc_type_kwd": "text",
+				})
+			}
+		}
+	}
+}
+
+// headingText returns the inline-text of a heading node by
+// concatenating every Leaf / Text child. Empty headings emit "".
+func headingText(h *ast.Heading) string {
+	var buf bytes.Buffer
+	for _, c := range h.GetChildren() {
+		buf.WriteString(leafText(c))
+	}
+	return strings.TrimSpace(buf.String())
+}
+
+// leafText mirrors gomarkdown's leaf walker: walks every descendant
+// leaf (Text or Inline content) and returns the concatenated UTF-8.
+// Non-text containers that have no leaf descendants return "".
+func leafText(n ast.Node) string {
+	var buf bytes.Buffer
+	walkLeaf(n, &buf)
+	return strings.TrimSpace(buf.String())
+}
+
+func walkLeaf(n ast.Node, buf *bytes.Buffer) {
+	switch t := n.(type) {
+	case *ast.Text:
+		buf.Write(t.Literal)
+	case *ast.Code:
+		buf.Write(t.Literal)
+	default:
+		for _, c := range n.GetChildren() {
+			walkLeaf(c, buf)
+		}
+	}
+}

internal/parser/parser/office_parsers_cgo_shared.go

@@ -0,0 +1,12 @@
+//go:build cgo
+
+package parser
+
+import "html"
+
+// OfficeOxide is the lib_type identifier for office_oxide backend.
+const OfficeOxide = "office_oxide"
+
+func htmlEscape(s string) string {
+	return html.EscapeString(s)
+}

internal/parser/parser/office_parsers_no_cgo.go

@@ -0,0 +1,136 @@
+//go:build !cgo
+
+package parser
+
+import (
+	"errors"
+	"fmt"
+)
+
+// OfficeOxide is the lib_type identifier for office_oxide backend.
+const OfficeOxide = "office_oxide"
+
+// ErrOfficeCGORequired is returned by ParseWithResult on every
+// office-parser family (DOC / DOCX / PPT / PPTX / XLS / XLSX)
+// when the build is not CGO-enabled. The CGO build's
+// implementation captures the office_oxide PlainText / ToMarkdown
+// output; this stub mirrors that surface so the package compiles
+// and existing tests pass.
+var ErrOfficeCGORequired = errors.New("parser: office family requires CGO (office_oxide)")
+
+// docxParseWithResultNoCGO is the no-CGO stub for the DOCX
+// family. The CGO build's implementation lives in docx_parser.go
+// under //go:build cgo.
+func (p *DOCXParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: docx", ErrOfficeCGORequired),
+	}
+}
+
+func (p *DOCParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: doc", ErrOfficeCGORequired),
+	}
+}
+
+func (p *XLSParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: xls", ErrOfficeCGORequired),
+	}
+}
+
+func (p *XLSXParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: xlsx", ErrOfficeCGORequired),
+	}
+}
+
+func (p *PPTParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: ppt", ErrOfficeCGORequired),
+	}
+}
+
+func (p *PPTXParser) ParseWithResult(filename string, _ []byte) ParseResult {
+	return ParseResult{
+		File: map[string]any{"name": filename},
+		Err:  fmt.Errorf("%w: pptx", ErrOfficeCGORequired),
+	}
+}
+
+type DOCParser struct {
+	libType string
+}
+
+func NewDOCParser(libType string) (*DOCParser, error) {
+	return nil, fmt.Errorf("DOC parser requires CGO (office_oxide)")
+}
+
+func (p *DOCParser) String() string {
+	return "DOCParser(no-cgo)"
+}
+
+type DOCXParser struct {
+	libType string
+}
+
+func NewDOCXParser(libType string) (*DOCXParser, error) {
+	return nil, fmt.Errorf("DOCX parser requires CGO (office_oxide)")
+}
+
+func (p *DOCXParser) String() string {
+	return "DOCXParser(no-cgo)"
+}
+
+type XLSParser struct {
+	libType string
+}
+
+func NewXLSParser(libType string) (*XLSParser, error) {
+	return nil, fmt.Errorf("XLS parser requires CGO (office_oxide)")
+}
+
+func (p *XLSParser) String() string {
+	return "XLSParser(no-cgo)"
+}
+
+type XLSXParser struct {
+	libType string
+}
+
+func NewXLSXParser(libType string) (*XLSXParser, error) {
+	return nil, fmt.Errorf("XLSX parser requires CGO (office_oxide)")
+}
+
+func (p *XLSXParser) String() string {
+	return "XLSXParser(no-cgo)"
+}
+
+type PPTParser struct {
+	libType string
+}
+
+func NewPPTParser(libType string) (*PPTParser, error) {
+	return nil, fmt.Errorf("PPT parser requires CGO (office_oxide)")
+}
+
+func (p *PPTParser) String() string {
+	return "PPTParser(no-cgo)"
+}
+
+type PPTXParser struct {
+	libType string
+}
+
+func NewPPTXParser(libType string) (*PPTXParser, error) {
+	return nil, fmt.Errorf("PPTX parser requires CGO (office_oxide)")
+}
+
+func (p *PPTXParser) String() string {
+	return "PPTXParser(no-cgo)"
+}

