// Copyright 2016 The Hugo 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 hugolib import ( "bytes" "errors" "fmt" "reflect" "github.com/bep/gitmap" "github.com/mitchellh/mapstructure" "github.com/spf13/hugo/helpers" "github.com/spf13/hugo/parser" "html/template" "io" "net/url" "path" "path/filepath" "regexp" "strings" "sync" "time" "unicode/utf8" "github.com/spf13/cast" bp "github.com/spf13/hugo/bufferpool" "github.com/spf13/hugo/source" ) var ( cjk = regexp.MustCompile(`\p{Han}|\p{Hangul}|\p{Hiragana}|\p{Katakana}`) allKinds = []string{KindPage, KindHome, KindSection, KindTaxonomy, KindTaxonomyTerm, kindRSS, kindSitemap, kindRobotsTXT, kind404} ) const ( KindPage = "page" // The rest are node types; home page, sections etc. KindHome = "home" KindSection = "section" KindTaxonomy = "taxonomy" KindTaxonomyTerm = "taxonomyTerm" // Temporary state. kindUnknown = "unknown" // The following are (currently) temporary nodes, // i.e. nodes we create just to render in isolation. kindRSS = "RSS" kindSitemap = "sitemap" kindRobotsTXT = "robotsTXT" kind404 = "404" ) type Page struct { *pageInit // Kind is the discriminator that identifies the different page types // in the different page collections. This can, as an example, be used // to to filter regular pages, find sections etc. // Kind will, for the pages available to the templates, be one of: // page, home, section, taxonomy and taxonomyTerm. // It is of string type to make it easy to reason about in // the templates. Kind string // Since Hugo 0.18 we got rid of the Node type. So now all pages are ... // pages (regular pages, home page, sections etc.). // Sections etc. will have child pages. These were earlier placed in .Data.Pages, // but can now be more intuitively also be fetched directly from .Pages. // This collection will be nil for regular pages. Pages Pages // translations will contain references to this page in other language // if available. translations Pages // Params contains configuration defined in the params section of page frontmatter. Params map[string]interface{} // Content sections Content template.HTML Summary template.HTML TableOfContents template.HTML Aliases []string Images []Image Videos []Video Truncated bool Draft bool Status string PublishDate time.Time ExpiryDate time.Time // PageMeta contains page stats such as word count etc. PageMeta // Markup contains the markup type for the content. Markup string extension string contentType string renderable bool Layout string layoutsCalculated []string linkTitle string frontmatter []byte // rawContent is the raw content read from the content file. rawContent []byte // workContent is a copy of rawContent that may be mutated during site build. workContent []byte // state telling if this is a "new page" or if we have rendered it previously. rendered bool // whether the content is in a CJK language. isCJKLanguage bool // shortcode state contentShortCodes map[string]func() (string, error) shortcodes map[string]shortcode // the content stripped for HTML plain string // TODO should be []byte plainWords []string // rendering configuration renderingConfig *helpers.Blackfriday // menus pageMenus PageMenus Source Position `json:"-"` GitInfo *gitmap.GitInfo // This was added as part of getting the Nodes (taxonomies etc.) to work as // Pages in Hugo 0.18. // It is deliberately named similar to Section, but not exported (for now). // We currently have only one level of section in Hugo, but the page can live // any number of levels down the file path. // To support taxonomies like /categories/hugo etc. we will need to keep track // of that information in a general way. // So, sections represents the path to the content, i.e. a content file or a // virtual content file in the situations where a taxonomy or a section etc. // isn't accomanied by one. sections []string s *Site // Pulled over from old Node. TODO(bep) reorg and group (embed) Site *SiteInfo `json:"-"` Title string Description string Keywords []string Data map[string]interface{} Date time.Time Lastmod time.Time Sitemap Sitemap RSSLink template.HTML URLPath permalink *url.URL relPermalink string paginator *Pager scratch *Scratch // It would be tempting to use the language set on the Site, but in they way we do // multi-site processing, these values may differ during the initial page processing. language *helpers.Language lang string } // pageInit lazy initializes different parts of the page. It is extracted // into its own type so we can easily create a copy of a given page. type pageInit struct { languageInit sync.Once pageMenusInit sync.Once pageMetaInit sync.Once paginatorInit sync.Once plainInit sync.Once plainWordsInit sync.Once renderingConfigInit sync.Once pageURLInit sync.Once relPermalinkInit sync.Once } // IsNode returns whether this is an item of one of the list types in Hugo, // i.e. not a regular content page. func (p *Page) IsNode() bool { return p.Kind != KindPage } // IsHome returns whether this is the home page. func (p *Page) IsHome() bool { return p.Kind == KindHome } // IsPage returns whether this is a regular content page. func (p *Page) IsPage() bool { return p.Kind == KindPage } type Source struct { Frontmatter []byte Content []byte source.File } type PageMeta struct { wordCount int fuzzyWordCount int readingTime int Weight int } type Position struct { Prev *Page Next *Page PrevInSection *Page NextInSection *Page } type Pages []*Page func (p Pages) String() string { return fmt.Sprintf("Pages(%d)", len(p)) } func (ps Pages) FindPagePosByFilePath(inPath string) int { for i, x := range ps { if x.Source.Path() == inPath { return i } } return -1 } // FindPagePos Given a page, it will find the position in Pages // will return -1 if not found func (ps Pages) FindPagePos(page *Page) int { for i, x := range ps { if x.Source.Path() == page.Source.Path() { return i } } return -1 } func (p *Page) createWorkContentCopy() { p.workContent = make([]byte, len(p.rawContent)) copy(p.workContent, p.rawContent) } func (p *Page) Plain() string { p.initPlain() return p.plain } func (p *Page) PlainWords() []string { p.initPlainWords() return p.plainWords } func (p *Page) initPlain() { p.plainInit.Do(func() { p.plain = helpers.StripHTML(string(p.Content)) return }) } func (p *Page) initPlainWords() { p.plainWordsInit.Do(func() { p.plainWords = strings.Fields(p.Plain()) return }) } // Param is a convenience method to do lookups in Page's and Site's Params map, // in that order. // // This method is also implemented on Node and SiteInfo. func (p *Page) Param(key interface{}) (interface{}, error) { keyStr, err := cast.ToStringE(key) if err != nil { return nil, err } keyStr = strings.ToLower(keyStr) result, _ := p.traverseDirect(keyStr) if result != nil { return result, nil } keySegments := strings.Split(keyStr, ".") if len(keySegments) == 1 { return nil, nil } return p.traverseNested(keySegments) } func (p *Page) traverseDirect(key string) (interface{}, error) { keyStr := strings.ToLower(key) if val, ok := p.Params[keyStr]; ok { return val, nil } return p.Site.Params[keyStr], nil } func (p *Page) traverseNested(keySegments []string) (interface{}, error) { result := traverse(keySegments, p.Params) if result != nil { return result, nil } result = traverse(keySegments, p.Site.Params) if result != nil { return result, nil } // Didn't find anything, but also no problems. return nil, nil } func traverse(keys []string, m map[string]interface{}) interface{} { // Shift first element off. firstKey, rest := keys[0], keys[1:] result := m[firstKey] // No point in continuing here. if result == nil { return result } if len(rest) == 0 { // That was the last key. return result } else { // That was not the last key. return traverse(rest, cast.ToStringMap(result)) } } func (p *Page) Author() Author { authors := p.Authors() for _, author := range authors { return author } return Author{} } func (p *Page) Authors() AuthorList { authorKeys, ok := p.Params["authors"] if !ok { return AuthorList{} } authors := authorKeys.([]string) if len(authors) < 1 || len(p.Site.Authors) < 1 { return AuthorList{} } al := make(AuthorList) for _, author := range authors { a, ok := p.Site.Authors[author] if ok { al[author] = a } } return al } func (p *Page) UniqueID() string { return p.Source.UniqueID() } // for logging func (p *Page) lineNumRawContentStart() int { return bytes.Count(p.frontmatter, []byte("\n")) + 1 } var ( internalSummaryDivider = []byte("HUGOMORE42") ) // We have to replace the with something that survives all the // rendering engines. // TODO(bep) inline replace func (p *Page) replaceDivider(content []byte) []byte { sections := bytes.Split(content, helpers.SummaryDivider) // If the raw content has nothing but whitespace after the summary // marker then the page shouldn't be marked as truncated. This check // is simplest against the raw content because different markup engines // (rst and asciidoc in particular) add div and p elements after the // summary marker. p.Truncated = (len(sections) == 2 && len(bytes.Trim(sections[1], " \n\r")) > 0) return bytes.Join(sections, internalSummaryDivider) } // Returns the page as summary and main if a user defined split is provided. func (p *Page) setUserDefinedSummaryIfProvided(rawContentCopy []byte) (*summaryContent, error) { sc, err := splitUserDefinedSummaryAndContent(p.Markup, rawContentCopy) if err != nil { return nil, err } if sc == nil { // No divider found return nil, nil } p.Summary = helpers.BytesToHTML(sc.summary) return sc, nil } // Make this explicit so there is no doubt about what is what. type summaryContent struct { summary []byte content []byte } func splitUserDefinedSummaryAndContent(markup string, c []byte) (sc *summaryContent, err error) { defer func() { if r := recover(); r != nil { err = fmt.Errorf("summary split failed: %s", r) } }() c = bytes.TrimSpace(c) startDivider := bytes.Index(c, internalSummaryDivider) if startDivider == -1 { return } endDivider := startDivider + len(internalSummaryDivider) endSummary := startDivider var ( startMarkup []byte endMarkup []byte addDiv bool ) switch markup { default: startMarkup = []byte("

