mindoc/models/BookModel.go

865 lines
27 KiB
Go
Raw Normal View History

2017-04-21 18:20:35 +08:00
package models
2017-04-22 17:24:17 +08:00
import (
2018-03-25 14:59:42 +08:00
"bytes"
"crypto/md5"
"errors"
"fmt"
"io"
"io/ioutil"
"os"
"path/filepath"
"regexp"
"strconv"
"strings"
2017-04-22 17:24:17 +08:00
"time"
2018-03-25 14:59:42 +08:00
"github.com/PuerkitoBio/goquery"
"github.com/astaxie/beego"
"github.com/astaxie/beego/logs"
"github.com/astaxie/beego/orm"
"github.com/lifei6671/mindoc/conf"
2018-03-24 17:24:02 +08:00
"github.com/lifei6671/mindoc/utils/cryptil"
2018-03-25 14:59:42 +08:00
"github.com/lifei6671/mindoc/utils/filetil"
2018-03-24 17:24:02 +08:00
"github.com/lifei6671/mindoc/utils/requests"
2018-03-25 14:59:42 +08:00
"github.com/lifei6671/mindoc/utils/ziptil"
2018-03-24 17:24:02 +08:00
"gopkg.in/russross/blackfriday.v2"
2017-04-22 17:24:17 +08:00
)
2017-04-21 18:20:35 +08:00
// Book struct .
type Book struct {
BookId int `orm:"pk;auto;unique;column(book_id)" json:"book_id"`
2017-04-21 18:20:35 +08:00
// BookName 项目名称.
BookName string `orm:"column(book_name);size(500)" json:"book_name"`
2017-04-21 18:20:35 +08:00
// Identify 项目唯一标识.
Identify string `orm:"column(identify);size(100);unique" json:"identify"`
//是否是自动发布 0 否/1 是
AutoRelease int `orm:"column(auto_release);type(int);default(0)" json:"auto_release"`
//是否开启下载功能 0 是/1 否
IsDownload int `orm:"column(is_download);type(int);default(0)" json:"is_download"`
2018-03-25 14:59:42 +08:00
OrderIndex int `orm:"column(order_index);type(int);default(0)" json:"order_index"`
2017-04-21 18:20:35 +08:00
// Description 项目描述.
Description string `orm:"column(description);size(2000)" json:"description"`
//发行公司
Publisher string `orm:"column(publisher);size(500)" json:"publisher"`
2018-03-25 14:59:42 +08:00
Label string `orm:"column(label);size(500)" json:"label"`
2017-04-21 18:20:35 +08:00
// PrivatelyOwned 项目私有: 0 公开/ 1 私有
PrivatelyOwned int `orm:"column(privately_owned);type(int);default(0)" json:"privately_owned"`
2017-04-21 18:20:35 +08:00
// 当项目是私有时的访问Token.
PrivateToken string `orm:"column(private_token);size(500);null" json:"private_token"`
//状态0 正常/1 已删除
Status int `orm:"column(status);type(int);default(0)" json:"status"`
2017-04-26 18:17:38 +08:00
//默认的编辑器.
Editor string `orm:"column(editor);size(50)" json:"editor"`
2017-04-21 18:20:35 +08:00
// DocCount 包含文档数量.
DocCount int `orm:"column(doc_count);type(int)" json:"doc_count"`
2017-04-22 17:24:17 +08:00
// CommentStatus 评论设置的状态:open 为允许所有人评论closed 为不允许评论, group_only 仅允许参与者评论 ,registered_only 仅允许注册者评论.
CommentStatus string `orm:"column(comment_status);size(20);default(open)" json:"comment_status"`
CommentCount int `orm:"column(comment_count);type(int)" json:"comment_count"`
//封面地址
Cover string `orm:"column(cover);size(1000)" json:"cover"`
//主题风格
Theme string `orm:"column(theme);size(255);default(default)" json:"theme"`
2017-04-21 18:20:35 +08:00
// CreateTime 创建时间 .
CreateTime time.Time `orm:"type(datetime);column(create_time);auto_now_add" json:"create_time"`
//每个文档保存的历史记录数量0 为不限制
2018-03-25 14:59:42 +08:00
HistoryCount int `orm:"column(history_count);type(int);default(0)" json:"history_count"`
2018-03-13 19:20:50 +08:00
//是否启用分享0启用/1不启用
2018-03-25 14:59:42 +08:00
IsEnableShare int `orm:"column(is_enable_share);type(int);default(0)" json:"is_enable_share"`
MemberId int `orm:"column(member_id);size(100)" json:"member_id"`
ModifyTime time.Time `orm:"type(datetime);column(modify_time);null;auto_now" json:"modify_time"`
Version int64 `orm:"type(bigint);column(version)" json:"version"`
//是否使用第一篇文章项目为默认首页,0 否/1 是
IsUseFirstDocument int `orm:"column(is_use_first_document);type(int);default(0)" json:"is_use_first_document"`
2017-04-21 18:20:35 +08:00
}
// TableName 获取对应数据库表名.
