Files
openclaw/scripts/docs-i18n/doc_chunked_raw.go
Mason 164287f056 docs-i18n: avoid ambiguous body-only wrapper unwrap (#63808)
* docs-i18n: avoid ambiguous body-only wrapper unwrap

* docs: clarify targeted testing tip

* changelog: include docs-i18n follow-up thanks
2026-04-10 00:01:17 +08:00

831 lines
24 KiB
Go

package main
import (
"context"
"fmt"
"log"
"os"
"regexp"
"slices"
"strconv"
"strings"
)
const defaultDocChunkMaxBytes = 12000
const defaultDocChunkPromptBudget = 15000
var (
docsFenceRE = regexp.MustCompile(`^\s*(` + "```" + `|~~~)`)
docsComponentTagRE = regexp.MustCompile(`<(/?)([A-Z][A-Za-z0-9]*)\b[^>]*?/?>`)
)
var docsProtocolTokens = []string{
frontmatterTagStart,
frontmatterTagEnd,
bodyTagStart,
bodyTagEnd,
"[[[FM_",
}
type docChunkStructure struct {
fenceCount int
tagCounts map[string]int
}
type docChunkSplitPlan struct {
groups [][]string
reason string
}
func translateDocBodyChunked(ctx context.Context, translator docsTranslator, relPath, body, srcLang, tgtLang string) (string, error) {
if strings.TrimSpace(body) == "" {
return body, nil
}
blocks := splitDocBodyIntoBlocks(body)
groups := groupDocBlocks(blocks, docsI18nDocChunkMaxBytes())
logDocChunkPlan(relPath, blocks, groups)
out := strings.Builder{}
for index, group := range groups {
chunkID := fmt.Sprintf("%s.chunk-%03d", relPath, index+1)
translated, err := translateDocBlockGroup(ctx, translator, chunkID, group, srcLang, tgtLang)
if err != nil {
return "", err
}
out.WriteString(translated)
}
return out.String(), nil
}
func translateDocBlockGroup(ctx context.Context, translator docsTranslator, chunkID string, blocks []string, srcLang, tgtLang string) (string, error) {
source := strings.Join(blocks, "")
if strings.TrimSpace(source) == "" {
return source, nil
}
if plan, ok := planDocChunkSplit(blocks, docsI18nDocChunkMaxBytes(), docsI18nDocChunkPromptBudget()); ok {
logDocChunkPlanSplit(chunkID, plan, source)
return translatePlannedDocChunkGroups(ctx, translator, chunkID, plan.groups, srcLang, tgtLang)
}
normalizedSource, commonIndent := stripCommonIndent(source)
log.Printf("docs-i18n: chunk start %s blocks=%d bytes=%d", chunkID, len(blocks), len(source))
translated, err := translator.TranslateRaw(ctx, normalizedSource, srcLang, tgtLang)
if err == nil {
translated = sanitizeDocChunkProtocolWrappers(source, translated)
translated = reapplyCommonIndent(translated, commonIndent)
if validationErr := validateDocChunkTranslation(source, translated); validationErr == nil {
log.Printf("docs-i18n: chunk done %s out_bytes=%d", chunkID, len(translated))
return translated, nil
} else {
err = validationErr
}
}
if len(blocks) <= 1 {
if fallback, fallbackErr := translateDocLeafBlock(ctx, translator, chunkID, source, srcLang, tgtLang); fallbackErr == nil {
return fallback, nil
}
if plan, ok := planSingletonDocChunkRetry(source, docsI18nDocChunkMaxBytes(), docsI18nDocChunkPromptBudget()); ok {
logDocChunkPlanSplit(chunkID, plan, source)
return translatePlannedDocChunkGroups(ctx, translator, chunkID, plan.groups, srcLang, tgtLang)
}
return "", fmt.Errorf("%s: %w", chunkID, err)
}
if plan, ok := planDocChunkSplit(blocks, docsI18nDocChunkMaxBytes(), docsI18nDocChunkPromptBudget()); ok {
logDocChunkSplit(chunkID, len(blocks), err)
return translatePlannedDocChunkGroups(ctx, translator, chunkID, plan.groups, srcLang, tgtLang)
}
if plan, ok := splitDocChunkBlocksMidpointSimple(blocks); ok {