") endMarkup = []byte("

") case "asciidoc": startMarkup = []byte("
") endMarkup = []byte("
") case "rst": startMarkup = []byte("

") endMarkup = []byte("

") addDiv = true } // Find the closest end/start markup string to the divider fromStart := -1 fromIdx := bytes.LastIndex(c[:startDivider], startMarkup) if fromIdx != -1 { fromStart = startDivider - fromIdx - len(startMarkup) } fromEnd := bytes.Index(c[endDivider:], endMarkup) if fromEnd != -1 && fromEnd <= fromStart { endSummary = startDivider + fromEnd + len(endMarkup) } else if fromStart != -1 && fromEnd != -1 { endSummary = startDivider - fromStart - len(startMarkup) } withoutDivider := bytes.TrimSpace(append(c[:startDivider], c[endDivider:]...)) var ( summary []byte ) if len(withoutDivider) > 0 { summary = bytes.TrimSpace(withoutDivider[:endSummary]) } if addDiv { // For the rst summary = append(append([]byte(nil), summary...), []byte("")...) } if err != nil { return } sc = &summaryContent{ summary: summary, content: withoutDivider, } return } func (p *Page) setAutoSummary() error { var summary string var truncated bool if p.isCJKLanguage { summary, truncated = helpers.TruncateWordsByRune(p.PlainWords(), helpers.SummaryLength) } else { summary, truncated = helpers.TruncateWordsToWholeSentence(p.Plain(), helpers.SummaryLength) } p.Summary = template.HTML(summary) p.Truncated = truncated return nil } func (p *Page) renderContent(content []byte) []byte { var fn helpers.LinkResolverFunc var fileFn helpers.FileResolverFunc if p.getRenderingConfig().SourceRelativeLinksEval { fn = func(ref string) (string, error) { return p.Site.SourceRelativeLink(ref, p) } fileFn = func(ref string) (string, error) { return p.Site.SourceRelativeLinkFile(ref, p) } } return p.s.ContentSpec.RenderBytes(&helpers.RenderingContext{ Content: content, RenderTOC: true, PageFmt: p.determineMarkupType(), Cfg: p.Language(), DocumentID: p.UniqueID(), DocumentName: p.Path(), Config: p.getRenderingConfig(), LinkResolver: fn, FileResolver: fileFn}) } func (p *Page) getRenderingConfig() *helpers.Blackfriday { p.renderingConfigInit.Do(func() { pageParam := cast.ToStringMap(p.GetParam("blackfriday")) if p.Language() == nil { panic(fmt.Sprintf("nil language for %s with source lang %s", p.BaseFileName(), p.lang)) } p.renderingConfig = p.s.ContentSpec.NewBlackfriday() if err := mapstructure.Decode(pageParam, p.renderingConfig); err != nil { p.s.Log.FATAL.Printf("Failed to get rendering config for %s:\n%s", p.BaseFileName(), err.Error()) } }) return p.renderingConfig } func (s *Site) newPage(filename string) *Page { sp := source.NewSourceSpec(s.Cfg, s.Fs) page := Page{ pageInit: &pageInit{}, Kind: kindFromFilename(filename), contentType: "", Source: Source{File: *sp.NewFile(filename)}, Keywords: []string{}, Sitemap: Sitemap{Priority: -1}, Params: make(map[string]interface{}), translations: make(Pages, 0), sections: sectionsFromFilename(filename), Site: &s.Info, s: s, } s.Log.DEBUG.Println("Reading from", page.File.Path()) return &page } func (p *Page) IsRenderable() bool { return p.renderable } func (p *Page) Type() string { if p.contentType != "" { return p.contentType } if x := p.Section(); x != "" { return x } return "page" } func (p *Page) Section() string { return p.Source.Section() } func (p *Page) layouts(l ...string) []string { if len(p.layoutsCalculated) > 0 { return p.layoutsCalculated } switch p.Kind { case KindHome: return p.s.appendThemeTemplates([]string{"index.html", "_default/list.html"}) case KindSection: section := p.sections[0] return p.s.appendThemeTemplates([]string{"section/" + section + ".html", "_default/section.html", "_default/list.html", "indexes/" + section + ".html", "_default/indexes.html"}) case KindTaxonomy: singular := p.s.taxonomiesPluralSingular[p.sections[0]] return p.s.appendThemeTemplates([]string{"taxonomy/" + singular + ".html", "indexes/" + singular + ".html", "_default/taxonomy.html", "_default/list.html"}) case KindTaxonomyTerm: singular := p.s.taxonomiesPluralSingular[p.sections[0]] return p.s.appendThemeTemplates([]string{"taxonomy/" + singular + ".terms.html", "_default/terms.html", "indexes/indexes.html"}) } // Regular Page handled below if p.Layout != "" { return layouts(p.Type(), p.Layout) } layout := "" if len(l) == 0 { layout = "single" } else { layout = l[0] } return layouts(p.Type(), layout) } // TODO(bep) consolidate and test these KindHome switches (see other layouts methods)s // rssLayouts returns RSS layouts to use for the RSS version of this page, nil // if no RSS should be rendered. func (p *Page) rssLayouts() []string { switch p.Kind { case KindHome: return []string{"rss.xml", "_default/rss.xml", "_internal/_default/rss.xml"} case KindSection: section := p.sections[0] return []string{"section/" + section + ".rss.xml", "_default/rss.xml", "rss.xml", "_internal/_default/rss.xml"} case KindTaxonomy: singular := p.s.taxonomiesPluralSingular[p.sections[0]] return []string{"taxonomy/" + singular + ".rss.xml", "_default/rss.xml", "rss.xml", "_internal/_default/rss.xml"} case KindTaxonomyTerm: // No RSS for taxonomy terms case KindPage: // No RSS for regular pages } return nil } func layouts(types string, layout string) (layouts []string) { t := strings.Split(types, "/") // Add type/layout.html for i := range t { search := t[:len(t)-i] layouts = append(layouts, fmt.Sprintf("%s/%s.html", strings.ToLower(path.Join(search...)), layout)) } // Add _default/layout.html layouts = append(layouts, fmt.Sprintf("_default/%s.html", layout)) // Add theme/type/layout.html & theme/_default/layout.html for _, l := range layouts { layouts = append(layouts, "theme/"+l) } return } func (s *Site) NewPageFrom(buf io.Reader, name string) (*Page, error) { p, err := s.NewPage(name) if err != nil { return p, err } _, err = p.ReadFrom(buf) return p, err } func (s *Site) NewPage(name string) (*Page, error) { if len(name) == 0 { return nil, errors.New("Zero length page name") } // Create new page p := s.newPage(name) p.s = s p.Site = &s.Info return p, nil } func (p *Page) ReadFrom(buf io.Reader) (int64, error) { // Parse for metadata & body if err := p.parse(buf); err != nil { p.s.Log.ERROR.Print(err) return 0, err } return int64(len(p.rawContent)), nil } func (p *Page) WordCount() int { p.analyzePage() return p.wordCount } func (p *Page) ReadingTime() int { p.analyzePage() return p.readingTime } func (p *Page) FuzzyWordCount() int { p.analyzePage() return p.fuzzyWordCount } func (p *Page) analyzePage() { p.pageMetaInit.Do(func() { if p.isCJKLanguage { p.wordCount = 0 for _, word := range p.PlainWords() { runeCount := utf8.RuneCountInString(word) if len(word) == runeCount { p.wordCount++ } else { p.wordCount += runeCount } } } else { p.wordCount = helpers.TotalWords(p.Plain()) } // TODO(bep) is set in a test. Fix that. if p.fuzzyWordCount == 0 { p.fuzzyWordCount = (p.wordCount + 100) / 100 * 100 } if p.isCJKLanguage { p.readingTime = (p.wordCount + 500) / 501 } else { p.readingTime = (p.wordCount + 212) / 213 } }) } func (p *Page) getPermalink() *url.URL { p.pageURLInit.Do(func() { u, err := p.createPermalink() if err != nil { p.s.Log.ERROR.Printf("Failed to create permalink for page %q: %s", p.FullFilePath(), err) p.permalink = new(url.URL) return } p.permalink = u }) // The link may be modified by the receiver, so create a copy. l := *p.permalink return &l } func (p *Page) createPermalink() (*url.URL, error) { // TODO(bep) this should probably be set once during build. Maybe. // And simplified. baseURL := string(p.Site.BaseURL) if p.IsNode() { // No permalink config for nodes (currently) pURL := strings.TrimSpace(p.s.PathSpec.URLize(p.URLPath.URL)) pURL = p.addLangPathPrefix(pURL) pURL = p.s.PathSpec.URLPrep(pURL) url := helpers.MakePermalink(baseURL, pURL) return url, nil } dir := strings.TrimSpace(p.s.PathSpec.MakePath(filepath.ToSlash(strings.ToLower(p.Source.Dir())))) pSlug := strings.TrimSpace(p.s.PathSpec.URLize(p.Slug)) pURL := strings.TrimSpace(p.s.PathSpec.URLize(p.URLPath.URL)) var permalink string var err error if len(pURL) > 0 { return helpers.MakePermalink(baseURL, pURL), nil } if override, ok := p.Site.Permalinks[p.Section()]; ok { permalink, err = override.Expand(p) if err != nil { return nil, err } } else { if len(pSlug) > 0 { permalink = p.s.PathSpec.URLPrep(path.Join(dir, p.Slug+"."+p.Extension())) } else { t := p.Source.TranslationBaseName() permalink = p.s.PathSpec.URLPrep(path.Join(dir, (strings.TrimSpace(t) + "." + p.Extension()))) } } permalink = p.addLangPathPrefix(permalink) return helpers.MakePermalink(baseURL, permalink), nil } func (p *Page) Extension() string { if p.extension != "" { return p.extension } return p.s.Cfg.GetString("defaultExtension") } // AllTranslations returns all translations, including the current Page. func (p *Page) AllTranslations() Pages { return p.translations } // IsTranslated returns whether this content file is translated to // other language(s). func (p *Page) IsTranslated() bool { return len(p.translations) > 1 } // Translations returns the translations excluding the current Page. func (p *Page) Translations() Pages { translations := make(Pages, 0) for _, t := range p.translations { if t.Lang() != p.Lang() { translations = append(translations, t) } } return translations } func (p *Page) LinkTitle() string { if len(p.linkTitle) > 0 { return p.linkTitle } return p.Title } func (p *Page) shouldBuild() bool { return shouldBuild(p.s.Cfg.GetBool("buildFuture"), p.s.Cfg.GetBool("buildExpired"), p.s.Cfg.GetBool("buildDrafts"), p.Draft, p.PublishDate, p.ExpiryDate) } func shouldBuild(buildFuture bool, buildExpired bool, buildDrafts bool, Draft bool, publishDate time.Time, expiryDate time.Time) bool { if !