internal/parser/parser/parse_result.go

@@ -0,0 +1,88 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+// ParseResult is the structured output contract for the Go parser
+// library. Port-rag-flow-pipeline-to-go.md §6.5 mandates that
+// parsers surface enough data to reconstruct a Python-compatible
+// stage-boundary payload:
+//
+//	output_format ∈ {"json","markdown","text","html"}
+//	file         (enriched metadata)
+//	exactly one payload family populated (matching output_format)
+//	err
+//
+// Go parser callers now consume only the structured ParseResult
+// contract. The legacy `Parse(filename, []byte) error` interface has
+// been removed so parser dispatch, ingestion, and service paths all
+// share the same typed payload contract.
+
+package parser
+
+// ParseResult is the structured return value of a successful parse.
+// Exactly one of the payload fields (JSON / Markdown / Text / HTML)
+// is populated on success, matching the Python contract — see
+// port-rag-flow-pipeline-to-go.md §4.2:
+//
+//   - OutputFormat = "json"     → JSON populated
+//   - OutputFormat = "markdown" → Markdown populated
+//   - OutputFormat = "text"     → Text populated
+//   - OutputFormat = "html"     → HTML populated
+//
+// On failure (Err != nil), all payload fields are zero values and
+// OutputFormat is empty.
+type ParseResult struct {
+	// OutputFormat is the wire-compatible format the parser
+	// chose. Empty when Err is non-nil.
+	OutputFormat string
+
+	// File is the enriched file metadata the parser emits. In
+	// Python this is the dict form of the original `file`
+	// descriptor, augmented with format-specific keys (e.g.
+	// `outline` on the PDF path, `page_count` for paginated
+	// formats). Nil when the parser did not enrich.
+	File map[string]any
+
+	// JSON is the structured payload when OutputFormat == "json".
+	// Shape depends on the parser family: PDF emits
+	// `[]map[string]any` with `text` + `doc_type_kwd` keys (and
+	// optional `image` / `layout` / `positions` fields);
+	// markdown / html / text emit normalized
+	// `{text, doc_type_kwd}` items; image emits OCR/VLM result
+	// items. Exactly one payload family is populated on success.
+	JSON []map[string]any
+
+	// Markdown is the string payload when OutputFormat ==
+	// "markdown". Empty otherwise.
+	Markdown string
+
+	// Text is the string payload when OutputFormat == "text".
+	// Empty otherwise.
+	Text string
+
+	// HTML is the string payload when OutputFormat == "html".
+	// Empty otherwise.
+	HTML string
+
+	// Err is the failure reason. On non-nil Err, all payload
+	// fields are zero values.
+	Err error
+}
+
+// ParseResultProducer is the parser package's single structured-output
+// contract. Every parser returned by GetParser must implement it.
+type ParseResultProducer interface {
+	ParseWithResult(filename string, data []byte) ParseResult
+}

internal/parser/parser/parse_result_test.go

@@ -0,0 +1,212 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+package parser
+
+import (
+	"strings"
+	"testing"
+)
+
+// TestParseResult_Contract pins the wire-shape guarantees
+// port-rag-flow-pipeline-to-go.md §6.5 requires:
+//
+//   - exactly one payload family populated on success
+//   - empty payload fields on failure (Err != nil)
+//   - OutputFormat is "" on failure
+//
+// These are the public-key invariants a component-level caller
+// (component/parser.go) relies on. The test does not exercise any
+// specific parser implementation — it pins the contract that the
+// ParseResult type itself enforces, so future parser additions
+// inherit the same guarantees.
+func TestParseResult_Contract(t *testing.T) {
+	cases := []struct {
+		name    string
+		in      ParseResult
+		wantErr bool
+		wantFmt string // expected OutputFormat on success
+	}{
+		{
+			name: "json family only",
+			in: ParseResult{
+				OutputFormat: "json",
+				JSON: []map[string]any{
+					{"text": "hello", "doc_type_kwd": "text"},
+				},
+			},
+			wantFmt: "json",
+		},
+		{
+			name: "markdown family only",
+			in: ParseResult{
+				OutputFormat: "markdown",
+				Markdown:     "# Title\n\nbody",
+			},
+			wantFmt: "markdown",
+		},
+		{
+			name: "text family only",
+			in: ParseResult{
+				OutputFormat: "text",
+				Text:         "raw text",
+			},
+			wantFmt: "text",
+		},
+		{