func (book *Book) TableName() string {
2017-04-21 18:20:35 +08:00
return "books"
}
2017-04-21 18:20:35 +08:00
// TableEngine 获取数据使用的引擎.
func (book *Book) TableEngine() string {
2017-04-21 18:20:35 +08:00
return "INNODB"
}
func (book *Book) TableNameWithPrefix() string {
return conf.GetDatabasePrefix() + book.TableName()
2017-04-22 17:24:17 +08:00
}
func NewBook() *Book {
return &Book{}
}
2018-03-24 17:24:02 +08:00
//添加一个项目
func (book *Book) Insert() error {
2017-04-22 17:24:17 +08:00
o := orm.NewOrm()
// o.Begin()
_, err := o.Insert(book)
2017-04-22 17:24:17 +08:00
if err == nil {
if book.Label != "" {
NewLabel().InsertOrUpdateMulti(book.Label)
}
relationship := NewRelationship()
relationship.BookId = book.BookId
relationship.RoleId = 0
relationship.MemberId = book.MemberId
err = relationship.Insert()
if err != nil {
logs.Error("插入项目与用户关联 => ", err)
2017-05-05 11:09:17 +08:00
//o.Rollback()
return err
}
document := NewDocument()
document.BookId = book.BookId
document.DocumentName = "空白文档"
document.MemberId = book.MemberId
err = document.InsertOrUpdate()
if err != nil {
2017-05-05 11:09:17 +08:00
//o.Rollback()
return err
}
2017-05-05 11:09:17 +08:00
//o.Commit()
2017-05-04 10:35:56 +08:00
return nil
}
2017-05-05 11:09:17 +08:00
//o.Rollback()
2017-04-22 17:24:17 +08:00
return err
}
func (book *Book) Find(id int) (*Book, error) {
if id <= 0 {
return book, ErrInvalidParameter
}
o := orm.NewOrm()
err := o.QueryTable(book.TableNameWithPrefix()).Filter("book_id", id).One(book)
2017-04-25 20:05:59 +08:00
return book, err
}
2018-03-25 14:59:42 +08:00
2018-03-24 17:24:02 +08:00
//更新一个项目
func (book *Book) Update(cols ...string) error {
2017-04-22 17:24:17 +08:00
o := orm.NewOrm()
temp := NewBook()
temp.BookId = book.BookId
if err := o.Read(temp); err != nil {
return err
}
2018-03-25 14:59:42 +08:00
if book.Label != "" || temp.Label != "" {
go NewLabel().InsertOrUpdateMulti(book.Label + "," + temp.Label)
}
_, err := o.Update(book, cols...)
2017-04-22 17:24:17 +08:00
return err
}
2018-07-10 18:53:41 +08:00
//复制项目
func (book *Book) Copy(identify string) error {
o := orm.NewOrm()
err := o.QueryTable(book.TableNameWithPrefix()).Filter("identify",identify).One(book)
if err != nil {
beego.Error("查询项目时出错 -> ",err)
return err
}
if err := o.Begin();err != nil {
beego.Error("开启事物时出错 -> ",err)
return err
}
bookId := book.BookId
book.BookId = 0
book.Identify = book.Identify + fmt.Sprintf("%s-%s",identify,strconv.FormatInt(time.Now().UnixNano(), 32))
book.BookName = book.BookName + "[副本]"
book.CreateTime = time.Now()
book.CommentCount = 0
book.HistoryCount = 0
if _,err := o.Insert(book);err != nil {
beego.Error("复制项目时出错 -> ",err)
o.Rollback()
return err
}
var rels []*Relationship
if _,err := o.QueryTable(NewRelationship().TableNameWithPrefix()).Filter("book_id",bookId).All(&rels); err != nil {
beego.Error("复制项目关系时出错 -> ",err)
o.Rollback()
return err
}
for _,rel := range rels {
rel.BookId = book.BookId
rel.RelationshipId = 0
if _,err := o.Insert(rel);err != nil {
beego.Error("复制项目关系时出错 -> ",err)
o.Rollback()
return err
}
}
var docs []*Document
if _,err := o.QueryTable(NewDocument().TableNameWithPrefix()).Filter("book_id",bookId).Filter("parent_id",0).All(&docs);err != nil && err != orm.ErrNoRows {
beego.Error("读取项目文档时出错 -> ",err)
o.Rollback()
return err
}
if len(docs) > 0 {
if err := recursiveInsertDocument(docs,o,book.BookId,0);err != nil {
beego.Error("复制项目时出错 -> ",err)
o.Rollback()
return err
}
}
return o.Commit()
}
//递归的复制文档
func recursiveInsertDocument(docs []*Document,o orm.Ormer,bookId int,parentId int) error {
for _,doc := range docs {
docId := doc.DocumentId
doc.DocumentId = 0
doc.ParentId = parentId
doc.BookId = bookId
doc.Version = time.Now().Unix()
if _,err := o.Insert(doc);err != nil {
beego.Error("插入项目时出错 -> ",err)
return err
}
var attachList []*Attachment
//读取所有附件列表
if _,err := o.QueryTable(NewAttachment().TableNameWithPrefix()).Filter("document_id",docId).All(&attachList); err == nil {
for _,attach := range attachList {
attach.BookId = bookId
attach.DocumentId = doc.DocumentId
attach.AttachmentId = 0
if _,err := o.Insert(attach);err != nil {
return err
}
}
}
var subDocs []*Document
if _,err := o.QueryTable(NewDocument().TableNameWithPrefix()).Filter("parent_id",docId).All(&subDocs);err != nil && err != orm.ErrNoRows {
beego.Error("读取文档时出错 -> ",err)
return err
}
if len(subDocs) > 0{
if err := recursiveInsertDocument(subDocs,o,bookId,doc.DocumentId);err != nil {
return err
}
}
}
return nil
}
//根据指定字段查询结果集.
