writefreely/read.go

341 lines
8.8 KiB
Go

/*
* Copyright © 2018-2021 Musing Studio LLC.
*
* This file is part of WriteFreely.
*
* WriteFreely is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License, included
* in the LICENSE file in this source code package.
*/
package writefreely
import (
"database/sql"
"fmt"
"html/template"
"math"
"net/http"
"strconv"
"time"
. "github.com/gorilla/feeds"
"github.com/gorilla/mux"
stripmd "github.com/writeas/go-strip-markdown/v2"
"github.com/writeas/impart"
"github.com/writeas/web-core/log"
"github.com/writeas/web-core/memo"
"github.com/writefreely/writefreely/page"
)
const (
tlFeedLimit = 100
tlAPIPageLimit = 10
tlMaxAuthorPosts = 5
tlPostsPerPage = 16
tlMaxPostCache = 250
tlCacheDur = 10 * time.Minute
)
type localTimeline struct {
m *memo.Memo
posts *[]PublicPost
// Configuration values
postsPerPage int
}
type readPublication struct {
page.StaticPage
Posts *[]PublicPost
CurrentPage int
TotalPages int
SelTopic string
IsAdmin bool
CanInvite bool
// Customizable page content
ContentTitle string
Content template.HTML
}
func initLocalTimeline(app *App) {
app.timeline = &localTimeline{
postsPerPage: tlPostsPerPage,
m: memo.New(app.FetchPublicPosts, tlCacheDur),
}
}
// satisfies memo.Func
func (app *App) FetchPublicPosts() (interface{}, error) {
// Conditions
limit := fmt.Sprintf("LIMIT %d", tlMaxPostCache)
// This is better than the hard limit when limiting posts from individual authors
// ageCond := `p.created >= ` + app.db.dateSub(3, "month") + ` AND `
// Finds all public posts and posts in a public collection published during the owner's active subscription period and within the last 3 months
rows, err := app.db.Query(`SELECT p.id, c.id, alias, c.title, p.slug, p.title, p.content, p.text_appearance, p.language, p.rtl, p.created, p.updated
FROM collections c
LEFT JOIN posts p ON p.collection_id = c.id
LEFT JOIN users u ON u.id = p.owner_id
WHERE c.privacy = 1 AND (p.created <= ` + app.db.now() + ` AND pinned_position IS NULL) AND u.status = 0
ORDER BY p.created DESC
` + limit)
if err != nil {
log.Error("Failed selecting from posts: %v", err)
return nil, impart.HTTPError{http.StatusInternalServerError, "Couldn't retrieve collection posts." + err.Error()}
}
defer rows.Close()
ap := map[string]uint{}
posts := []PublicPost{}
for rows.Next() {
p := &Post{}
c := &Collection{}
var alias, title sql.NullString
err = rows.Scan(&p.ID, &c.ID, &alias, &title, &p.Slug, &p.Title, &p.Content, &p.Font, &p.Language, &p.RTL, &p.Created, &p.Updated)
if err != nil {
log.Error("[READ] Unable to scan row, skipping: %v", err)
continue
}
c.hostName = app.cfg.App.Host
isCollectionPost := alias.Valid
if isCollectionPost {
c.Alias = alias.String
if c.Alias != "" && ap[c.Alias] == tlMaxAuthorPosts {
// Don't add post if we've hit the post-per-author limit
continue
}
c.Public = true
c.Title = title.String
c.Monetization = app.db.GetCollectionAttribute(c.ID, "monetization_pointer")
}
p.extractData()
p.handlePremiumContent(c, false, false, app.cfg)
p.HTMLContent = template.HTML(applyMarkdown([]byte(p.Content), "", app.cfg))
fp := p.processPost()
if isCollectionPost {
fp.Collection = &CollectionObj{Collection: *c}
}
posts = append(posts, fp)
ap[c.Alias]++
}
return posts, nil
}
func viewLocalTimelineAPI(app *App, w http.ResponseWriter, r *http.Request) error {
updateTimelineCache(app.timeline, false)
skip, _ := strconv.Atoi(r.FormValue("skip"))
posts := []PublicPost{}
for i := skip; i < skip+tlAPIPageLimit && i < len(*app.timeline.posts); i++ {
posts = append(posts, (*app.timeline.posts)[i])
}
return impart.WriteSuccess(w, posts, http.StatusOK)
}
func viewLocalTimeline(app *App, w http.ResponseWriter, r *http.Request) error {
if !app.cfg.App.LocalTimeline {
return impart.HTTPError{http.StatusNotFound, "Page doesn't exist."}
}
vars := mux.Vars(r)
var p int
page := 1
p, _ = strconv.Atoi(vars["page"])
if p > 0 {
page = p
}
return showLocalTimeline(app, w, r, page, vars["author"], vars["tag"])
}
// updateTimelineCache will reset and update the cache if it is stale or
// the boolean passed in is true.
func updateTimelineCache(tl *localTimeline, reset bool) {
if reset {
tl.m.Reset()
}
// Fetch posts if the cache is empty, has been reset or enough time has
// passed since last cache.