+			name: "html family only",
+			in: ParseResult{
+				OutputFormat: "html",
+				HTML:         "<p>x</p>",
+			},
+			wantFmt: "html",
+		},
+		{
+			name:    "failure clears payload",
+			in:      ParseResult{Err: errSentinel},
+			wantErr: true,
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			if tc.wantErr {
+				if tc.in.Err == nil {
+					t.Fatal("Err: want non-nil, got nil")
+				}
+				if tc.in.OutputFormat != "" {
+					t.Errorf("OutputFormat on failure = %q, want empty", tc.in.OutputFormat)
+				}
+				if tc.in.JSON != nil || tc.in.Markdown != "" || tc.in.Text != "" || tc.in.HTML != "" {
+					t.Errorf("payload fields must be zero on failure; got %+v", tc.in)
+				}
+				return
+			}
+			if tc.in.Err != nil {
+				t.Errorf("Err on success: want nil, got %v", tc.in.Err)
+			}
+			if tc.in.OutputFormat != tc.wantFmt {
+				t.Errorf("OutputFormat = %q, want %q", tc.in.OutputFormat, tc.wantFmt)
+			}
+			// Exactly-one-payload-family: pick the family declared
+			// by OutputFormat; every other field must be zero.
+			active := payloadFamily(tc.in)
+			for _, other := range allPayloadFamilies {
+				if other == active {
+					continue
+				}
+				if !isPayloadZero(tc.in, other) {
+					t.Errorf("OutputFormat=%s: secondary family %s should be zero", tc.in.OutputFormat, other)
+				}
+			}
+		})
+	}
+}
+
+// errSentinel is a tiny stand-in so the test does not need to
+// import errors just to declare one.
+var errSentinel = sentinelErr("sentinel")
+
+type sentinelErr string
+
+func (s sentinelErr) Error() string { return string(s) }
+
+// TestMarkdownParser_ParseWithResult pins the migration exemplar
+// — port-rag-flow-pipeline-to-go.md §6.5 marks MarkdownParser as
+// the first ported format whose ParseWithResult surface emulates
+// the Python "json" output_format. The fixture is intentionally
+// tiny: 1 heading, 1 paragraph, 1 unordered list item, no nested
+// formatting.
+func TestMarkdownParser_ParseWithResult(t *testing.T) {
+	p, err := NewMarkdownParser(GoMarkdown)
+	if err != nil {
+		t.Fatalf("NewMarkdownParser: %v", err)
+	}
+	src := []byte("# Title\n\nFirst paragraph.\n\n- Item one\n")
+	res := p.ParseWithResult("doc.md", src)
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if res.OutputFormat != "json" {
+		t.Errorf("OutputFormat = %q, want json", res.OutputFormat)
+	}
+	if res.File == nil || res.File["name"] != "doc.md" {
+		t.Errorf("File.name = %v, want doc.md", res.File)
+	}
+	if res.JSON == nil {
+		t.Fatal("JSON: want non-nil slice, got nil")
+	}
+	// 1 heading + 1 paragraph + 1 list = 3 items.
+	if len(res.JSON) != 3 {
+		t.Fatalf("JSON len = %d, want 3 (heading, paragraph, list)", len(res.JSON))
+	}
+	heading := res.JSON[0]
+	if txt, _ := heading["text"].(string); !strings.Contains(txt, "Title") {
+		t.Errorf("JSON[0].text = %q, want contains 'Title'", txt)
+	}
+	if got, _ := heading["ck_type"].(string); got != "heading" {
+		t.Errorf("JSON[0].ck_type = %q, want heading", got)
+	}
+	for i, it := range res.JSON {
+		if _, ok := it["doc_type_kwd"]; !ok {
+			t.Errorf("JSON[%d] missing doc_type_kwd: %+v", i, it)
+		}
+	}
+}
+
+// TestParseResultProducer_PDFIsProducer pins the explicit
+// PDFParser ParseResultProducer wiring. The dispatch seam routes
+// PDFs through ParseWithResult regardless of whether the current
+// build has the cgo-backed DeepDOC engine enabled.
+func TestParseResultProducer_PDFIsProducer(t *testing.T) {
+	pdf := &PDFParser{}
+	if _, ok := any(pdf).(ParseResultProducer); !ok {
+		t.Error("PDFParser must implement ParseResultProducer so the " +
+			"dispatch seam routes PDFs through ParseWithResult")
+	}
+}
+
+// payloadFamily returns the field name that should be populated
+// given the declared OutputFormat. "" when OutputFormat is
+// unrecognized.
+func payloadFamily(r ParseResult) string {
+	switch r.OutputFormat {
+	case "json":
+		return "json"
+	case "markdown":
+		return "markdown"
+	case "text":
+		return "text"
+	case "html":
+		return "html"
+	}
+	return ""
+}
+
+var allPayloadFamilies = []string{"json", "markdown", "text", "html"}
+
+func isPayloadZero(r ParseResult, family string) bool {
+	switch family {
+	case "json":
+		return r.JSON == nil
+	case "markdown":
+		return r.Markdown == ""
+	case "text":
+		return r.Text == ""
+	case "html":
+		return r.HTML == ""
+	}
+	return false
+}