2018-03-25 14:59:42 +08:00
func (book *Book) FindByField(field string, value interface{},cols ...string) ([]*Book, error) {
2017-04-22 17:24:17 +08:00
o := orm.NewOrm()
var books []*Book
2018-03-25 14:59:42 +08:00
_, err := o.QueryTable(book.TableNameWithPrefix()).Filter(field, value).All(&books,cols...)
2017-04-22 17:24:17 +08:00
return books, err
2017-04-22 17:24:17 +08:00
}
//根据指定字段查询一个结果.
func (book *Book) FindByFieldFirst(field string, value interface{}) (*Book, error) {
o := orm.NewOrm()
err := o.QueryTable(book.TableNameWithPrefix()).Filter(field, value).One(book)
return book, err
}
2018-03-24 17:24:02 +08:00
//根据项目标识查询项目
func (book *Book) FindByIdentify(identify string) (*Book, error) {
o := orm.NewOrm()
err := o.QueryTable(book.TableNameWithPrefix()).Filter("identify", identify).One(book)
return book, err
}
//分页查询指定用户的项目
func (book *Book) FindToPager(pageIndex, pageSize, memberId int) (books []*BookResult, totalCount int, err error) {
2017-04-22 17:24:17 +08:00
relationship := NewRelationship()
o := orm.NewOrm()
sql1 := "SELECT COUNT(book.book_id) AS total_count FROM " + book.TableNameWithPrefix() + " AS book LEFT JOIN " +
relationship.TableNameWithPrefix() + " AS rel ON book.book_id=rel.book_id AND rel.member_id = ? WHERE rel.relationship_id > 0 "
2017-04-22 17:24:17 +08:00
err = o.Raw(sql1, memberId).QueryRow(&totalCount)
2017-04-22 17:24:17 +08:00
if err != nil {
return
}
offset := (pageIndex - 1) * pageSize
sql2 := "SELECT book.*,rel.member_id,rel.role_id,m.account as create_name FROM " + book.TableNameWithPrefix() + " AS book" +
" LEFT JOIN " + relationship.TableNameWithPrefix() + " AS rel ON book.book_id=rel.book_id AND rel.member_id = ?" +
" LEFT JOIN " + relationship.TableNameWithPrefix() + " AS rel1 ON book.book_id=rel1.book_id AND rel1.role_id=0" +
" LEFT JOIN " + NewMember().TableNameWithPrefix() + " AS m ON rel1.member_id=m.member_id " +
" WHERE rel.relationship_id > 0 ORDER BY book.order_index DESC,book.book_id DESC LIMIT " + fmt.Sprintf("%d,%d", offset, pageSize)
_, err = o.Raw(sql2, memberId).QueryRows(&books)
if err != nil {
logs.Error("分页查询项目列表 => ", err)
return
}
2017-04-22 17:24:17 +08:00
sql := "SELECT m.account,doc.modify_time FROM md_documents AS doc LEFT JOIN md_members AS m ON doc.modify_at=m.member_id WHERE book_id = ? LIMIT 1 ORDER BY doc.modify_time DESC"
if err == nil && len(books) > 0 {
for index, book := range books {
var text struct {
Account string
2017-04-22 17:24:17 +08:00
ModifyTime time.Time
}
err1 := o.Raw(sql, book.BookId).QueryRow(&text)
2017-04-22 17:24:17 +08:00
if err1 == nil {
books[index].LastModifyText = text.Account + " 于 " + text.ModifyTime.Format("2006-01-02 15:04:05")
}
if book.RoleId == 0 {
2017-04-22 17:24:17 +08:00
book.RoleName = "创始人"
} else if book.RoleId == 1 {
2017-04-22 17:24:17 +08:00
book.RoleName = "管理员"
} else if book.RoleId == 2 {
2017-04-22 17:24:17 +08:00
book.RoleName = "编辑者"
} else if book.RoleId == 3 {
2017-04-22 17:24:17 +08:00
book.RoleName = "观察者"
}
}
}
return
2017-04-23 12:48:46 +08:00
}
// 彻底删除项目.