logDocChunkSplit(chunkID, len(blocks), err)
return translatePlannedDocChunkGroups(ctx, translator, chunkID, plan.groups, srcLang, tgtLang)
}
return "", fmt.Errorf("%s: %w", chunkID, err)
}
func translateDocLeafBlock(ctx context.Context, translator docsTranslator, chunkID, source, srcLang, tgtLang string) (string, error) {
sourceStructure := summarizeDocChunkStructure(source)
if sourceStructure.fenceCount != 0 {
return "", fmt.Errorf("%s: raw leaf fallback not applicable", chunkID)
}
normalizedSource, commonIndent := stripCommonIndent(source)
maskedSource, placeholders := maskDocComponentTags(normalizedSource)
translated, err := translator.Translate(ctx, maskedSource, srcLang, tgtLang)
if err != nil {
return "", err
}
translated, err = restoreDocComponentTags(translated, placeholders)
if err != nil {
return "", err
}
translated = sanitizeDocChunkProtocolWrappers(source, translated)
translated = reapplyCommonIndent(translated, commonIndent)
if validationErr := validateDocChunkTranslation(source, translated); validationErr != nil {
return "", validationErr
}
log.Printf("docs-i18n: chunk leaf-fallback done %s out_bytes=%d", chunkID, len(translated))
return translated, nil
}
func splitDocBodyIntoBlocks(body string) []string {
if body == "" {
return nil
}
lines := strings.SplitAfter(body, "\n")
blocks := make([]string, 0, len(lines))
var current strings.Builder
fenceDelimiter := ""
for _, line := range lines {
current.WriteString(line)
fenceDelimiter, _ = updateFenceDelimiter(fenceDelimiter, line)
inFence := fenceDelimiter != ""
if !inFence && strings.TrimSpace(line) == "" {
blocks = append(blocks, current.String())
current.Reset()
}
}
if current.Len() > 0 {
blocks = append(blocks, current.String())
}
if len(blocks) == 0 {
return []string{body}
}
return blocks
}
func groupDocBlocks(blocks []string, maxBytes int) [][]string {
if len(blocks) == 0 {
return nil
}
if maxBytes <= 0 {
maxBytes = defaultDocChunkMaxBytes
}
groups := make([][]string, 0, len(blocks))
current := make([]string, 0, 8)
currentBytes := 0
flush := func() {
if len(current) == 0 {
return
}
groups = append(groups, current)
current = make([]string, 0, 8)
currentBytes = 0
}
for _, block := range blocks {
blockBytes := len(block)
if len(current) > 0 && currentBytes+blockBytes > maxBytes {
flush()
}
if blockBytes > maxBytes {
groups = append(groups, []string{block})
continue
}
current = append(current, block)
currentBytes += blockBytes
}
flush()
return groups
}
func validateDocChunkTranslation(source, translated string) error {
if hasUnexpectedTopLevelProtocolWrapper(source, translated) {
return fmt.Errorf("protocol token leaked: top-level wrapper")
}
sourceLower := strings.ToLower(source)
translatedLower := strings.ToLower(translated)
for _, token := range docsProtocolTokens {
tokenLower := strings.ToLower(token)
if strings.Contains(sourceLower, tokenLower) {
continue
}
if strings.Contains(translatedLower, tokenLower) {
return fmt.Errorf("protocol token leaked: %s", token)
}
}
sourceStructure := summarizeDocChunkStructure(source)
translatedStructure := summarizeDocChunkStructure(translated)
if sourceStructure.fenceCount != translatedStructure.fenceCount {
return fmt.Errorf("code fence mismatch: source=%d translated=%d", sourceStructure.fenceCount, translatedStructure.fenceCount)
}
if !slices.Equal(sortedKeys(sourceStructure.tagCounts), sortedKeys(translatedStructure.tagCounts)) {
return fmt.Errorf("component tag set mismatch")