internal/parser/parser/parse_with_result_test.go

@@ -0,0 +1,199 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+// Slice 1 tests for port-rag-flow-pipeline-to-go.md Phase 2.5.
+// These pin the new ParseWithResult contracts for the parsers
+// that did not previously satisfy ParseResultProducer:
+//
+//   - HTMLParser — block-level walker that emits the python-compatible
+//     {text, doc_type_kwd, ck_type} shape.
+//   - TextParser — paragraph-splitting for the text&code family
+//     (.txt / .py / .js / .java / .c / .cpp / .h / .php / .go / .ts
+//     / .sh / .cs / .kt / .sql).
+//
+// MarkdownParser's ParseWithResult is already pinned by
+// parse_result_test.go (prior slice). PDFParser and the office
+// variants remain deferred to a follow-up slice that wires
+// them to the existing internal/deepdoc/parser/pdf pipeline and
+// office_oxide libraries.
+
+package parser
+
+import (
+	"strings"
+	"testing"
+
+	"ragflow/internal/utility"
+)
+
+// TestTextParser_ParseWithResult_ParaSplit pins the paragraph-split
+// rule. A blank-line-separated input yields one item per
+// paragraph; the python TxtParser does the same.
+func TestTextParser_ParseWithResult_ParaSplit(t *testing.T) {
+	p, err := NewTextParser("")
+	if err != nil {
+		t.Fatalf("NewTextParser: %v", err)
+	}
+	src := []byte("First paragraph.\n\nSecond paragraph.\n\nThird.")
+	res := p.ParseWithResult("doc.txt", src)
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if res.OutputFormat != "json" {
+		t.Errorf("OutputFormat = %q, want json", res.OutputFormat)
+	}
+	if got, want := res.File["name"], "doc.txt"; got != want {
+		t.Errorf("File.name = %v, want %v", got, want)
+	}
+	if len(res.JSON) != 3 {
+		t.Fatalf("JSON len = %d, want 3 (one per paragraph)", len(res.JSON))
+	}
+	if got, want := res.JSON[0]["text"], "First paragraph."; got != want {
+		t.Errorf("JSON[0].text = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[2]["text"], "Third."; got != want {
+		t.Errorf("JSON[2].text = %v, want %v", got, want)
+	}
+}
+
+// TestTextParser_ParseWithResult_Empty pins the empty-input
+// fallback (one empty item, not nil) so the downstream chunker
+// sees a non-nil JSON slice. Mirrors the MarkdownParser convention
+// at markdown_parser.go:71-76.
+func TestTextParser_ParseWithResult_Empty(t *testing.T) {
+	p, _ := NewTextParser("")
+	res := p.ParseWithResult("empty.txt", []byte{})
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if len(res.JSON) != 1 {
+		t.Errorf("JSON len = %d, want 1 (empty-input fallback)", len(res.JSON))
+	}
+}
+
+// TestTextParser_ParseWithResult_LongParagraphSlicing pins the
+// maxItemBytes boundary behaviour. A single paragraph longer
+// than 8192 bytes is sliced at the nearest line boundary.
+func TestTextParser_ParseWithResult_LongParagraphSlicing(t *testing.T) {
+	p, _ := NewTextParser("")
+	long := strings.Repeat("a", 9000)
+	res := p.ParseWithResult("long.txt", []byte(long))
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if len(res.JSON) < 2 {
+		t.Errorf("JSON len = %d, want >=2 (sliced at maxItemBytes)", len(res.JSON))
+	}
+	for i, it := range res.JSON {
+		if txt, _ := it["text"].(string); len(txt) > 8192 {
+			t.Errorf("JSON[%d].text len = %d, exceeds maxItemBytes=8192", i, len(txt))
+		}
+	}
+}
+
+// TestTextParser_ParseWithResult_InvalidUTF8 pins the UTF-8
+// validation rule. Invalid bytes produce an error in the result
+// (matching the python TxtParser's behaviour).
+func TestTextParser_ParseWithResult_InvalidUTF8(t *testing.T) {
+	p, _ := NewTextParser("")
+	bad := []byte{0xff, 0xfe, 0xfd}
+	res := p.ParseWithResult("bad.txt", bad)
+	if res.Err == nil {
+		t.Fatal("want error for invalid UTF-8, got nil")
+	}
+}
+
+// TestHTMLParser_ParseWithResult_BlockSplit pins the HTML walker.
+// Three block elements (heading, paragraph, list) yield three
+// items with the python-compatible ck_type vocabulary.
+func TestHTMLParser_ParseWithResult_BlockSplit(t *testing.T) {
+	p, err := NewHTMLParser(Official)
+	if err != nil {
+		t.Fatalf("NewHTMLParser: %v", err)
+	}
+	src := []byte(`<!DOCTYPE html><html><body>
+<h1>Title</h1>
+<p>First paragraph.</p>
+<ul><li>Item one</li></ul>
+</body></html>`)
+	res := p.ParseWithResult("doc.html", src)
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if res.OutputFormat != "json" {
+		t.Errorf("OutputFormat = %q, want json", res.OutputFormat)
+	}
+	if len(res.JSON) != 3 {
+		t.Fatalf("JSON len = %d, want 3 (h1, p, ul)", len(res.JSON))
+	}
+	if got, want := res.JSON[0]["ck_type"], "heading"; got != want {
+		t.Errorf("JSON[0].ck_type = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[0]["text"], "Title"; got != want {
+		t.Errorf("JSON[0].text = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[1]["ck_type"], "paragraph"; got != want {
+		t.Errorf("JSON[1].ck_type = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[1]["text"], "First paragraph."; got != want {
+		t.Errorf("JSON[1].text = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[2]["ck_type"], "list"; got != want {
+		t.Errorf("JSON[2].ck_type = %v, want %v", got, want)
+	}
+	if got, want := res.JSON[2]["text"], "Item one"; got != want {
+		t.Errorf("JSON[2].text = %v, want %v", got, want)
+	}
+}
+
+// TestHTMLParser_ParseWithResult_SkipsScriptAndStyle pins the
+// rule that <script> / <style> subtrees are skipped entirely so
+// they don't pollute the downstream chunker input.
+func TestHTMLParser_ParseWithResult_SkipsScriptAndStyle(t *testing.T) {
+	p, _ := NewHTMLParser(Official)
+	src := []byte(`<html><body>
+<p>Visible.</p>
+<script>alert("x")</script>
+<style>body { color: red; }</style>
+<p>Also visible.</p>
+</body></html>`)
+	res := p.ParseWithResult("doc.html", src)
+	if res.Err != nil {
+		t.Fatalf("ParseWithResult: %v", res.Err)
+	}
+	if len(res.JSON) != 2 {
+		t.Errorf("JSON len = %d, want 2 (script+style skipped)", len(res.JSON))
+	}
+	for _, it := range res.JSON {
+		if txt, _ := it["text"].(string); strings.Contains(txt, "alert") || strings.Contains(txt, "color") {
+			t.Errorf("item text leaks script/style content: %q", txt)
+		}
+	}
+}
+
+// TestGetParser_RoutesTextAndCode pins the parser-type switch
+// routing for the text&code family. After the Slice 1 additions
+// `utility.FileTypeTXT` resolves to a TextParser that satisfies
+// ParseResultProducer.
+func TestGetParser_RoutesTextAndCode(t *testing.T) {
+	p, err := GetParser(utility.FileTypeTXT, map[string]string{"lib_type": ""})
+	if err != nil {
+		t.Fatalf("GetParser(FileTypeTXT): %v", err)
+	}
+	if _, ok := p.(ParseResultProducer); !ok {
+		t.Fatal("TextParser does not implement ParseResultProducer")
+	}
+}

internal/parser/parser/parser_type.go

@@ -0,0 +1,53 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"fmt"
+	"ragflow/internal/utility"
+)
+
+func GetParser(fileType utility.FileType, config map[string]string) (ParseResultProducer, error) {
+	libType, ok := config["lib_type"]
+	if !ok {
+		return nil, fmt.Errorf("missing lib_type config")
+	}
+	switch fileType {
+	case utility.FileTypePPTX:
+		return NewPPTXParser(libType)
+	case utility.FileTypePPT:
+		return NewPPTParser(libType)
+	case utility.FileTypeXLSX:
+		return NewXLSXParser(libType)
+	case utility.FileTypeXLS:
+		return NewXLSParser(libType)
+	case utility.FileTypeDOCX:
+		return NewDOCXParser(libType)
+	case utility.FileTypeDOC:
+		return NewDOCParser(libType)
+	case utility.FileTypePDF:
+		return NewPDFParser(), nil
+	case utility.FileTypeHTML:
+		return NewHTMLParser(Official)
+	case utility.FileTypeMarkdown:
+		return NewMarkdownParser(GoMarkdown)
+	case utility.FileTypeTXT:
+		return NewTextParser(libType)
+	default:
+		return nil, fmt.Errorf("unsupported file type: %s", fileType)
+	}
+}