func (book *Book) ThoroughDeleteBook(id int) error {
if id <= 0 {
return ErrInvalidParameter
}
o := orm.NewOrm()
2018-03-25 14:59:42 +08:00
book, err := book.Find(id)
if err != nil {
return err
}
o.Begin()
sql2 := "DELETE FROM " + NewDocument().TableNameWithPrefix() + " WHERE book_id = ?"
_, err = o.Raw(sql2, book.BookId).Exec()
if err != nil {
o.Rollback()
return err
}
sql3 := "DELETE FROM " + book.TableNameWithPrefix() + " WHERE book_id = ?"
_, err = o.Raw(sql3, book.BookId).Exec()
if err != nil {
o.Rollback()
return err
}
sql4 := "DELETE FROM " + NewRelationship().TableNameWithPrefix() + " WHERE book_id = ?"
_, err = o.Raw(sql4, book.BookId).Exec()
if err != nil {
o.Rollback()
2018-04-02 15:52:36 +08:00
return err
}
if book.Label != "" {
NewLabel().InsertOrUpdateMulti(book.Label)
}
2018-03-25 14:59:42 +08:00
os.RemoveAll(filepath.Join(conf.WorkingDirectory, "uploads", "books", strconv.Itoa(id)))
return o.Commit()
}
2017-05-13 12:12:37 +08:00
//分页查找系统首页数据.
func (book *Book) FindForHomeToPager(pageIndex, pageSize, member_id int) (books []*BookResult, totalCount int, err error) {
o := orm.NewOrm()
offset := (pageIndex - 1) * pageSize
//如果是登录用户
if member_id > 0 {
sql1 := "SELECT COUNT(*) FROM md_books AS book LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.member_id = ? WHERE relationship_id > 0 OR book.privately_owned = 0"
err = o.Raw(sql1, member_id).QueryRow(&totalCount)
if err != nil {
return
}
sql2 := `SELECT book.*,rel1.*,member.account AS create_name,member.real_name FROM md_books AS book
LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.member_id = ?
LEFT JOIN md_relationship AS rel1 ON rel1.book_id = book.book_id AND rel1.role_id = 0
LEFT JOIN md_members AS member ON rel1.member_id = member.member_id
WHERE rel.relationship_id > 0 OR book.privately_owned = 0 ORDER BY order_index DESC ,book.book_id DESC LIMIT ?,?`
_, err = o.Raw(sql2, member_id, offset, pageSize).QueryRows(&books)
} else {
count, err1 := o.QueryTable(book.TableNameWithPrefix()).Filter("privately_owned", 0).Count()
if err1 != nil {
err = err1
return
}
totalCount = int(count)
sql := `SELECT book.*,rel.*,member.account AS create_name,member.real_name FROM md_books AS book
LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.role_id = 0
LEFT JOIN md_members AS member ON rel.member_id = member.member_id
WHERE book.privately_owned = 0 ORDER BY order_index DESC ,book.book_id DESC LIMIT ?,?`
_, err = o.Raw(sql, offset, pageSize).QueryRows(&books)
}
return
}
//分页全局搜索.
func (book *Book) FindForLabelToPager(keyword string, pageIndex, pageSize, memberId int) (books []*BookResult, totalCount int, err error) {
o := orm.NewOrm()
keyword = "%" + keyword + "%"
offset := (pageIndex - 1) * pageSize
//如果是登录用户
if memberId > 0 {
sql1 := "SELECT COUNT(*) FROM md_books AS book LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.member_id = ? WHERE (relationship_id > 0 OR book.privately_owned = 0) AND book.label LIKE ?"
err = o.Raw(sql1, memberId, keyword).QueryRow(&totalCount)
if err != nil {
return
}
sql2 := `SELECT book.*,rel1.*,member.account AS create_name FROM md_books AS book
LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.member_id = ?