}
for _, key := range sortedKeys(sourceStructure.tagCounts) {
if sourceStructure.tagCounts[key] != translatedStructure.tagCounts[key] {
return fmt.Errorf("component tag mismatch for %s: source=%d translated=%d", key, sourceStructure.tagCounts[key], translatedStructure.tagCounts[key])
}
}
return nil
}
func sanitizeDocChunkProtocolWrappers(source, translated string) string {
if !containsProtocolWrapperToken(translated) {
return translated
}
trimmedTranslated := strings.TrimSpace(translated)
if !hasUnexpectedTopLevelProtocolWrapper(source, trimmedTranslated) {
return translated
}
if !hasAmbiguousTaggedBodyClose(source, trimmedTranslated) {
_, body, err := parseTaggedDocument(trimmedTranslated)
if err == nil {
if strings.TrimSpace(body) == "" {
return translated
}
return body
}
}
body, ok := stripBodyOnlyWrapper(source, trimmedTranslated)
if !ok || strings.TrimSpace(body) == "" {
return translated
}
return body
}
func stripBodyOnlyWrapper(source, text string) (string, bool) {
sourceLower := strings.ToLower(source)
// When the source itself documents <body> tokens, a bare body-only payload is
// ambiguous: the trailing </body> can be literal translated content instead of
// a real wrapper close. Keep it for validation/retry instead of truncating.
if strings.Contains(sourceLower, strings.ToLower(bodyTagStart)) || strings.Contains(sourceLower, strings.ToLower(bodyTagEnd)) {
return "", false
}
lower := strings.ToLower(text)
bodyStartLower := strings.ToLower(bodyTagStart)
bodyEndLower := strings.ToLower(bodyTagEnd)
if !strings.HasPrefix(lower, bodyStartLower) || !strings.HasSuffix(lower, bodyEndLower) {
return "", false
}
body := text[len(bodyTagStart) : len(text)-len(bodyTagEnd)]
bodyLower := lower[len(bodyTagStart) : len(lower)-len(bodyTagEnd)]
if strings.Contains(bodyLower, bodyStartLower) || strings.Contains(bodyLower, bodyEndLower) {
return "", false
}
return trimTagNewlines(body), true
}
func hasAmbiguousTaggedBodyClose(source, translated string) bool {
sourceLower := strings.ToLower(source)
if !strings.Contains(sourceLower, strings.ToLower(bodyTagStart)) && !strings.Contains(sourceLower, strings.ToLower(bodyTagEnd)) {
return false
}
translatedLower := strings.ToLower(translated)
if !strings.Contains(translatedLower, strings.ToLower(frontmatterTagStart)) {
return false
}
return strings.Count(translatedLower, strings.ToLower(bodyTagEnd)) == 1
}
func maskDocComponentTags(text string) (string, []string) {
placeholders := make([]string, 0, 4)
masked := docsComponentTagRE.ReplaceAllStringFunc(text, func(match string) string {
placeholder := fmt.Sprintf("__OC_DOC_TAG_%03d__", len(placeholders))
placeholders = append(placeholders, match)
return placeholder
})
return masked, placeholders
}
func restoreDocComponentTags(text string, placeholders []string) (string, error) {
restored := text
for index, original := range placeholders {
placeholder := fmt.Sprintf("__OC_DOC_TAG_%03d__", index)
if !strings.Contains(restored, placeholder) {
return "", fmt.Errorf("component tag placeholder missing: %s", placeholder)
}
restored = strings.ReplaceAll(restored, placeholder, original)
}
return restored, nil
}
func logDocChunkSplit(chunkID string, blockCount int, err error) {
if docsI18nVerboseLogs() || blockCount >= 16 {
log.Printf("docs-i18n: chunk split %s blocks=%d err=%v", chunkID, blockCount, err)
}
}
func logDocChunkPlanSplit(chunkID string, plan docChunkSplitPlan, source string) {