(buildDrafts || !Draft) { return false } if !buildFuture && !publishDate.IsZero() && publishDate.After(time.Now()) { return false } if !buildExpired && !expiryDate.IsZero() && expiryDate.Before(time.Now()) { return false } return true } func (p *Page) IsDraft() bool { return p.Draft } func (p *Page) IsFuture() bool { if p.PublishDate.IsZero() { return false } return p.PublishDate.After(time.Now()) } func (p *Page) IsExpired() bool { if p.ExpiryDate.IsZero() { return false } return p.ExpiryDate.Before(time.Now()) } func (p *Page) Permalink() string { return p.getPermalink().String() } func (p *Page) URL() string { if p.IsPage() && p.URLPath.URL != "" { // This is the url set in front matter return p.URLPath.URL } // Fall back to the relative permalink. u := p.RelPermalink() return u } func (p *Page) RelPermalink() string { p.relPermalinkInit.Do(func() { link := p.getPermalink() if p.s.Info.canonifyURLs { // replacements for relpermalink with baseURL on the form http://myhost.com/sub/ will fail later on // have to return the URL relative from baseURL relpath, err := helpers.GetRelativePath(link.String(), string(p.Site.BaseURL)) if err != nil { return } relpath = filepath.ToSlash(relpath) if relpath[0] == '.' { relpath = relpath[1:] } if !strings.HasPrefix(relpath, "/") { relpath = "/" + relpath } p.relPermalink = relpath return } link.Scheme = "" link.Host = "" link.User = nil link.Opaque = "" p.relPermalink = link.String() }) return p.relPermalink } var ErrHasDraftAndPublished = errors.New("both draft and published parameters were found in page's frontmatter") func (p *Page) update(f interface{}) error { if f == nil { return errors.New("no metadata found") } m := f.(map[string]interface{}) // Needed for case insensitive fetching of params values helpers.ToLowerMap(m) var err error var draft, published, isCJKLanguage *bool for k, v := range m { loki := strings.ToLower(k) switch loki { case "title": p.Title = cast.ToString(v) case "linktitle": p.linkTitle = cast.ToString(v) case "description": p.Description = cast.ToString(v) p.Params["description"] = p.Description case "slug": p.Slug = cast.ToString(v) case "url": if url := cast.ToString(v); strings.HasPrefix(url, "http://") || strings.HasPrefix(url, "https://") { return fmt.Errorf("Only relative URLs are supported, %v provided", url) } p.URLPath.URL = cast.ToString(v) case "type": p.contentType = cast.ToString(v) case "extension", "ext": p.extension = cast.ToString(v) case "keywords": p.Keywords = cast.ToStringSlice(v) case "date": p.Date, err = cast.ToTimeE(v) if err != nil { p.s.Log.ERROR.Printf("Failed to parse date '%v' in page %s", v, p.File.Path()) } case "lastmod": p.Lastmod, err = cast.ToTimeE(v) if err != nil { p.s.Log.ERROR.Printf("Failed to parse lastmod '%v' in page %s", v, p.File.Path()) } case "publishdate", "pubdate": p.PublishDate, err = cast.ToTimeE(v) if err != nil { p.s.Log.ERROR.Printf("Failed to parse publishdate '%v' in page %s", v, p.File.Path()) } case "expirydate", "unpublishdate": p.ExpiryDate, err = cast.ToTimeE(v) if err != nil { p.s.Log.ERROR.Printf("Failed to parse expirydate '%v' in page %s", v, p.File.Path()) } case "draft": draft = new(bool) *draft = cast.ToBool(v) case "published": // Intentionally undocumented published = new(bool) *published = cast.ToBool(v) case "layout": p.Layout = cast.ToString(v) case "markup": p.Markup = cast.ToString(v) case "weight": p.Weight = cast.ToInt(v) case "aliases": p.Aliases = cast.ToStringSlice(v) for _, alias := range p.Aliases { if strings.HasPrefix(alias, "http://") || strings.HasPrefix(alias, "https://") { return fmt.Errorf("Only relative aliases are supported, %v provided", alias) } } case "status": p.Status = cast.ToString(v) case "sitemap": p.Sitemap = parseSitemap(cast.ToStringMap(v)) case "iscjklanguage": isCJKLanguage = new(bool) *isCJKLanguage = cast.ToBool(v) default: // If not one of the explicit values, store in Params switch vv := v.(type) { case bool: p.Params[loki] = vv case string: p.Params[loki] = vv case int64, int32, int16, int8, int: p.Params[loki] = vv case float64, float32: p.Params[loki] = vv case time.Time: p.Params[loki] = vv default: // handle array of strings as well switch vvv := vv.(type) { case []interface{}: if len(vvv) > 0 { switch vvv[0].(type) { case map[interface{}]interface{}: // Proper parsing structured array from YAML based FrontMatter p.Params[loki] = vvv case map[string]interface{}: // Proper parsing structured array from JSON based FrontMatter p.Params[loki] = vvv case []interface{}: p.Params[loki] = vvv default: a := make([]string, len(vvv)) for i, u := range vvv { a[i] = cast.ToString(u) } p.Params[loki] = a } } else { p.Params[loki] = []string{} } default: p.Params[loki] = vv } } } } if draft != nil && published != nil { p.Draft = *draft p.s.Log.ERROR.Printf("page %s has both draft and published settings in its frontmatter. Using draft.", p.File.Path()) return ErrHasDraftAndPublished } else if draft != nil { p.Draft = *draft } else if published != nil { p.Draft = !