internal/parser/parser/pdf_parser_cgo.go

@@ -0,0 +1,26 @@
+//go:build cgo
+
+package parser
+
+import (
+	"context"
+	"errors"
+	"fmt"
+
+	deepdocpdf "ragflow/internal/deepdoc/parser/pdf"
+	deepdoctype "ragflow/internal/deepdoc/parser/type"
+)
+
+func (p *PDFParser) ParseWithResult(filename string, data []byte) ParseResult {
+	cfg := deepdoctype.DefaultParserConfig()
+	cfg.SkipOCR = false
+	parser := deepdocpdf.NewParser(cfg)
+	res := parsePDFWithDeepDoc(context.Background(), filename, data, parser.Parse)
+	if res.Err != nil && errors.Is(res.Err, deepdocpdf.ErrNoPDFData) {
+		return ParseResult{Err: fmt.Errorf("%w: %s", ErrPDFEngineUnavailable, filename)}
+	}
+	if res.Err != nil && res.Err.Error() == "deepdoc/pdf: cgo required" {
+		return ParseResult{Err: fmt.Errorf("%w: %s", ErrPDFEngineUnavailable, filename)}
+	}
+	return res
+}

internal/parser/parser/pdf_parser_cgo_test.go

@@ -0,0 +1,49 @@
+//go:build cgo
+
+package parser
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestPDFParser_ParseWithResult_CGOInvalidPDF(t *testing.T) {
+	t.Setenv("DEEPDOC_URL", "")
+	t.Setenv("OSSDEEPDOC_URL", "")
+
+	pdf := NewPDFParser()
+	res := pdf.ParseWithResult("bad.pdf", []byte("not a valid pdf"))
+	if res.Err == nil {
+		t.Fatal("want parse error for invalid PDF bytes, got nil")
+	}
+}
+
+func TestPDFParser_ParseWithResult_CGOEmpty(t *testing.T) {
+	pdf := NewPDFParser()
+	res := pdf.ParseWithResult("empty.pdf", nil)
+	if res.Err != nil {
+		t.Fatalf("empty input: want nil err, got %v", res.Err)
+	}
+	if res.OutputFormat != "json" {
+		t.Fatalf("OutputFormat = %q, want json", res.OutputFormat)
+	}
+	if got, want := res.File["name"], "empty.pdf"; got != want {
+		t.Fatalf("File.name = %v, want %v", got, want)
+	}
+	if got := res.File["page_count"]; got != 0 {
+		t.Fatalf("File.page_count = %v, want 0", got)
+	}
+	outline, ok := res.File["outline"].([]map[string]any)
+	if !ok {
+		t.Fatalf("File.outline type = %T, want []map[string]any", res.File["outline"])
+	}
+	if len(outline) != 0 {
+		t.Fatalf("File.outline len = %d, want 0", len(outline))
+	}
+	if len(res.JSON) != 1 {
+		t.Fatalf("JSON len = %d, want 1", len(res.JSON))
+	}
+	if text, _ := res.JSON[0]["text"].(string); strings.TrimSpace(text) != "" {
+		t.Fatalf("JSON[0].text = %q, want empty", text)
+	}
+}

internal/parser/parser/pdf_parser_common.go

@@ -0,0 +1,181 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"context"
+	"encoding/base64"
+	"errors"
+	"fmt"
+	"os"
+	"strings"
+
+	deepdocpdf "ragflow/internal/deepdoc/parser/pdf"
+	"ragflow/internal/deepdoc/parser/pdf/inference"
+	pdflayout "ragflow/internal/deepdoc/parser/pdf/layout"
+	deepdoctype "ragflow/internal/deepdoc/parser/type"
+)
+
+// ErrPDFEngineUnavailable is returned by PDFParser.ParseWithResult
+// when the current build cannot construct the DeepDOC PDF backend.
+// The normal reason is a non-cgo build, because the pdfoxide bridge
+// is compiled behind `//go:build cgo`.
+var ErrPDFEngineUnavailable = errors.New("parser: PDF backend unavailable in this build")
+
+type PDFParser struct {
+	ParserType string // DeepDoc, PaddleOCR, MinerU
+	Model      string // DeepDoc@buildin@ragflow
+	LibType    string // pdf_oxide, used by DeepDoc
+}
+
+func NewPDFParser() *PDFParser {
+	return &PDFParser{
+		ParserType: "DeepDoc",
+		Model:      "DeepDoc@buildin@ragflow",
+		LibType:    "pdf_oxide",
+	}
+}
+
+func (p *PDFParser) String() string {
+	return "PDFParser"
+}
+
+func emptyPDFResult(filename string) ParseResult {
+	return ParseResult{
+		OutputFormat: "json",
+		File: map[string]any{
+			"name":       filename,
+			"page_count": 0,
+			"outline":    []map[string]any{},
+		},
+		JSON: []map[string]any{{"text": "", "doc_type_kwd": "text"}},
+	}
+}
+
+func deepDocAnalyzerFromEnv() deepdoctype.DocAnalyzer {
+	baseURL := strings.TrimSpace(os.Getenv("DEEPDOC_URL"))
+	if baseURL == "" {
+		baseURL = strings.TrimSpace(os.Getenv("OSSDEEPDOC_URL"))
+	}
+	if baseURL == "" {
+		return &deepdocpdf.MockDocAnalyzer{Healthy: true}
+	}
+	client, err := inference.NewClient(baseURL)
+	if err != nil {
+		return &deepdocpdf.MockDocAnalyzer{Healthy: true}
+	}
+	if !client.Health() {
+		return &deepdocpdf.MockDocAnalyzer{Healthy: true}
+	}
+	return client
+}
+
+func pdfParseResultToJSON(filename string, parsed *deepdoctype.ParseResult) ParseResult {
+	if parsed == nil {
+		return ParseResult{Err: fmt.Errorf("parser: nil DeepDOC PDF result for %s", filename)}
+	}
+	items := pdflayout.SectionsToJSON(parsed.Sections)
+	if len(items) == 0 {
+		items = []map[string]any{{"text": "", "doc_type_kwd": "text"}}
+	}
+	for i := range items {
+		if layoutType, _ := items[i]["layout_type"].(string); layoutType != "" {
+			items[i]["layout"] = layoutType
+		}
+		if _, ok := items[i]["page_number"]; !ok {
+			items[i]["page_number"] = firstPageNumber(items[i]["_pdf_positions"])
+		}
+		if img, _ := items[i]["image"].(string); img != "" {
+			items[i]["image"] = "data:image/png;base64," + img
+		}
+	}
+	return ParseResult{
+		OutputFormat: "json",
+		File: map[string]any{
+			"name":       filename,
+			"page_count": len(parsed.PageImages),
+			"outline":    outlinesToFileMeta(parsed.Outlines),
+		},
+		JSON: items,
+	}
+}
+
+func outlinesToFileMeta(outlines []deepdoctype.Outline) []map[string]any {
+	if len(outlines) == 0 {
+		return []map[string]any{}
+	}
+	result := make([]map[string]any, 0, len(outlines))
+	for _, o := range outlines {
+		result = append(result, map[string]any{
+			"title":       o.Title,
+			"level":       o.Level,
+			"page_number": o.PageNumber,
+		})
+	}
+	return result
+}
+
+func firstPageNumber(raw any) int {
+	positions, ok := raw.([][]any)
+	if !ok || len(positions) == 0 || len(positions[0]) == 0 {
+		return 0
+	}
+	pages, ok := positions[0][0].([]any)
+	if !ok || len(pages) == 0 {
+		return 0
+	}
+	switch v := pages[0].(type) {
+	case int:
+		return v
+	case int64:
+		return int(v)
+	case float64:
+		return int(v)
+	default:
+		return 0
+	}
+}
+
+func inlinePNGDataURL(raw string) string {
+	if raw == "" {
+		return ""
+	}
+	if strings.HasPrefix(raw, "data:image/") {
+		return raw
+	}
+	if _, err := base64.StdEncoding.DecodeString(raw); err != nil {
+		return raw
+	}
+	return "data:image/png;base64," + raw
+}
+
+func parsePDFWithDeepDoc(ctx context.Context, filename string, data []byte, parseFn func(context.Context, []byte, deepdoctype.DocAnalyzer) (*deepdoctype.ParseResult, error)) ParseResult {
+	if len(data) == 0 {
+		return emptyPDFResult(filename)
+	}
+	parsed, err := parseFn(ctx, data, deepDocAnalyzerFromEnv())
+	if err != nil {
+		return ParseResult{Err: err}
+	}
+	res := pdfParseResultToJSON(filename, parsed)
+	for i := range res.JSON {
+		if img, _ := res.JSON[i]["image"].(string); img != "" {
+			res.JSON[i]["image"] = inlinePNGDataURL(img)
+		}
+	}
+	return res
+}