if plan.reason == "" {
plan.reason = "unknown"
}
log.Printf("docs-i18n: chunk pre-split %s reason=%s groups=%d bytes=%d", chunkID, plan.reason, len(plan.groups), len(source))
}
func summarizeDocChunkStructure(text string) docChunkStructure {
counts := map[string]int{}
lines := strings.Split(text, "\n")
fenceDelimiter := ""
for _, line := range lines {
var toggled bool
fenceDelimiter, toggled = updateFenceDelimiter(fenceDelimiter, line)
if toggled {
counts["__fence_toggle__"]++
}
for _, match := range docsComponentTagRE.FindAllStringSubmatch(line, -1) {
if len(match) < 3 {
continue
}
fullToken := match[0]
tagName := match[2]
direction := "open"
if match[1] == "/" {
direction = "close"
}
if strings.HasSuffix(fullToken, "/>") {
direction = "self"
}
counts[tagName+":"+direction]++
}
}
return docChunkStructure{
fenceCount: counts["__fence_toggle__"],
tagCounts: countsWithoutFence(counts),
}
}
func countsWithoutFence(counts map[string]int) map[string]int {
filtered := map[string]int{}
for key, value := range counts {
if key == "__fence_toggle__" {
continue
}
filtered[key] = value
}
return filtered
}
func sortedKeys(counts map[string]int) []string {
keys := make([]string, 0, len(counts))
for key := range counts {
keys = append(keys, key)
}
slices.Sort(keys)
return keys
}
func updateFenceDelimiter(current, line string) (string, bool) {
delimiter := leadingFenceDelimiter(line)
if delimiter == "" {
return current, false
}
if current == "" {
return delimiter, true
}
if delimiter[0] == current[0] && len(delimiter) >= len(current) && isClosingFenceLine(line, delimiter) {
return "", true
}
return current, false
}
func leadingFenceDelimiter(line string) string {
trimmed := strings.TrimLeft(line, " \t")
if len(trimmed) < 3 {
return ""
}
switch trimmed[0] {
case '`', '~':
default:
return ""
}
marker := trimmed[0]
index := 0
for index < len(trimmed) && trimmed[index] == marker {
index++
}
if index < 3 {
return ""
}
return trimmed[:index]
}
func isClosingFenceLine(line, delimiter string) bool {
trimmed := strings.TrimLeft(line, " \t")
if !strings.HasPrefix(trimmed, delimiter) {
return false
}
return strings.TrimSpace(trimmed[len(delimiter):]) == ""
}
func hasUnexpectedTopLevelProtocolWrapper(source, translated string) bool {
sourceTrimmed := strings.ToLower(strings.TrimSpace(source))
translatedTrimmed := strings.ToLower(strings.TrimSpace(translated))
checks := []struct {
token string
match func(string) bool
}{
{token: frontmatterTagStart, match: func(text string) bool { return strings.HasPrefix(text, strings.ToLower(frontmatterTagStart)) }},
{token: bodyTagStart, match: func(text string) bool { return strings.HasPrefix(text, strings.ToLower(bodyTagStart)) }},
{token: frontmatterTagEnd, match: func(text string) bool { return strings.HasSuffix(text, strings.ToLower(frontmatterTagEnd)) }},
{token: bodyTagEnd, match: func(text string) bool { return strings.HasSuffix(text, strings.ToLower(bodyTagEnd)) }},
}
for _, check := range checks {
if check.match(translatedTrimmed) && !check.match(sourceTrimmed) {
return true
}
}
return false
}
func containsProtocolWrapperToken(text string) bool {
lower := strings.ToLower(text)
return strings.Contains(lower, strings.ToLower(bodyTagStart)) || strings.Contains(lower, strings.ToLower(frontmatterTagStart))
}
func translatePlannedDocChunkGroups(ctx context.Context, translator docsTranslator, chunkID string, groups [][]string, srcLang, tgtLang string) (string, error) {