*published } if p.Date.IsZero() && p.s.Cfg.GetBool("useModTimeAsFallback") { fi, err := p.s.Fs.Source.Stat(filepath.Join(p.s.PathSpec.AbsPathify(p.s.Cfg.GetString("contentDir")), p.File.Path())) if err == nil { p.Date = fi.ModTime() } } if p.Lastmod.IsZero() { p.Lastmod = p.Date } if isCJKLanguage != nil { p.isCJKLanguage = *isCJKLanguage } else if p.s.Cfg.GetBool("hasCJKLanguage") { if cjk.Match(p.rawContent) { p.isCJKLanguage = true } else { p.isCJKLanguage = false } } return nil } func (p *Page) GetParam(key string) interface{} { return p.getParam(key, true) } func (p *Page) getParam(key string, stringToLower bool) interface{} { v := p.Params[strings.ToLower(key)] if v == nil { return nil } switch val := v.(type) { case bool: return val case string: if stringToLower { return strings.ToLower(val) } return val case int64, int32, int16, int8, int: return cast.ToInt(v) case float64, float32: return cast.ToFloat64(v) case time.Time: return val case []string: if stringToLower { return helpers.SliceToLower(val) } return v case map[string]interface{}: // JSON and TOML return v case map[interface{}]interface{}: // YAML return v } p.s.Log.ERROR.Printf("GetParam(\"%s\"): Unknown type %s\n", key, reflect.TypeOf(v)) return nil } func (p *Page) HasMenuCurrent(menuID string, me *MenuEntry) bool { sectionPagesMenu := p.Site.sectionPagesMenu // page is labeled as "shadow-member" of the menu with the same identifier as the section if sectionPagesMenu != "" && p.Section() != "" && sectionPagesMenu == menuID && p.Section() == me.Identifier { return true } if !me.HasChildren() { return false } menus := p.Menus() if m, ok := menus[menuID]; ok { for _, child := range me.Children { if child.IsEqual(m) { return true } if p.HasMenuCurrent(menuID, child) { return true } } } if p.IsPage() { return false } // The following logic is kept from back when Hugo had both Page and Node types. // TODO(bep) consolidate / clean nme := MenuEntry{Name: p.Title, URL: p.URL()} for _, child := range me.Children { if nme.IsSameResource(child) { return true } if p.HasMenuCurrent(menuID, child) { return true } } return false } func (p *Page) IsMenuCurrent(menuID string, inme *MenuEntry) bool { menus := p.Menus() if me, ok := menus[menuID]; ok { if me.IsEqual(inme) { return true } } if p.IsPage() { return false } // The following logic is kept from back when Hugo had both Page and Node types. // TODO(bep) consolidate / clean me := MenuEntry{Name: p.Title, URL: p.URL()} if !me.IsSameResource(inme) { return false } // this resource may be included in several menus // search for it to make sure that it is in the menu with the given menuId if menu, ok := (*p.Site.Menus)[menuID]; ok { for _, menuEntry := range *menu { if menuEntry.IsSameResource(inme) { return true } descendantFound := p.isSameAsDescendantMenu(inme, menuEntry) if descendantFound { return descendantFound } } } return false } func (p *Page) isSameAsDescendantMenu(inme *MenuEntry, parent *MenuEntry) bool { if parent.HasChildren() { for _, child := range parent.Children { if child.IsSameResource(inme) { return true } descendantFound := p.isSameAsDescendantMenu(inme, child) if descendantFound { return descendantFound } } } return false } func (p *Page) Menus() PageMenus { p.pageMenusInit.Do(func() { p.pageMenus = PageMenus{} if ms, ok := p.Params["menu"]; ok { link := p.RelPermalink() me := MenuEntry{Name: p.LinkTitle(), Weight: p.Weight, URL: link} // Could be the name of the menu to attach it to mname, err := cast.ToStringE(ms) if err == nil { me.Menu = mname p.pageMenus[mname] = &me return } // Could be a slice of strings mnames, err := cast.ToStringSliceE(ms) if err == nil { for _, mname := range mnames { me.Menu = mname p.pageMenus[mname] = &me } return } // Could be a structured menu entry menus, err := cast.ToStringMapE(ms) if err != nil { p.s.Log.ERROR.Printf("unable to process menus for %q\n", p.Title) } for name, menu := range menus { menuEntry := MenuEntry{Name: p.LinkTitle(), URL: link, Weight: p.Weight, Menu: name} if menu != nil { p.s.Log.DEBUG.Printf("found menu: %q, in %q\n", name, p.Title) ime, err := cast.ToStringMapE(menu) if err != nil { p.s.Log.ERROR.Printf("unable to process menus for %q: %s", p.Title, err) } menuEntry.marshallMap(ime) } p.pageMenus[name] = &menuEntry } } }) return p.pageMenus } func (p *Page) Render(layout ...string) template.HTML { var l []string if len(layout) > 0 { l = layouts(p.Type(), layout[0]) } else { l = p.layouts() } return p.s.Tmpl.ExecuteTemplateToHTML(p, l...) } func (p *Page) determineMarkupType() string { // Try markup explicitly set in the frontmatter p.Markup = helpers.GuessType(p.Markup) if p.Markup == "unknown" { // Fall back to file extension (might also return "unknown") p.Markup = helpers.GuessType(p.Source.Ext()) } return p.Markup } func (p *Page) parse(reader io.Reader) error { psr, err := parser.ReadFrom(reader) if err != nil { return err } p.renderable = psr.IsRenderable() p.frontmatter = psr.FrontMatter() p.rawContent = psr.Content() p.lang = p.Source.File.Lang() meta, err := psr.Metadata() if meta != nil { if err != nil { p.s.Log.ERROR.Printf("Error parsing page meta data for %s", p.File.Path()) p.s.Log.ERROR.Println(err) return err } if err = p.update(meta); err != nil { return err } } return nil } func (p *Page) RawContent() string { return string(p.rawContent) } func (p *Page) SetSourceContent(content []byte) { p.Source.Content = content } func (p *Page) SetSourceMetaData(in interface{}, mark rune) (err error) { // See https://github.com/spf13/hugo/issues/2458 defer func() { if r := recover(); r != nil { var ok bool err, ok = r.