hugo/hugolib/pagecollections.go
Bjørn Erik Pedersen 9f5a92078a
Add Hugo Modules
This commit implements Hugo Modules.

This is a broad subject, but some keywords include:

* A new `module` configuration section where you can import almost anything. You can configure both your own file mounts nd the file mounts of the modules you import. This is the new recommended way of configuring what you earlier put in `configDir`, `staticDir` etc. And it also allows you to mount folders in non-Hugo-projects, e.g. the `SCSS` folder in the Bootstrap GitHub project.
* A module consists of a set of mounts to the standard 7 component types in Hugo: `static`, `content`, `layouts`, `data`, `assets`, `i18n`, and `archetypes`. Yes, Theme Components can now include content, which should be very useful, especially in bigger multilingual projects.
* Modules not in your local file cache will be downloaded automatically and even "hot replaced" while the server is running.
* Hugo Modules supports and encourages semver versioned modules, and uses the minimal version selection algorithm to resolve versions.
* A new set of CLI commands are provided to manage all of this: `hugo mod init`,  `hugo mod get`,  `hugo mod graph`,  `hugo mod tidy`, and  `hugo mod vendor`.

All of the above is backed by Go Modules.

Fixes #5973
Fixes #5996
Fixes #6010
Fixes #5911
Fixes #5940
Fixes #6074
Fixes #6082
Fixes #6092
2019-07-24 09:35:53 +02:00