var out strings.Builder
for index, group := range groups {
translated, err := translateDocBlockGroup(ctx, translator, fmt.Sprintf("%s.%02d", chunkID, index+1), group, srcLang, tgtLang)
if err != nil {
return "", err
}
out.WriteString(translated)
}
return out.String(), nil
}
func planDocChunkSplit(blocks []string, maxBytes, promptBudget int) (docChunkSplitPlan, bool) {
if len(blocks) == 0 {
return docChunkSplitPlan{}, false
}
source := strings.Join(blocks, "")
if strings.TrimSpace(source) == "" {
return docChunkSplitPlan{}, false
}
normalizedSource, _ := stripCommonIndent(source)
estimatedPromptCost := estimateDocPromptCost(normalizedSource)
if len(blocks) > 1 && promptBudget > 0 && estimatedPromptCost > promptBudget {
return splitDocChunkBlocksMidpoint(blocks, estimatedPromptCost, promptBudget)
}
if len(blocks) == 1 {
return planSingletonDocChunk(blocks[0], maxBytes, promptBudget)
}
return docChunkSplitPlan{}, false
}
func splitDocChunkBlocksMidpoint(blocks []string, estimatedPromptCost, promptBudget int) (docChunkSplitPlan, bool) {
if len(blocks) <= 1 {
return docChunkSplitPlan{}, false
}
mid := len(blocks) / 2
if mid <= 0 || mid >= len(blocks) {
return docChunkSplitPlan{}, false
}
return docChunkSplitPlan{
groups: [][]string{blocks[:mid], blocks[mid:]},
reason: fmt.Sprintf("prompt-budget:%d>%d", estimatedPromptCost, promptBudget),
}, true
}
func splitDocChunkBlocksMidpointSimple(blocks []string) (docChunkSplitPlan, bool) {
if len(blocks) <= 1 {
return docChunkSplitPlan{}, false
}
mid := len(blocks) / 2
if mid <= 0 || mid >= len(blocks) {
return docChunkSplitPlan{}, false
}
return docChunkSplitPlan{
groups: [][]string{blocks[:mid], blocks[mid:]},
reason: "retry-midpoint",
}, true
}
func planSingletonDocChunk(block string, maxBytes, promptBudget int) (docChunkSplitPlan, bool) {
normalizedBlock, _ := stripCommonIndent(block)
estimatedPromptCost := estimateDocPromptCost(normalizedBlock)
overBytes := maxBytes > 0 && len(block) > maxBytes
overPrompt := promptBudget > 0 && estimatedPromptCost > promptBudget
if !overBytes && !overPrompt {
return docChunkSplitPlan{}, false
}
return planSingletonDocChunkWithMode(block, maxBytes, promptBudget, false)
}
func planSingletonDocChunkRetry(block string, maxBytes, promptBudget int) (docChunkSplitPlan, bool) {
return planSingletonDocChunkWithMode(block, maxBytes, promptBudget, true)
}
func planSingletonDocChunkWithMode(block string, maxBytes, promptBudget int, force bool) (docChunkSplitPlan, bool) {
if sections := splitDocBlockSections(block); len(sections) > 1 {
if groups := wrapDocChunkSections(sections); len(groups) > 1 {
reason := "singleton-structural"
if force {
reason = "singleton-retry-structural"
}
return docChunkSplitPlan{
groups: groups,
reason: reason,
}, true
}
}
if groups, ok := splitPureFencedDocSectionWithMode(block, maxBytes, promptBudget, force); ok {
reason := "singleton-fence"
if force {
reason = "singleton-retry-fence"
}
return docChunkSplitPlan{
groups: groups,
reason: reason,
}, true
}
if groups, ok := splitPlainDocSectionWithMode(block, maxBytes, promptBudget, force); ok {
reason := "singleton-lines"
if force {
reason = "singleton-retry-lines"
}
return docChunkSplitPlan{
groups: groups,
reason: reason,
}, true
}
return docChunkSplitPlan{}, false
}
func wrapDocChunkSections(sections []string) [][]string {
groups := make([][]string, 0, len(sections))