LEFT JOIN md_relationship AS rel1 ON rel1.book_id = book.book_id AND rel1.role_id = 0
LEFT JOIN md_members AS member ON rel1.member_id = member.member_id
WHERE (rel.relationship_id > 0 OR book.privately_owned = 0) AND book.label LIKE ? ORDER BY order_index DESC ,book.book_id DESC LIMIT ?,?`
_, err = o.Raw(sql2, memberId, keyword, offset, pageSize).QueryRows(&books)
return
} else {
count, err1 := o.QueryTable(NewBook().TableNameWithPrefix()).Filter("privately_owned", 0).Filter("label__icontains", keyword).Count()
if err1 != nil {
err = err1
return
}
totalCount = int(count)
sql := `SELECT book.*,rel.*,member.account AS create_name FROM md_books AS book
LEFT JOIN md_relationship AS rel ON rel.book_id = book.book_id AND rel.role_id = 0
LEFT JOIN md_members AS member ON rel.member_id = member.member_id
WHERE book.privately_owned = 0 AND book.label LIKE ? ORDER BY order_index DESC ,book.book_id DESC LIMIT ?,?`
_, err = o.Raw(sql, keyword, offset, pageSize).QueryRows(&books)
return
}
}
2018-03-25 14:59:42 +08:00
//发布文档
func (book *Book) ReleaseContent(bookId int) {
o := orm.NewOrm()
var docs []*Document
_, err := o.QueryTable(NewDocument().TableNameWithPrefix()).Filter("book_id", bookId).All(&docs, "document_id", "identify", "content")
if err != nil {
2018-04-02 15:52:36 +08:00
beego.Error("发布失败 =>",bookId, err)
2018-03-25 14:59:42 +08:00
return
}
for _, item := range docs {
if item.Content != "" {
item.Release = item.Content
bufio := bytes.NewReader([]byte(item.Content))
//解析文档中非本站的链接,并设置为新窗口打开
if content, err := goquery.NewDocumentFromReader(bufio); err == nil {
content.Find("a").Each(func(i int, contentSelection *goquery.Selection) {
if src, ok := contentSelection.Attr("href"); ok {
if strings.HasPrefix(src, "http://") || strings.HasPrefix(src, "https://") {
//beego.Info(src,conf.BaseUrl,strings.HasPrefix(src,conf.BaseUrl))
if conf.BaseUrl != "" && !strings.HasPrefix(src, conf.BaseUrl) {
contentSelection.SetAttr("target", "_blank")
if html, err := content.Html(); err == nil {
item.Release = html
}
}
}
}
})
}
}
attachList, err := NewAttachment().FindListByDocumentId(item.DocumentId)
if err == nil && len(attachList) > 0 {
content := bytes.NewBufferString("<div class=\"attach-list\"><strong>附件</strong><ul>")
for _, attach := range attachList {
if strings.HasPrefix(attach.HttpPath, "/") {
attach.HttpPath = strings.TrimSuffix(conf.BaseUrl, "/") + attach.HttpPath
}
li := fmt.Sprintf("<li><a href=\"%s\" target=\"_blank\" title=\"%s\">%s</a></li>", attach.HttpPath, attach.FileName, attach.FileName)
content.WriteString(li)
}
content.WriteString("</ul></div>")
item.Release += content.String()
}
_, err = o.Update(item, "release")
if err != nil {
beego.Error(fmt.Sprintf("发布失败 => %+v", item), err)
} else {
//当文档发布后,需要清除已缓存的转换文档和文档缓存
if doc, err := NewDocument().Find(item.DocumentId); err == nil {
doc.PutToCache()
} else {
doc.RemoveCache()
}
2018-04-02 15:52:36 +08:00
if err := os.RemoveAll(filepath.Join(conf.WorkingDirectory, "uploads", "books", strconv.Itoa(bookId))); err != nil {
beego.Error("删除已缓存的文档目录失败 => ",filepath.Join(conf.WorkingDirectory, "uploads", "books", strconv.Itoa(bookId)))
}
2018-03-25 14:59:42 +08:00
}
}
}
//重置文档数量
func (book *Book) ResetDocumentNumber(bookId int) {
o := orm.NewOrm()
totalCount, err := o.QueryTable(NewDocument().TableNameWithPrefix()).Filter("book_id", bookId).Count()
if err == nil {
2018-04-02 15:52:36 +08:00
_,err = o.Raw("UPDATE md_books SET doc_count = ? WHERE book_id = ?", int(totalCount), bookId).Exec()
if err != nil {
beego.Error("重置文档数量失败 =>",bookId,err)
}
} else {
2018-04-02 15:52:36 +08:00