(error) if !ok { err = fmt.Errorf("error from marshal: %v", r) } } }() var by []byte by, err = parser.InterfaceToFrontMatter(in, mark) if err != nil { return } by = append(by, '\n') p.Source.Frontmatter = by return } func (p *Page) SafeSaveSourceAs(path string) error { return p.saveSourceAs(path, true) } func (p *Page) SaveSourceAs(path string) error { return p.saveSourceAs(path, false) } func (p *Page) saveSourceAs(path string, safe bool) error { b := bp.GetBuffer() defer bp.PutBuffer(b) b.Write(p.Source.Frontmatter) b.Write(p.Source.Content) bc := make([]byte, b.Len(), b.Len()) copy(bc, b.Bytes()) return p.saveSource(bc, path, safe) } func (p *Page) saveSource(by []byte, inpath string, safe bool) (err error) { if !filepath.IsAbs(inpath) { inpath = p.s.PathSpec.AbsPathify(inpath) } p.s.Log.INFO.Println("creating", inpath) if safe { err = helpers.SafeWriteToDisk(inpath, bytes.NewReader(by), p.s.Fs.Source) } else { err = helpers.WriteToDisk(inpath, bytes.NewReader(by), p.s.Fs.Source) } if err != nil { return } return nil } func (p *Page) SaveSource() error { return p.SaveSourceAs(p.FullFilePath()) } func (p *Page) ProcessShortcodes() { tmpContent, tmpContentShortCodes, _ := extractAndRenderShortcodes(string(p.workContent), p) p.workContent = []byte(tmpContent) p.contentShortCodes = tmpContentShortCodes } func (p *Page) FullFilePath() string { return filepath.Join(p.Dir(), p.LogicalName()) } func (p *Page) TargetPath() (outfile string) { switch p.Kind { case KindHome: return p.addLangFilepathPrefix(helpers.FilePathSeparator) case KindSection: return p.addLangFilepathPrefix(p.sections[0]) case KindTaxonomy: return p.addLangFilepathPrefix(filepath.Join(p.sections...)) case KindTaxonomyTerm: return p.addLangFilepathPrefix(filepath.Join(p.sections...)) } // Always use URL if it's specified if len(strings.TrimSpace(p.URLPath.URL)) > 2 { outfile = strings.TrimSpace(p.URLPath.URL) if strings.HasSuffix(outfile, "/") { outfile = outfile + "index.html" } outfile = filepath.FromSlash(outfile) return } // If there's a Permalink specification, we use that if override, ok := p.Site.Permalinks[p.Section()]; ok { var err error outfile, err = override.Expand(p) if err == nil { outfile, _ = url.QueryUnescape(outfile) if strings.HasSuffix(outfile, "/") { outfile += "index.html" } outfile = filepath.FromSlash(outfile) outfile = p.addLangFilepathPrefix(outfile) return } } if len(strings.TrimSpace(p.Slug)) > 0 { outfile = strings.TrimSpace(p.Slug) + "." + p.Extension() } else { // Fall back to filename outfile = (p.Source.TranslationBaseName() + "." + p.Extension()) } return p.addLangFilepathPrefix(filepath.Join(strings.ToLower( p.s.PathSpec.MakePath(p.Source.Dir())), strings.TrimSpace(outfile))) } // Pre render prepare steps func (p *Page) prepareLayouts() error { // TODO(bep): Check the IsRenderable logic. if p.Kind == KindPage { var layouts []string if !p.IsRenderable() { self := "__" + p.TargetPath() _, err := p.s.Tmpl.GetClone().New(self).Parse(string(p.Content)) if err != nil { return err } layouts = append(layouts, self) } else { layouts = append(layouts, p.layouts()...) } p.layoutsCalculated = layouts } return nil } func (p *Page) prepareData(s *Site) error { var pages Pages p.Data = make(map[string]interface{}) switch p.Kind { case KindPage: case KindHome: pages = s.RegularPages case KindSection: sectionData, ok := s.Sections[p.sections[0]] if !ok { return fmt.Errorf("Data for section %s not found", p.Section()) } pages = sectionData.Pages() case KindTaxonomy: plural := p.sections[0] term := p.sections[1] if s.Info.preserveTaxonomyNames { if v, ok := s.taxonomiesOrigKey[fmt.Sprintf("%s-%s", plural, term)]; ok { term = v } } singular := s.taxonomiesPluralSingular[plural] taxonomy := s.Taxonomies[plural].Get(term) p.Data[singular] = taxonomy p.Data["Singular"] = singular p.Data["Plural"] = plural pages = taxonomy.Pages() case KindTaxonomyTerm: plural := p.sections[0] singular := s.taxonomiesPluralSingular[plural] p.Data["Singular"] = singular p.Data["Plural"] = plural p.Data["Terms"] = s.Taxonomies[plural] // keep the following just for legacy reasons p.Data["OrderedIndex"] = p.Data["Terms"] p.Data["Index"] = p.Data["Terms"] } p.Data["Pages"] = pages p.Pages = pages // Now we know enough to set missing dates on home page etc. p.updatePageDates() return nil } func (p *Page) updatePageDates() { // TODO(bep) there is a potential issue with page sorting for home pages // etc. without front matter dates set, but let us wrap the head around // that in another time. if !p.IsNode() { return } if !p.Date.IsZero() { if p.Lastmod.IsZero() { p.Lastmod = p.Date } return } else if !p.Lastmod.IsZero() { if p.Date.IsZero() { p.Date = p.Lastmod } return } // Set it to the first non Zero date in children var