for _, section := range sections {
if strings.TrimSpace(section) == "" {
continue
}
groups = append(groups, []string{section})
}
return groups
}
func splitDocBlockSections(block string) []string {
lines := strings.SplitAfter(block, "\n")
if len(lines) == 0 {
return nil
}
sections := make([]string, 0, len(lines))
var current strings.Builder
fenceDelimiter := ""
for _, line := range lines {
lineDelimiter := leadingFenceDelimiter(line)
if fenceDelimiter == "" && lineDelimiter != "" {
if current.Len() > 0 {
sections = append(sections, current.String())
current.Reset()
}
current.WriteString(line)
fenceDelimiter = lineDelimiter
continue
}
current.WriteString(line)
if fenceDelimiter != "" {
if lineDelimiter != "" && lineDelimiter[0] == fenceDelimiter[0] && len(lineDelimiter) >= len(fenceDelimiter) && isClosingFenceLine(line, fenceDelimiter) {
sections = append(sections, current.String())
current.Reset()
fenceDelimiter = ""
}
continue
}
if strings.TrimSpace(line) == "" {
sections = append(sections, current.String())
current.Reset()
}
}
if current.Len() > 0 {
sections = append(sections, current.String())
}
if len(sections) <= 1 {
return nil
}
return sections
}
func splitPureFencedDocSection(block string, maxBytes, promptBudget int) ([][]string, bool) {
return splitPureFencedDocSectionWithMode(block, maxBytes, promptBudget, false)
}
func splitPureFencedDocSectionWithMode(block string, maxBytes, promptBudget int, force bool) ([][]string, bool) {
lines := strings.SplitAfter(block, "\n")
if len(lines) < 2 {
return nil, false
}
openingIndex := firstNonEmptyLineIndex(lines)
closingIndex := lastNonEmptyLineIndex(lines)
if openingIndex == -1 || closingIndex <= openingIndex {
return nil, false
}
opening := lines[openingIndex]
delimiter := leadingFenceDelimiter(opening)
if delimiter == "" || !isClosingFenceLine(lines[closingIndex], delimiter) {
return nil, false
}
prefix := strings.Join(lines[:openingIndex], "")
suffix := strings.Join(lines[closingIndex+1:], "")
if strings.TrimSpace(prefix) != "" || strings.TrimSpace(suffix) != "" {
return nil, false
}
closing := lines[closingIndex]
inner := strings.Join(lines[openingIndex+1:closingIndex], "")
groups, ok := splitPlainDocSectionWithMode(inner, maxBytes-len(opening)-len(closing), promptBudget, force)
if !ok {
return nil, false
}
for index, group := range groups {
joined := strings.Join(group, "")
groups[index] = []string{opening + joined + closing}
}
return groups, true
}
func splitPlainDocSection(text string, maxBytes, promptBudget int) ([][]string, bool) {
return splitPlainDocSectionWithMode(text, maxBytes, promptBudget, false)
}
func splitPlainDocSectionWithMode(text string, maxBytes, promptBudget int, force bool) ([][]string, bool) {
if maxBytes <= 0 {
maxBytes = len(text)
}
if promptBudget <= 0 {
promptBudget = defaultDocChunkPromptBudget
}
lines := strings.SplitAfter(text, "\n")
if len(lines) <= 1 {
return nil, false
}
groups := make([][]string, 0, len(lines))
var current strings.Builder
currentBytes := 0
currentPrompt := 0
for _, line := range lines {
linePrompt := estimateDocPromptCost(line)
if len(line) > maxBytes || linePrompt > promptBudget {
return nil, false
}
if currentBytes > 0 && (currentBytes+len(line) > maxBytes || currentPrompt+linePrompt > promptBudget) {
groups = append(groups, []string{current.String()})
current.Reset()
currentBytes = 0
currentPrompt = 0
}
current.WriteString(line)
currentBytes += len(line)