beego.Error("获取文档数量失败 =>",bookId,err)
}
}
2018-03-24 17:24:02 +08:00
2018-07-10 18:53:41 +08:00
//导入项目
2018-03-25 14:59:42 +08:00
func (book *Book) ImportBook(zipPath string) error {
2018-03-24 17:24:02 +08:00
if !filetil.FileExists(zipPath) {
2018-03-25 14:59:42 +08:00
return errors.New("文件不存在 => " + zipPath)
2018-03-24 17:24:02 +08:00
}
w := md5.New()
2018-03-25 14:59:42 +08:00
io.WriteString(w, zipPath) //将str写入到w中
2018-03-24 17:24:02 +08:00
io.WriteString(w, time.Now().String())
io.WriteString(w, book.BookName)
2018-03-24 17:24:02 +08:00
md5str := fmt.Sprintf("%x", w.Sum(nil)) //w.Sum(nil)将w的hash转成[]byte格式
2018-03-25 14:59:42 +08:00
tempPath := filepath.Join(os.TempDir(), md5str)
2018-03-24 17:24:02 +08:00
2018-04-02 15:52:36 +08:00
if err := os.MkdirAll(tempPath, 0766); err != nil {
beego.Error("创建导入目录出错 => ",err)
}
2018-03-24 17:24:02 +08:00
//如果加压缩失败
2018-03-25 14:59:42 +08:00
if err := ziptil.Unzip(zipPath, tempPath); err != nil {
2018-03-24 17:24:02 +08:00
return err
}
2018-03-25 14:59:42 +08:00
//当导入结束后,删除临时文件
2018-07-23 09:54:00 +08:00
//defer os.RemoveAll(tempPath)
2018-03-25 14:59:42 +08:00
for {
//如果当前目录下只有一个目录,则重置根目录
if entries, err := ioutil.ReadDir(tempPath); err == nil && len(entries) == 1 {
dir := entries[0]
if dir.IsDir() && dir.Name() != "." && dir.Name() != ".." {
tempPath = filepath.Join(tempPath, dir.Name())
break
}
} else {
break
}
}
2018-03-24 17:24:02 +08:00
2018-03-25 14:59:42 +08:00
tempPath = strings.Replace(tempPath, "\\", "/", -1)
2018-03-25 14:59:42 +08:00
docMap := make(map[string]int, 0)
o := orm.NewOrm()
o.Insert(book)
relationship := NewRelationship()
relationship.BookId = book.BookId
relationship.RoleId = 0
relationship.MemberId = book.MemberId
relationship.Insert()
2018-03-24 17:24:02 +08:00
err := filepath.Walk(tempPath, func(path string, info os.FileInfo, err error) error {
2018-03-25 14:59:42 +08:00
path = strings.Replace(path, "\\", "/", -1)
2018-03-24 17:24:02 +08:00
if path == tempPath {
return nil
}
if !info.IsDir() {
ext := filepath.Ext(info.Name())
//如果是Markdown文件
2018-03-25 14:59:42 +08:00
if strings.EqualFold(ext, ".md") || strings.EqualFold(ext, ".markdown") {
beego.Info("正在处理 =>",path,info.Name())
2018-03-24 17:24:02 +08:00
doc := NewDocument()
doc.BookId = book.BookId
doc.MemberId = book.MemberId
2018-03-24 17:24:02 +08:00
docIdentify := strings.Replace(strings.TrimPrefix(path, tempPath+"/"), "/", "-", -1)
if ok, err := regexp.MatchString(`[a-z]+[a-zA-Z0-9_.\-]*$`, docIdentify); !ok || err != nil {
docIdentify = "import-" + docIdentify
}
doc.Identify = docIdentify
//匹配图片,如果图片语法是在代码块中,这里同样会处理
2018-03-24 17:24:02 +08:00
re := regexp.MustCompile(`!\[(.*?)\]\((.*?)\)`)
markdown, err := filetil.ReadFileAndIgnoreUTF8BOM(path)
2018-03-24 17:24:02 +08:00
if err != nil {
return err
}
2018-03-25 10:56:52 +08:00
//处理图片
2018-03-24 17:24:02 +08:00
doc.Markdown = re.ReplaceAllStringFunc(string(markdown), func(image string) string {
2018-03-25 14:59:42 +08:00
images := re.FindAllSubmatch([]byte(image), -1)
2018-03-24 17:24:02 +08:00
if len(images) <= 0 || len(images[0]) < 3 {
return image
}
originalImageUrl := string(images[0][2])
imageUrl := strings.Replace(string(originalImageUrl), "\\", "/", -1)
2018-03-24 17:24:02 +08:00
//如果是本地路径,则需要将图片复制到项目目录
2018-07-23 09:54:00 +08:00
if !strings.HasPrefix(imageUrl, "http://") &&
!strings.HasPrefix(imageUrl, "https://") &&
!strings.HasPrefix(imageUrl,"ftp://") {
//如果路径中存在参数
if l := strings.Index(imageUrl,"?"); l > 0 {
imageUrl = imageUrl[:l]
}
2018-03-24 17:24:02 +08:00
if strings.HasPrefix(imageUrl, "/") {
imageUrl = filepath.Join(tempPath, imageUrl)