internal/parser/parser/pdf_parser_nocgo.go

@@ -0,0 +1,16 @@
+//go:build !cgo
+
+package parser
+
+import (
+	"fmt"
+)
+
+func (p *PDFParser) ParseWithResult(filename string, data []byte) ParseResult {
+	if len(data) == 0 {
+		return emptyPDFResult(filename)
+	}
+	return ParseResult{
+		Err: fmt.Errorf("%w: %s", ErrPDFEngineUnavailable, filename),
+	}
+}

internal/parser/parser/pdf_parser_nocgo_test.go

@@ -0,0 +1,31 @@
+//go:build !cgo
+
+package parser
+
+import (
+	"errors"
+	"testing"
+)
+
+func TestPDFParser_ParseWithResult_NoCGO(t *testing.T) {
+	pdf := NewPDFParser()
+
+	empty := pdf.ParseWithResult("empty.pdf", nil)
+	if empty.Err != nil {
+		t.Fatalf("empty input: want nil err, got %v", empty.Err)
+	}
+	if empty.OutputFormat != "json" {
+		t.Fatalf("empty input OutputFormat = %q, want json", empty.OutputFormat)
+	}
+	if len(empty.JSON) != 1 {
+		t.Fatalf("empty input JSON len = %d, want 1", len(empty.JSON))
+	}
+
+	res := pdf.ParseWithResult("a.pdf", []byte("%PDF-1.4"))
+	if res.Err == nil {
+		t.Fatal("want ErrPDFEngineUnavailable, got nil")
+	}
+	if !errors.Is(res.Err, ErrPDFEngineUnavailable) {
+		t.Fatalf("err = %v, want wraps ErrPDFEngineUnavailable", res.Err)
+	}
+}

internal/parser/parser/ppt_parser.go

@@ -0,0 +1,58 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"fmt"
+)
+
+type PPTParser struct {
+	libType string
+}
+
+func NewPPTParser(libType string) (*PPTParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &PPTParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported PPT library type: %s", libType)
+	}
+}
+
+func (p *PPTParser) String() string {
+	return "PPTParser"
+}
+
+// ParseWithResult delegates to PPTXParser's structured output
+// for the legacy PPT format. The two file families differ only
+// in the binary container; the python parser.py:slides branch
+// treats them uniformly.
+func (p *PPTParser) ParseWithResult(filename string, data []byte) ParseResult {
+	delegate, err := NewPPTXParser(OfficeOxide)
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("ppt delegate: %w", err)}
+	}
+	res := delegate.ParseWithResult(filename, data)
+	if res.File != nil {
+		res.File["format"] = "ppt"
+	}
+	return res
+}