currentPrompt += linePrompt
}
if current.Len() > 0 {
groups = append(groups, []string{current.String()})
}
if len(groups) <= 1 {
if !force {
return nil, false
}
return splitPlainDocSectionMidpoint(lines)
}
return groups, true
}
func splitPlainDocSectionMidpoint(lines []string) ([][]string, bool) {
if len(lines) <= 1 {
return nil, false
}
mid := len(lines) / 2
if mid <= 0 || mid >= len(lines) {
return nil, false
}
left := strings.Join(lines[:mid], "")
right := strings.Join(lines[mid:], "")
if strings.TrimSpace(left) == "" || strings.TrimSpace(right) == "" {
return nil, false
}
return [][]string{{left}, {right}}, true
}
func firstNonEmptyLineIndex(lines []string) int {
for index, line := range lines {
if strings.TrimSpace(line) != "" {
return index
}
}
return -1
}
func lastNonEmptyLineIndex(lines []string) int {
for index := len(lines) - 1; index >= 0; index-- {
if strings.TrimSpace(lines[index]) != "" {
return index
}
}
return -1
}
func docsI18nDocChunkMaxBytes() int {
value := strings.TrimSpace(os.Getenv("OPENCLAW_DOCS_I18N_DOC_CHUNK_MAX_BYTES"))
if value == "" {
return defaultDocChunkMaxBytes
}
parsed, err := strconv.Atoi(value)
if err != nil || parsed <= 0 {
return defaultDocChunkMaxBytes
}
return parsed
}
func docsI18nDocChunkPromptBudget() int {
value := strings.TrimSpace(os.Getenv("OPENCLAW_DOCS_I18N_DOC_CHUNK_PROMPT_BUDGET"))
if value == "" {
return defaultDocChunkPromptBudget
}
parsed, err := strconv.Atoi(value)
if err != nil || parsed <= 0 {
return defaultDocChunkPromptBudget
}
return parsed
}
func estimateDocPromptCost(text string) int {
cost := len(text)
cost += strings.Count(text, "`") * 6
cost += strings.Count(text, "|") * 4
cost += strings.Count(text, "{") * 4
cost += strings.Count(text, "}") * 4
cost += strings.Count(text, "[") * 4
cost += strings.Count(text, "]") * 4
cost += strings.Count(text, ":") * 2
cost += strings.Count(text, "<") * 4
cost += strings.Count(text, ">") * 4
return cost
}
func stripCommonIndent(text string) (string, string) {
lines := strings.SplitAfter(text, "\n")
common := ""
for _, line := range lines {
trimmed := strings.TrimRight(line, "\r\n")
if strings.TrimSpace(trimmed) == "" {
continue
}
indent := leadingIndent(trimmed)
if common == "" {
common = indent
continue
}
common = commonIndentPrefix(common, indent)
if common == "" {
return text, ""
}
}
if common == "" {
return text, ""
}
var out strings.Builder
for _, line := range lines {
trimmed := strings.TrimRight(line, "\r\n")
if strings.TrimSpace(trimmed) == "" {
out.WriteString(line)
continue
}
if strings.HasPrefix(line, common) {
out.WriteString(strings.TrimPrefix(line, common))
continue
}
out.WriteString(line)
}
return out.String(), common
}
func reapplyCommonIndent(text, indent string) string {
if indent == "" || text == "" {
return text
}
lines := strings.SplitAfter(text, "\n")
var out strings.Builder
for _, line := range lines {
trimmed := strings.TrimRight(line, "\r\n")
if strings.TrimSpace(trimmed) == "" {
out.WriteString(line)
continue
}
out.WriteString(indent)
out.WriteString(line)
}
return out.String()
}
func leadingIndent(line string) string {
index := 0
for index < len(line) {
if line[index] != ' ' && line[index] != '\t' {
break
}
index++
}
return line[:index]
}
func commonIndentPrefix(a, b string) string {
limit := len(a)
if len(b) < limit {
limit = len(b)
}
index := 0
for index < limit && a[index] == b[index] {
index++
}
return a[:index]
}