if tl.posts == nil || reset || tl.m.Invalidate() {
log.Info("[READ] Updating post cache")
postsInterfaces, err := tl.m.Get()
if err != nil {
log.Error("[READ] Unable to cache posts: %v", err)
} else {
castPosts := postsInterfaces.([]PublicPost)
tl.posts = &castPosts
}
}
}
func showLocalTimeline(app *App, w http.ResponseWriter, r *http.Request, page int, author, tag string) error {
updateTimelineCache(app.timeline, false)
pl := len(*(app.timeline.posts))
ttlPages := int(math.Ceil(float64(pl) / float64(app.timeline.postsPerPage)))
start := 0
if page > 1 {
start = app.timeline.postsPerPage * (page - 1)
if start > pl {
return impart.HTTPError{http.StatusFound, fmt.Sprintf("/read/p/%d", ttlPages)}
}
}
end := app.timeline.postsPerPage * page
if end > pl {
end = pl
}
var posts []PublicPost
if author != "" {
posts = []PublicPost{}
for _, p := range *app.timeline.posts {
if author == "anonymous" {
if p.Collection == nil {
posts = append(posts, p)
}
} else if p.Collection != nil && p.Collection.Alias == author {
posts = append(posts, p)
}
}
} else if tag != "" {
posts = []PublicPost{}
for _, p := range *app.timeline.posts {
if p.HasTag(tag) {
posts = append(posts, p)
}
}
} else {
posts = *app.timeline.posts
posts = posts[start:end]
}
d := &readPublication{
StaticPage: pageForReq(app, r),
Posts: &posts,
CurrentPage: page,
TotalPages: ttlPages,
SelTopic: tag,
}
u := getUserSession(app, r)
d.IsAdmin = u != nil && u.IsAdmin()
d.CanInvite = canUserInvite(app.cfg, d.IsAdmin)
c, err := getReaderSection(app)
if err != nil {
return err
}
d.ContentTitle = c.Title.String
d.Content = template.HTML(applyMarkdown([]byte(c.Content), "", app.cfg))
err = templates["read"].ExecuteTemplate(w, "base", d)
if err != nil {
log.Error("Unable to render reader: %v", err)
fmt.Fprintf(w, ":(")
}
return nil
}
// NextPageURL provides a full URL for the next page of collection posts
func (c *readPublication) NextPageURL(n int) string {
return fmt.Sprintf("/read/p/%d", n+1)
}
// PrevPageURL provides a full URL for the previous page of collection posts,
// returning a /page/N result for pages >1
func (c *readPublication) PrevPageURL(n int) string {
if n == 2 {
// Previous page is 1; no need for /p/ prefix
return "/read"
}
return fmt.Sprintf("/read/p/%d", n-1)
}
// handlePostIDRedirect handles a route where a post ID is given and redirects
// the user to the canonical post URL.
func handlePostIDRedirect(app *App, w http.ResponseWriter, r *http.Request) error {
vars := mux.Vars(r)
postID := vars["post"]
p, err := app.db.GetPost(postID, 0)
if err != nil {
return err
}
if !p.CollectionID.Valid {
// No collection; send to normal URL
// NOTE: not handling single user blogs here since this handler is only used for the Reader
return impart.HTTPError{http.StatusFound, app.cfg.App.Host + "/" + postID + ".md"}
}
c, err := app.db.GetCollectionBy("id = ?", fmt.Sprintf("%d", p.CollectionID.Int64))
if err != nil {
return err
}
c.hostName = app.cfg.App.Host
// Retrieve collection information and send user to canonical URL
return impart.HTTPError{http.StatusFound, c.CanonicalURL() + p.Slug.String}
}
func viewLocalTimelineFeed(app *App, w http.ResponseWriter, req *http.Request) error {
if !app.cfg.App.LocalTimeline {
return impart.HTTPError{http.StatusNotFound, "Page doesn't exist."}
}
updateTimelineCache(app.timeline, false)
feed := &Feed{
Title: app.cfg.App.SiteName + " Reader",
Link: &Link{Href: app.cfg.App.Host},
Description: "Read the latest posts from " + app.cfg.App.SiteName + ".",
Created: time.Now(),
}
c := 0
var title, permalink, author string
for _, p := range *app.timeline.posts {
if c == tlFeedLimit {
break
}
title = p.PlainDisplayTitle()
permalink = p.CanonicalURL(app.cfg.App.Host)
if p.Collection != nil {
author = p.Collection.Title
} else {
author = "Anonymous"
}
i := &Item{
Id: app.cfg.App.Host + "/read/a/" + p.ID,
Title: title,
Link: &Link{Href: permalink},
Description: "<![CDATA[" + stripmd.Strip(p.Content) + "]]>",
Content: applyMarkdown([]byte(p.Content), "", app.cfg),
Author: &Author{author, ""},
Created: p.Created,
Updated: p.Updated,
}
feed.Items = append(feed.Items, i)
c++
}
rss, err := feed.ToRss()
if err != nil {
return err
}
fmt.Fprint(w, rss)
return nil
}