internal/parser/parser/pptx_parser.go

@@ -0,0 +1,85 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"fmt"
+	"strings"
+
+	officeOxide "github.com/yfedoseev/office_oxide/go"
+)
+
+type PPTXParser struct {
+	libType string
+}
+
+func NewPPTXParser(libType string) (*PPTXParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &PPTXParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported PPTX library type: %s", libType)
+	}
+}
+
+func (p *PPTXParser) String() string {
+	return "PPTXParser"
+}
+
+// ParseWithResult emits one JSON item per slide with the slide's
+// plain text. Mirrors the python parser.py:slides branch which
+// forces output_format="json" for the slide family.
+func (p *PPTXParser) ParseWithResult(filename string, data []byte) ParseResult {
+	doc, err := officeOxide.OpenFromBytes(data, "pptx")
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("pptx open: %w", err)}
+	}
+	defer doc.Close()
+
+	text, err := doc.PlainText()
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("pptx plain-text: %w", err)}
+	}
+
+	// Split on form-feed (the python TxtParser convention used by
+	// ragflow's slide parser) — each block becomes a JSON item.
+	var items []map[string]any
+	for i, raw := range strings.Split(text, "\f") {
+		trimmed := strings.TrimSpace(raw)
+		if trimmed == "" {
+			continue
+		}
+		items = append(items, map[string]any{
+			"text":         trimmed,
+			"doc_type_kwd": "text",
+			"slide_number": i + 1,
+		})
+	}
+	if items == nil {
+		items = []map[string]any{{"text": strings.TrimSpace(text), "doc_type_kwd": "text"}}
+	}
+
+	return ParseResult{
+		OutputFormat: "json",
+		File:         map[string]any{"name": filename, "format": "pptx"},
+		JSON:         items,
+	}
+}

internal/parser/parser/text_parser.go

@@ -0,0 +1,178 @@
+//
+//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+//  Licensed under the Apache License, Version 2.0 (the "License");
+//  you may not use this file except in compliance with the License.
+//  You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+//  Unless required by applicable law or agreed to in writing, software
+//  distributed under the License is distributed on an "AS IS" BASIS,
+//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+//  See the License for the specific language governing permissions and
+//  limitations under the License.
+//
+
+// TextParser (port-rag-flow-pipeline-to-go.md Phase 2.5 Slice 1).
+//
+// The python rag/flow/parser/parser.py:_code path (L1066) routes
+// .txt / .py / .js / .java / .c / .cpp / .h / .php / .go / .ts / .sh
+// / .cs / .kt / .sql files through deepdoc.parser.TxtParser. The Go
+// side needs a parser for these families so `text&code` resolves to a
+// real ParseResultProducer.
+//
+// TextParser fills that gap with a minimal but real implementation:
+// it splits the input into paragraph-sized items and emits the
+// python-compatible `{text, doc_type_kwd:"text"}` shape. The
+// python TxtParser additionally does layout-aware section
+// detection; the Go version is intentionally simpler because (a)
+// no production template currently relies on text&code for richer
+// structure than paragraph items.
+
+package parser
+
+import (
+	"bytes"
+	"strings"
+)
+
+const TextParserLibType = "text"
+
+// TextParser is the text&code family parser. It implements the
+// structured ParseResultProducer contract directly.
+type TextParser struct {
+	// maxItemBytes caps each emitted item's text length. The
+	// python TxtParser uses similar paragraph-style chunking;
+	// 8192 bytes is a conservative ceiling that prevents the
+	// downstream chunker from receiving oversized inputs.
+	maxItemBytes int
+}
+
+// NewTextParser constructs a TextParser. The libType argument
+// preserves the parser-library constructor signature for
+// consistency with the other family parsers; the value is ignored
+// (TextParser has no alternative backend).
+func NewTextParser(_ string) (*TextParser, error) {
+	return &TextParser{maxItemBytes: 8192}, nil
+}
+
+// ParseWithResult emits one item per non-empty paragraph. The
+// output format is "json" to mirror the python TxtParser's
+// behaviour (it emits a list of items with text + doc_type_kwd).
+//
+// The items slice is always non-nil so downstream chunkers see a
+// non-empty JSON payload even for an empty input (mirrors the
+// MarkdownParser convention at markdown_parser.go:71-76).
+func (p *TextParser) ParseWithResult(filename string, data []byte) ParseResult {
+	if !utf8Valid(data) {
+		return ParseResult{Err: errInvalidUTF8}
+	}
+	items := textParserItems(data, p.maxItemBytes)
+	if items == nil {
+		items = []map[string]any{{"text": "", "doc_type_kwd": "text"}}
+	}
+	return ParseResult{
+		OutputFormat: "json",
+		File: map[string]any{
+			"name":     filename,
+			"size":     len(data),
+			"encoding": "utf-8",
+		},
+		JSON: items,
+	}
+}
+
+func (p *TextParser) String() string {
+	return "TextParser"
+}
+
+// errInvalidUTF8 is returned when the input bytes fail UTF-8
+// validation. Matches the python TxtParser's behaviour of
+// surfacing a clear error rather than emitting replacement bytes.
+var errInvalidUTF8 = errInvalidUTF8Sentinel("parser: text input is not valid UTF-8")
+
+type errInvalidUTF8Sentinel string
+
+func (e errInvalidUTF8Sentinel) Error() string { return string(e) }
+
+// utf8Valid is a tiny stdlib-free validator. We avoid
+// unicode/utf8.Valid to keep this file dependency-light; the
+// validation rule is the same (decode without rejecting bytes).
+func utf8Valid(data []byte) bool {
+	for i := 0; i < len(data); {
+		r, size := decodeRune(data[i:])
+		if r == 0xFFFD && size == 1 {
+			return false
+		}
+		i += size
+	}
+	return true
+}
+
+// decodeRune is a minimal UTF-8 decoder that mirrors
+// utf8.DecodeRune's signature: returns the rune and its byte
+// width. Returns (RuneError, 1) on invalid sequences, matching
+// the stdlib contract.
+func decodeRune(p []byte) (rune, int) {
+	if len(p) == 0 {
+		return 0xFFFD, 0
+	}
+	c := p[0]
+	switch {
+	case c < 0x80:
+		return rune(c), 1
+	case c < 0xC2:
+		return 0xFFFD, 1
+	case c < 0xE0:
+		if len(p) < 2 || p[1]&0xC0 != 0x80 {
+			return 0xFFFD, 1
+		}
+		return rune(c&0x1F)<<6 | rune(p[1]&0x3F), 2
+	case c < 0xF0:
+		if len(p) < 3 || p[1]&0xC0 != 0x80 || p[2]&0xC0 != 0x80 {
+			return 0xFFFD, 1
+		}
+		return rune(c&0x0F)<<12 | rune(p[1]&0x3F)<<6 | rune(p[2]&0x3F), 3
+	case c < 0xF5:
+		if len(p) < 4 || p[1]&0xC0 != 0x80 || p[2]&0xC0 != 0x80 || p[3]&0xC0 != 0x80 {
+			return 0xFFFD, 1
+		}
+		return rune(c&0x07)<<18 | rune(p[1]&0x3F)<<12 | rune(p[2]&0x3F)<<6 | rune(p[3]&0x3F), 4
+	}
+	return 0xFFFD, 1
+}
+
+// textParserItems splits `data` into paragraph-sized chunks. The
+// split rule mirrors the python TxtParser: blank lines separate
+// paragraphs; long paragraphs are sliced at maxItemBytes boundaries.
+func textParserItems(data []byte, maxItemBytes int) []map[string]any {
+	var items []map[string]any
+	for _, raw := range bytes.Split(data, []byte("\n\n")) {
+		text := strings.TrimSpace(string(raw))
+		if text == "" {
+			continue
+		}
+		if maxItemBytes > 0 && len(text) > maxItemBytes {
+			// Slice at the nearest newline below maxItemBytes;
+			// falls back to a hard slice when no newline exists.
+			cut := strings.LastIndex(text[:maxItemBytes], "\n")
+			if cut <= 0 {
+				cut = maxItemBytes
+			}
+			items = append(items, map[string]any{
+				"text":         strings.TrimSpace(text[:cut]),
+				"doc_type_kwd": "text",
+			})
+			text = strings.TrimSpace(text[cut:])
+			if text == "" {
+				continue
+			}
+		}
+		items = append(items, map[string]any{
+			"text":         text,
+			"doc_type_kwd": "text",
+		})
+	}
+	return items
+}

