HTML encoding is not autodetected properly

[Dinver]

Hi! When I try to recognize the encoding on sites with windows-1251, I get: 2023/08/23 21:45:10 ÑÄÎ «Ïðîìåòåé» | ÎÎÎ «Âèðòóàëüíûå òåõíîëîãèè â îáðàçîâàíèè» 2023/08/23 21:45:10 Ýëåêòðîííûå êóðñû 2023/08/23 21:45:10 Ïðîäóêòû

Example:

package main

import (
	"log"

	"github.com/gocolly/colly"
)

func main() {
	c := colly.NewCollector(
		colly.DetectCharset(),
		colly.Async(true),
	)
	c.OnHTML("title", func(e *colly.HTMLElement) {
		title := e.Text
		log.Println(title)
	})
	c.OnHTML("a[href]", func(e *colly.HTMLElement) {
		title := e.Text
		log.Println(title)
	})

	c.OnHTML("img", func(e *colly.HTMLElement) {
		title := e.Attr("alt")
		log.Println(title)
	})

	c.Visit("https://prometeus.ru/")
	c.Wait()
}

colly.DetectCharset() / c.DetectCharset = true - does not working.

[blagoySimandov]

Pretty sure this is not a problem with Colly but with the terminal. Most terminals do not support cyrillic output. If you put everything in a database everything should look fine (I've crawled cyrillic pages before and I know that it works). But in case you really need to have the output displayed in the terminal try using something like Windows PowerShell ISE - it has a fairly good support for displaying Unicode.

[Dinver]

Pretty sure this is not a problem with Colly but with the terminal. Most terminals do not support cyrillic output. If you put everything in a database everything should look fine (I've crawled cyrillic pages before and I know that it works). But in case you really need to have the output displayed in the terminal try using something like Windows PowerShell ISE - it has a fairly good support for displaying Unicode.

It's not about the terminal, this example is just to reproduce the error. The on API data is also sent incorrectly.

[WGH-]

Yeah, I can reproduce it with colly/v2, too

[Dinver]

Solved the problem, by adding a check meta[http-equiv='Content-Type'] in body, in the absence of a "charset" but with "text/html" in the header. I don't know if this is the correct approach, but it solves the problem.

response.go:

package colly

import (
	"bytes"
	"fmt"
	"io/ioutil"
	"mime"
	"net/http"
	"strings"

	"github.com/PuerkitoBio/goquery"
	"github.com/saintfish/chardet"
	"golang.org/x/net/html/charset"
)

// Response is the representation of a HTTP response made by a Collector
type Response struct {
	// StatusCode is the status code of the Response
	StatusCode int
	// Body is the content of the Response
	Body []byte
	// Ctx is a context between a Request and a Response
	Ctx *Context
	// Request is the Request object of the response
	Request *Request
	// Headers contains the Response's HTTP headers
	Headers *http.Header
	// Trace contains the HTTPTrace for the request. Will only be set by the
	// collector if Collector.TraceHTTP is set to true.
	Trace *HTTPTrace
}

// Save writes response body to disk
func (r *Response) Save(fileName string) error {
	return ioutil.WriteFile(fileName, r.Body, 0644)
}

// FileName returns the sanitized file name parsed from "Content-Disposition"
// header or from URL
func (r *Response) FileName() string {
	_, params, err := mime.ParseMediaType(r.Headers.Get("Content-Disposition"))
	if fName, ok := params["filename"]; ok && err == nil {
		return SanitizeFileName(fName)
	}
	if r.Request.URL.RawQuery != "" {
		return SanitizeFileName(fmt.Sprintf("%s_%s", r.Request.URL.Path, r.Request.URL.RawQuery))
	}
	return SanitizeFileName(strings.TrimPrefix(r.Request.URL.Path, "/"))
}

func (r *Response) fixCharset(detectCharset bool, defaultEncoding string) error {
	if len(r.Body) == 0 {
		return nil
	}
	if defaultEncoding != "" {
		tmpBody, err := encodeBytes(r.Body, "text/plain; charset="+defaultEncoding)
		if err != nil {
			return err
		}
		r.Body = tmpBody
		return nil
	}
	contentType := strings.ToLower(r.Headers.Get("Content-Type"))

	if strings.Contains(contentType, "image/") ||
		strings.Contains(contentType, "video/") ||
		strings.Contains(contentType, "audio/") ||
		strings.Contains(contentType, "font/") {
		// These MIME types should not have textual data.

		return nil
	}

	if !strings.Contains(contentType, "charset") && strings.Contains(contentType, "text/html") {
		if !detectCharset {
			return nil
		}
		contentTypeBody := checkContentTypeInBody(string(r.Body))
		if contentTypeBody != "" {
			contentType = contentTypeBody
		}
	}

	if !strings.Contains(contentType, "charset") {
		if !detectCharset {
			return nil
		}
		d := chardet.NewTextDetector()
		r, err := d.DetectBest(r.Body)
		if err != nil {
			return err
		}
		contentType = "text/plain; charset=" + r.Charset
	}
	if strings.Contains(contentType, "utf-8") || strings.Contains(contentType, "utf8") {
		return nil
	}
	tmpBody, err := encodeBytes(r.Body, contentType)
	if err != nil {
		return err
	}
	r.Body = tmpBody
	return nil
}

func encodeBytes(b []byte, contentType string) ([]byte, error) {
	r, err := charset.NewReader(bytes.NewReader(b), contentType)
	if err != nil {
		return nil, err
	}
	return ioutil.ReadAll(r)
}

func checkContentTypeInBody(b string) string {
	reader := strings.NewReader(b)
	doc, err := goquery.NewDocumentFromReader(reader)
	if err != nil {
		fmt.Println(err)
	}
	metaContent, exists := doc.Find("meta[http-equiv='Content-Type']").Attr("content")
	if exists {
		return metaContent
	} else {
		return ""
	}
}

[WGH-]

There's a specific algorithm for detecting the encoding of an HTML document defined here: https://html.spec.whatwg.org/multipage/parsing.html#determining-the-character-encoding. It also handles the <meta tags.

It's implemented in Go here: https://pkg.go.dev/golang.org/x/net/html/charset#DetermineEncoding

There's even a recipe how to integrate it into goquery: https://github.com/PuerkitoBio/goquery/wiki/Tips-and-tricks/7fad3f848d40fbc4504912e57fb52f8fcee7e348

We really should incorporate it into Colly.

[blagoySimandov]

Just did some testing. Apparently the default colly charset detection thinks the encoding is actually ISO-8859-1. I checked that by just having the "fixCharset" function, in the response file, print out the encoding. Maybe we can try to implement a new type of encoding detection or try to fix any bugs in the current ?