foundDate, foundLastMod bool for _, child := range p.Pages { if !child.Date.IsZero() { p.Date = child.Date foundDate = true } if !child.Lastmod.IsZero() { p.Lastmod = child.Lastmod foundLastMod = true } if foundDate && foundLastMod { break } } } // copy creates a copy of this page with the lazy sync.Once vars reset // so they will be evaluated again, for word count calculations etc. func (p *Page) copy() *Page { c := *p c.pageInit = &pageInit{} return &c } func (p *Page) Now() time.Time { // Delete in Hugo 0.21 helpers.Deprecated("Page", "Now", "Use now (the template func)", false) return time.Now() } func (p *Page) Hugo() *HugoInfo { return hugoInfo } func (p *Page) RSSlink() template.HTML { // TODO(bep) we cannot have two of these // Remove in Hugo 0.20 helpers.Deprecated(".Page", "Use RSSlink", "RSSLink", true) return p.RSSLink } func (p *Page) Ref(ref string) (string, error) { return p.Site.Ref(ref, nil) } func (p *Page) RelRef(ref string) (string, error) { return p.Site.RelRef(ref, nil) } func (p *Page) String() string { return fmt.Sprintf("Page(%q)", p.Title) } type URLPath struct { URL string Permalink string Slug string Section string } // Scratch returns the writable context associated with this Page. func (p *Page) Scratch() *Scratch { if p.scratch == nil { p.scratch = newScratch() } return p.scratch } func (p *Page) Language() *helpers.Language { p.initLanguage() return p.language } func (p *Page) Lang() string { // When set, Language can be different from lang in the case where there is a // content file (doc.sv.md) with language indicator, but there is no language // config for that language. Then the language will fall back on the site default. if p.Language() != nil { return p.Language().Lang } return p.lang } func (p *Page) isNewTranslation(candidate *Page) bool { if p.Kind != candidate.Kind { return false } if p.Kind == KindPage || p.Kind == kindUnknown { panic("Node type not currently supported for this op") } // At this point, we know that this is a traditional Node (home page, section, taxonomy) // It represents the same node, but different language, if the sections is the same. if len(p.sections) != len(candidate.sections) { return false } for i := 0; i < len(p.sections); i++ { if p.sections[i] != candidate.sections[i] { return false } } // Finally check that it is not already added. for _, translation := range p.translations { if candidate == translation { return false } } return true } func (p *Page) shouldAddLanguagePrefix() bool { if !p.Site.IsMultiLingual() { return false } if p.Lang() == "" { return false } if !p.Site.defaultContentLanguageInSubdir && p.Lang() == p.Site.multilingual.DefaultLang.Lang { return false } return true } func (p *Page) initLanguage() { p.languageInit.Do(func() { if p.language != nil { return } ml := p.Site.multilingual if ml == nil { panic("Multilanguage not set") } if p.lang == "" { p.lang = ml.DefaultLang.Lang p.language = ml.DefaultLang return } language := ml.Language(p.lang) if language == nil { // It can be a file named stefano.chiodino.md. p.s.Log.WARN.Printf("Page language (if it is that) not found in multilang setup: %s.", p.lang) language = ml.DefaultLang } p.language = language }) } func (p *Page) LanguagePrefix() string { return p.Site.LanguagePrefix } func (p *Page) addLangPathPrefix(outfile string) string { return p.addLangPathPrefixIfFlagSet(outfile, p.shouldAddLanguagePrefix()) } func (p *Page) addLangPathPrefixIfFlagSet(outfile string, should bool) string { if helpers.IsAbsURL(outfile) { return outfile } if !should { return outfile } hadSlashSuffix := strings.HasSuffix(outfile, "/") outfile = "/" + path.Join(p.Lang(), outfile) if hadSlashSuffix { outfile += "/" } return outfile } func (p *Page) addLangFilepathPrefix(outfile string) string { if outfile == "" { outfile = helpers.FilePathSeparator } if !p.shouldAddLanguagePrefix() { return outfile } return helpers.FilePathSeparator + filepath.Join(p.Lang(), outfile) } func sectionsFromFilename(filename string) []string { var sections []string dir, _ := filepath.Split(filename) dir = strings.TrimSuffix(dir, helpers.FilePathSeparator) if dir == "" { return sections } sections = strings.Split(dir, helpers.FilePathSeparator) return sections } func kindFromFilename(filename string) string { if !strings.Contains(filename, "_index") { return KindPage } if strings.HasPrefix(filename, "_index") { return KindHome } // We don't know enough yet to determine the type. return kindUnknown } func (p *Page) setValuesForKind(s *Site) { if p.Kind == kindUnknown { // This is either a taxonomy list, taxonomy term or a section nodeType := s.kindFromSections(p.sections) if nodeType == kindUnknown { panic(fmt.Sprintf("Unable to determine page kind from %q", p.sections)) } p.Kind = nodeType } switch p.Kind { case KindHome: p.URLPath.URL = "/" case KindSection: p.URLPath.URL = "/" + p.sections[0] + "/" case KindTaxonomy: p.URLPath.URL = "/" + path.Join(p.sections...) + "/" case KindTaxonomyTerm: p.URLPath.URL = "/" + path.Join(p.sections...) + "/" } }