} else if strings.HasPrefix(imageUrl, "./") {
imageUrl = filepath.Join(filepath.Dir(path), strings.TrimPrefix(imageUrl, "./"))
} else if strings.HasPrefix(imageUrl, "../") {
imageUrl = filepath.Join(filepath.Dir(path), imageUrl)
} else {
imageUrl = filepath.Join(filepath.Dir(path), imageUrl)
}
imageUrl = strings.Replace(imageUrl, "\\", "/", -1)
dstFile := filepath.Join(conf.WorkingDirectory, "uploads", time.Now().Format("200601"), strings.TrimPrefix(imageUrl, tempPath))
2018-03-24 17:24:02 +08:00
if filetil.FileExists(imageUrl) {
filetil.CopyFile(imageUrl, dstFile)
2018-03-24 17:24:02 +08:00
imageUrl = strings.TrimPrefix(strings.Replace(dstFile,"\\","/",-1), strings.Replace(conf.WorkingDirectory,"\\","/",-1))
2018-03-24 17:24:02 +08:00
if !strings.HasPrefix(imageUrl, "/") && !strings.HasPrefix(imageUrl, "\\") {
2018-03-24 17:24:02 +08:00
imageUrl = "/" + imageUrl
}
}
} else {
2018-03-24 17:24:02 +08:00
imageExt := cryptil.Md5Crypt(imageUrl) + filepath.Ext(imageUrl)
dstFile := filepath.Join(conf.WorkingDirectory, "uploads", time.Now().Format("200601"), imageExt)
2018-03-24 17:24:02 +08:00
if err := requests.DownloadAndSaveFile(imageUrl, dstFile); err == nil {
imageUrl = strings.TrimPrefix(strings.Replace(dstFile, "\\", "/", -1), strings.Replace(conf.WorkingDirectory, "\\", "/", -1))
if !strings.HasPrefix(imageUrl, "/") && !strings.HasPrefix(imageUrl, "\\") {
2018-03-24 17:24:02 +08:00
imageUrl = "/" + imageUrl
}
}
}
imageUrl = strings.Replace(strings.TrimSuffix(image, originalImageUrl+")")+ conf.URLForWithCdnImage(imageUrl) +")", "\\", "/", -1)
2018-03-24 17:24:02 +08:00
return imageUrl
})
2018-03-25 10:56:52 +08:00
linkRegexp := regexp.MustCompile(`\[(.*?)\]\((.*?)\)`)
2018-07-23 09:54:00 +08:00
//处理链接
2018-03-25 10:56:52 +08:00
doc.Markdown = linkRegexp.ReplaceAllStringFunc(doc.Markdown, func(link string) string {
2018-03-25 14:59:42 +08:00
links := linkRegexp.FindAllStringSubmatch(link, -1)
originalLink := links[0][2]
2018-07-23 09:54:00 +08:00
var linkPath string
var err error
//如果是从根目录开始,
if strings.HasPrefix(originalLink,"/") {
linkPath,err = filepath.Abs(filepath.Join(tempPath,originalLink))
}else if strings.HasPrefix(originalLink, "./"){
linkPath, err = filepath.Abs(filepath.Join(filepath.Dir(path), originalLink[1:]))
}else{
linkPath, err = filepath.Abs(filepath.Join(filepath.Dir(path), originalLink))
}
2018-03-25 10:56:52 +08:00
if err == nil {
//如果本地存在该链接
if filetil.FileExists(linkPath) {
ext := filepath.Ext(linkPath)
2018-07-23 11:02:03 +08:00
//beego.Info("当前后缀 -> ",ext)
2018-03-25 10:56:52 +08:00
//如果链接是Markdown文件则生成文档标识,否则,将目标文件复制到项目目录
2018-03-25 14:59:42 +08:00
if strings.EqualFold(ext, ".md") || strings.EqualFold(ext, ".markdown") {
docIdentify := strings.Replace(strings.TrimPrefix(strings.Replace(linkPath, "\\", "/", -1), tempPath+"/"), "/", "-", -1)
//beego.Info(originalLink, "|", linkPath, "|", docIdentify)
2018-03-25 10:56:52 +08:00
if ok, err := regexp.MatchString(`[a-z]+[a-zA-Z0-9_.\-]*$`, docIdentify); !ok || err != nil {
docIdentify = "import-" + docIdentify
}
2018-03-25 14:59:42 +08:00
docIdentify = strings.TrimSuffix(docIdentify, "-README.md")
link = strings.TrimSuffix(link, originalLink+")") + conf.URLFor("DocumentController.Read", ":key", book.Identify, ":id", docIdentify) + ")"
2018-07-23 09:54:00 +08:00
2018-03-25 14:59:42 +08:00
} else {
dstPath := filepath.Join(conf.WorkingDirectory, "uploads", time.Now().Format("200601"), originalLink)
filetil.CopyFile(linkPath, dstPath)
tempLink := conf.BaseUrl + strings.TrimPrefix(strings.Replace(dstPath, "\\", "/", -1), strings.Replace(conf.WorkingDirectory, "\\", "/", -1))