391 lines
9.9 KiB
Go

// Copyright 2019 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 (
"fmt"
"path"
"path/filepath"
"strings"
"sync"
"github.com/pkg/errors"
"github.com/gohugoio/hugo/cache"
"github.com/gohugoio/hugo/helpers"
"github.com/gohugoio/hugo/resources/page"
)
// Used in the page cache to mark more than one hit for a given key.
var ambiguityFlag = &pageState{}
// PageCollections contains the page collections for a site.
type PageCollections struct {
// Includes absolute all pages (of all types), including drafts etc.
rawAllPages pageStatePages
// rawAllPages plus additional pages created during the build process.
workAllPages pageStatePages
// Includes headless bundles, i.e. bundles that produce no output for its content page.
headlessPages pageStatePages
// Lazy initialized page collections
pages *lazyPagesFactory
regularPages *lazyPagesFactory
allPages *lazyPagesFactory
allRegularPages *lazyPagesFactory
// The index for .Site.GetPage etc.
pageIndex *cache.Lazy
}
// Pages returns all pages.
// This is for the current language only.
func (c *PageCollections) Pages() page.Pages {
return c.pages.get()
}
// RegularPages returns all the regular pages.
// This is for the current language only.
func (c *PageCollections) RegularPages() page.Pages {
return c.regularPages.get()
}
// AllPages returns all pages for all languages.
func (c *PageCollections) AllPages() page.Pages {
return c.allPages.get()
}
// AllPages returns all regular pages for all languages.
func (c *PageCollections) AllRegularPages() page.Pages {
return c.allRegularPages.get()
}
// Get initializes the index if not already done so, then
// looks up the given page ref, returns nil if no value found.
func (c *PageCollections) getFromCache(ref string) (page.Page, error) {
v, found, err := c.pageIndex.Get(ref)
if err != nil {
return nil, err
}
if !found {
return nil, nil
}
p := v.(page.Page)
if p != ambiguityFlag {
return p, nil
}
return nil, fmt.Errorf("page reference %q is ambiguous", ref)
}
type lazyPagesFactory struct {
pages page.Pages
init sync.Once
factory page.PagesFactory
}
func (l *lazyPagesFactory) get() page.Pages {
l.init.Do(func() {
l.pages = l.factory()
})
return l.pages
}
func newLazyPagesFactory(factory page.PagesFactory) *lazyPagesFactory {
return &lazyPagesFactory{factory: factory}
}
func newPageCollections() *PageCollections {
return newPageCollectionsFromPages(nil)
}
func newPageCollectionsFromPages(pages pageStatePages) *PageCollections {
c := &PageCollections{rawAllPages: pages}
c.pages = newLazyPagesFactory(func() page.Pages {
pages := make(page.Pages, len(c.workAllPages))
for i, p := range c.workAllPages {
pages[i] = p
}
return pages
})
c.regularPages = newLazyPagesFactory(func() page.Pages {
return c.findPagesByKindInWorkPages(page.KindPage, c.workAllPages)
})
c.pageIndex = cache.NewLazy(func() (map[string]interface{}, error) {
index := make(map[string]interface{})
add := func(ref string, p page.Page) {
ref = strings.ToLower(ref)
existing := index[ref]
if existing == nil {
index[ref] = p
} else if existing != ambiguityFlag && existing != p {
index[ref] = ambiguityFlag
}
}
for _, pageCollection := range []pageStatePages{c.workAllPages, c.headlessPages} {
for _, p := range pageCollection {
if p.IsPage() {
sourceRef := p.sourceRef()
if sourceRef != "" {
// index the canonical ref
// e.g. /section/article.md
add(sourceRef, p)
}
// Ref/Relref supports this potentially ambiguous lookup.
add(p.File().LogicalName(), p)
translationBaseName := p.File().TranslationBaseName()
dir, _ := path.Split(sourceRef)
dir = strings.TrimSuffix(dir, "/")
if translationBaseName == "index" {
add(dir, p)
add(path.Base(dir), p)
} else {
add(translationBaseName, p)
}
// We need a way to get to the current language version.
pathWithNoExtensions := path.Join(dir, translationBaseName)
add(pathWithNoExtensions, p)
} else {
// index the canonical, unambiguous ref for any backing file
// e.g. /section/_index.md
sourceRef := p.sourceRef()
if sourceRef != "" {
add(sourceRef, p)
}
ref := p.SectionsPath()
// index the canonical, unambiguous virtual ref
// e.g. /section
// (this may already have been indexed above)
add("/"+ref, p)
}
}
}
return index, nil
})
return c
}
// This is an adapter func for the old API with Kind as first argument.
// This is invoked when you do .Site.GetPage. We drop the Kind and fails
// if there are more than 2 arguments, which would be ambigous.
func (c *PageCollections) getPageOldVersion(ref ...string) (page.Page, error) {
var refs []string
for _, r := range ref {
// A common construct in the wild is
// .Site.GetPage "home" "" or
// .Site.GetPage "home" "/"
if r != "" && r != "/" {
refs = append(refs, r)
}
}
var key string
if len(refs) > 2 {
// This was allowed in Hugo <= 0.44, but we cannot support this with the
// new API. This should be the most unusual case.
return nil, fmt.Errorf(`too many arguments to .Site.GetPage: %v. Use lookups on the form {{ .Site.GetPage "/posts/mypage-md" }}`, ref)
}
if len(refs) == 0 || refs[0] == page.KindHome {
key = "/"
} else if len(refs) == 1 {
if len(ref) == 2 && refs[0] == page.KindSection {
// This is an old style reference to the "Home Page section".
// Typically fetched via {{ .Site.GetPage "section" .Section }}
// See https://github.com/gohugoio/hugo/issues/4989
key = "/"
} else {
key = refs[0]
}
} else {
key = refs[1]
}
key = filepath.ToSlash(key)
if !strings.HasPrefix(key, "/") {
key = "/" + key
}
return c.getPageNew(nil, key)
}
// Only used in tests.
func (c *PageCollections) getPage(typ string, sections ...string) page.Page {
refs := append([]string{typ}, path.Join(sections...))
p, _ := c.getPageOldVersion(refs...)
return p
}
// Case insensitive page lookup.
func (c *PageCollections) getPageNew(context page.Page, ref string) (page.Page, error) {
var anError error
ref = strings.ToLower(ref)
// Absolute (content root relative) reference.
if strings.HasPrefix(ref, "/") {
p, err := c.getFromCache(ref)
if err == nil && p != nil {
return p, nil
}
if err != nil {
anError = err
}
} else if context != nil {
// Try the page-relative path.
ppath := path.Join("/", strings.ToLower(context.SectionsPath()), ref)
p, err := c.getFromCache(ppath)
if err == nil && p != nil {
return p, nil
}
if err != nil {
anError = err
}
}
if !strings.HasPrefix(ref, "/") {
// Many people will have "post/foo.md" in their content files.
p, err := c.getFromCache("/" + ref)
if err == nil && p != nil {
if context != nil {
// TODO(bep) remove this case and the message below when the storm has passed
err := wrapErr(errors.Errorf(`make non-relative ref/relref page reference(s) in page %q absolute, e.g. {{< ref "/blog/my-post.md" >}}`, context.Path()), context)
helpers.DistinctWarnLog.Println(err)
}
return p, nil
}
if err != nil {
anError = err
}
}
// Last try.
ref = strings.TrimPrefix(ref, "/")
p, err := c.getFromCache(ref)
if err != nil {
anError = err
}
if p == nil && anError != nil {
return nil, wrapErr(errors.Wrap(anError, "failed to resolve ref"), context)
}
return p, nil
}
func (*PageCollections) findPagesByKindIn(kind string, inPages page.Pages) page.Pages {
var pages page.Pages
for _, p := range inPages {
if p.Kind() == kind {
pages = append(pages, p)
}
}
return pages
}
func (c *PageCollections) findPagesByKind(kind string) page.Pages {
return c.findPagesByKindIn(kind, c.Pages())
}
func (c *PageCollections) findWorkPagesByKind(kind string) pageStatePages {
var pages pageStatePages
for _, p := range c.workAllPages {
if p.Kind() == kind {
pages = append(pages, p)
}
}
return pages
}
func (*PageCollections) findPagesByKindInWorkPages(kind string, inPages pageStatePages) page.Pages {
var pages page.Pages
for _, p := range inPages {
if p.Kind() == kind {
pages = append(pages, p)
}
}
return pages
}
func (c *PageCollections) findFirstWorkPageByKindIn(kind string) *pageState {
for _, p := range c.workAllPages {
if p.Kind() == kind {
return p
}
}
return nil
}
func (c *PageCollections) addPage(page *pageState) {
c.rawAllPages = append(c.rawAllPages, page)
}
func (c *PageCollections) removePageFilename(filename string) {
if i := c.rawAllPages.findPagePosByFilename(filename); i >= 0 {
c.clearResourceCacheForPage(c.rawAllPages[i])
c.rawAllPages = append(c.rawAllPages[:i], c.rawAllPages[i+1:]...)
}
}
func (c *PageCollections) removePage(page *pageState) {
if i := c.rawAllPages.findPagePos(page); i >= 0 {
c.clearResourceCacheForPage(c.rawAllPages[i])
c.rawAllPages = append(c.rawAllPages[:i], c.rawAllPages[i+1:]...)
}
}
func (c *PageCollections) findPagesByShortcode(shortcode string) page.Pages {
var pages page.Pages
for _, p := range c.rawAllPages {
if p.HasShortcode(shortcode) {
pages = append(pages, p)
}
}
return pages
}
func (c *PageCollections) replacePage(page *pageState) {
// will find existing page that matches filepath and remove it
c.removePage(page)
c.addPage(page)
}
func (c *PageCollections) clearResourceCacheForPage(page *pageState) {
if len(page.resources) > 0 {
page.s.ResourceSpec.DeleteCacheByPrefix(page.targetPaths().SubResourceBaseTarget)
}
}