internal/parser/parser/xls_parser.go

@@ -0,0 +1,85 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+
+	"github.com/xuri/excelize/v2"
+)
+
+type XLSParser struct {
+	libType string
+}
+
+func NewXLSParser(libType string) (*XLSParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &XLSParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported XLS library type: %s", libType)
+	}
+}
+
+func (p *XLSParser) String() string {
+	return "XLSParser"
+}
+
+// ParseWithResult delegates to excelize which handles both .xls
+// and .xlsx through the same API. The python ExcelParser falls
+// back to a similar delegation; on the Go side excelize is the
+// single library for both extensions.
+func (p *XLSParser) ParseWithResult(filename string, data []byte) ParseResult {
+	f, err := excelize.OpenReader(bytes.NewReader(data))
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("xls open: %w", err)}
+	}
+	defer f.Close()
+
+	var html strings.Builder
+	html.WriteString("<html><body>")
+	for _, sheet := range f.GetSheetList() {
+		html.WriteString("<h3>")
+		html.WriteString(sheet)
+		html.WriteString("</h3>")
+		rows, _ := f.GetRows(sheet)
+		html.WriteString("<table>")
+		for _, row := range rows {
+			html.WriteString("<tr>")
+			for _, cell := range row {
+				html.WriteString("<td>")
+				html.WriteString(htmlEscape(cell))
+				html.WriteString("</td>")
+			}
+			html.WriteString("</tr>")
+		}
+		html.WriteString("</table>")
+	}
+	html.WriteString("</body></html>")
+
+	return ParseResult{
+		OutputFormat: "html",
+		File:         map[string]any{"name": filename, "format": "xls"},
+		HTML:         html.String(),
+	}
+}

internal/parser/parser/xlsx_parser.go

@@ -0,0 +1,91 @@
+//go:build cgo
+
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package parser
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+
+	"github.com/xuri/excelize/v2"
+)
+
+type XLSXParser struct {
+	libType string
+}
+
+func NewXLSXParser(libType string) (*XLSXParser, error) {
+	switch libType {
+	case OfficeOxide:
+		return &XLSXParser{
+			libType: OfficeOxide,
+		}, nil
+	default:
+		return nil, fmt.Errorf("unsupported XLSX library type: %s", libType)
+	}
+}
+
+func (p *XLSXParser) String() string {
+	return "XLSXParser"
+}
+
+// ParseWithResult renders the spreadsheet as HTML — the python
+// ExcelParser shape. Each sheet becomes a <table> with row /
+// column structure preserved; sheet names become <h3> headings
+// so a downstream title chunker can pick them up. Implementation
+// uses excelize (already in go.mod) instead of office_oxide's
+// PlainText/ToMarkdown so cell boundaries survive the round-trip.
+func (p *XLSXParser) ParseWithResult(filename string, data []byte) ParseResult {
+	f, err := excelize.OpenReader(bytes.NewReader(data))
+	if err != nil {
+		return ParseResult{Err: fmt.Errorf("xlsx open: %w", err)}
+	}
+	defer f.Close()
+
+	sheets := f.GetSheetList()
+	var html strings.Builder
+	html.WriteString("<html><body>")
+	for _, sheet := range sheets {
+		html.WriteString("<h3>")
+		html.WriteString(sheet)
+		html.WriteString("</h3>")
+		rows, err := f.GetRows(sheet)
+		if err != nil {
+			continue
+		}
+		html.WriteString("<table>")
+		for _, row := range rows {
+			html.WriteString("<tr>")
+			for _, cell := range row {
+				html.WriteString("<td>")
+				html.WriteString(htmlEscape(cell))
+				html.WriteString("</td>")
+			}
+			html.WriteString("</tr>")
+		}
+		html.WriteString("</table>")
+	}
+	html.WriteString("</body></html>")
+
+	return ParseResult{
+		OutputFormat: "html",
+		File:         map[string]any{"name": filename, "format": "xlsx", "sheets": len(sheets)},
+		HTML:         html.String(),
+	}
+}