2018-03-25 14:59:42 +08:00
link = strings.TrimSuffix(link, originalLink+")") + tempLink + ")"
2018-03-25 14:59:42 +08:00
2018-03-25 10:56:52 +08:00
}
2018-07-23 09:54:00 +08:00
}else{
beego.Info("文件不存在 ->",linkPath)
2018-03-25 10:56:52 +08:00
}
}
return link
})
2018-03-26 13:32:02 +08:00
//codeRe := regexp.MustCompile("```\\w+")
//doc.Markdown = codeRe.ReplaceAllStringFunc(doc.Markdown, func(s string) string {
// //beego.Info(s)
// return strings.Replace(s,"```","``` ",-1)
//})
doc.Content = string(blackfriday.Run([]byte(doc.Markdown)))
2018-03-25 14:59:42 +08:00
doc.Version = time.Now().Unix()
//解析文档名称默认使用第一个h标签为标题
docName := strings.TrimSuffix(info.Name(), ext)
for _, line := range strings.Split(doc.Markdown, "\n") {
if strings.HasPrefix(line, "#") {
docName = strings.TrimLeft(line, "#")
break
}
}
doc.DocumentName = strings.TrimSpace(docName)
parentId := 0
2018-03-25 14:59:42 +08:00
parentIdentify := strings.Replace(strings.Trim(strings.TrimSuffix(strings.TrimPrefix(path, tempPath), info.Name()), "/"), "/", "-", -1)
if parentIdentify != "" {
2018-03-24 23:17:50 +08:00
if ok, err := regexp.MatchString(`[a-z]+[a-zA-Z0-9_.\-]*$`, parentIdentify); !ok || err != nil {
parentIdentify = "import-" + parentIdentify
}
if id, ok := docMap[parentIdentify]; ok {
parentId = id
}
}
2018-03-24 23:17:50 +08:00
if strings.EqualFold(info.Name(), "README.md") {
2018-03-25 14:59:42 +08:00
beego.Info(path, "|", info.Name(), "|", parentIdentify, "|", parentId)
2018-03-24 23:17:50 +08:00
}
isInsert := false
//如果当前文件是README.md则将内容更新到父级
2018-03-24 23:17:50 +08:00
if strings.EqualFold(info.Name(), "README.md") && parentId != 0 {
doc.DocumentId = parentId
2018-03-24 23:17:50 +08:00
//beego.Info(path,"|",parentId)
} else {
2018-03-24 23:17:50 +08:00
//beego.Info(path,"|",parentIdentify)
doc.ParentId = parentId
isInsert = true
}
2018-03-25 14:59:42 +08:00
if err := doc.InsertOrUpdate("document_name", "markdown", "content"); err != nil {
beego.Error(doc.DocumentId, err)
}
if isInsert {
docMap[docIdentify] = doc.DocumentId
}
}
2018-03-25 14:59:42 +08:00
} else {
//如果当前目录下存在Markdown文件则需要创建此节点
2018-03-25 14:59:42 +08:00
if filetil.HasFileOfExt(path, []string{".md", ".markdown"}) {
beego.Info("正在处理 =>",path,info.Name())
identify := strings.Replace(strings.Trim(strings.TrimPrefix(path, tempPath), "/"), "/", "-", -1)
if ok, err := regexp.MatchString(`[a-z]+[a-zA-Z0-9_.\-]*$`, identify); !ok || err != nil {
identify = "import-" + identify
}
2018-03-24 23:17:50 +08:00
parentDoc := NewDocument()
parentDoc.MemberId = book.MemberId
parentDoc.BookId = book.BookId
parentDoc.Identify = identify
parentDoc.Version = time.Now().Unix()
parentDoc.DocumentName = "空白文档"
parentId := 0
2018-03-24 23:17:50 +08:00
2018-03-25 14:59:42 +08:00
parentIdentify := strings.TrimSuffix(identify, "-"+info.Name())
2018-03-25 14:59:42 +08:00
if id, ok := docMap[parentIdentify]; ok {
parentId = id
}
parentDoc.ParentId = parentId
2018-03-25 14:59:42 +08:00
if err := parentDoc.InsertOrUpdate(); err != nil {
beego.Error(err)
}
2018-03-24 17:24:02 +08:00
docMap[identify] = parentDoc.DocumentId
2018-03-25 14:59:42 +08:00
//beego.Info(path,"|",parentDoc.DocumentId,"|",identify,"|",info.Name(),"|",parentIdentify)
2018-03-24 17:24:02 +08:00
}
}
return nil
})
2018-03-25 14:59:42 +08:00
if err != nil {
beego.Error("导入项目异常 => ", err)
book.Description = "【项目导入存在错误:" + err.Error() + "】"
}
beego.Info("项目导入完毕 => ",book.BookName)
2018-03-25 14:59:42 +08:00
book.ReleaseContent(book.BookId)
2018-03-24 17:24:02 +08:00
return err
2018-03-25 14:59:42 +08:00
}