mirror of
https://github.com/miniflux/v2.git
synced 2025-06-27 16:36:00 +00:00
- Replace a completely overkill regex - Use `.Remove()` instead of a hand-rolled loop - Use a strings.Builder instead of a bytes.NewBufferString - Replace a call to Fprintf with string concatenation, as the latter are much faster - Remove a superfluous cast - Delay some computations - Add some tests
163 lines
3.3 KiB
Go
163 lines
3.3 KiB
Go
// SPDX-FileCopyrightText: Copyright The Miniflux Authors. All rights reserved.
|
|
// SPDX-License-Identifier: Apache-2.0
|
|
|
|
package readability // import "miniflux.app/v2/internal/reader/readability"
|
|
|
|
import (
|
|
"strings"
|
|
"testing"
|
|
)
|
|
|
|
func TestBaseURL(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<base href="https://example.org/ ">
|
|
</head>
|
|
<body>
|
|
<article>
|
|
Some content
|
|
</article>
|
|
</body>
|
|
</html>`
|
|
|
|
baseURL, _, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
if baseURL != "https://example.org/" {
|
|
t.Errorf(`Unexpected base URL, got %q instead of "https://example.org/"`, baseURL)
|
|
}
|
|
}
|
|
|
|
func TestMultipleBaseURL(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<base href="https://example.org/ ">
|
|
<base href="https://example.com/ ">
|
|
</head>
|
|
<body>
|
|
<article>
|
|
Some content
|
|
</article>
|
|
</body>
|
|
</html>`
|
|
|
|
baseURL, _, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
if baseURL != "https://example.org/" {
|
|
t.Errorf(`Unexpected base URL, got %q instead of "https://example.org/"`, baseURL)
|
|
}
|
|
}
|
|
|
|
func TestRelativeBaseURL(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<base href="/test/ ">
|
|
</head>
|
|
<body>
|
|
<article>
|
|
Some content
|
|
</article>
|
|
</body>
|
|
</html>`
|
|
|
|
baseURL, _, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
if baseURL != "" {
|
|
t.Errorf(`Unexpected base URL, got %q`, baseURL)
|
|
}
|
|
}
|
|
|
|
func TestWithoutBaseURL(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<title>Test</title>
|
|
</head>
|
|
<body>
|
|
<article>
|
|
Some content
|
|
</article>
|
|
</body>
|
|
</html>`
|
|
|
|
baseURL, _, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
if baseURL != "" {
|
|
t.Errorf(`Unexpected base URL, got %q instead of ""`, baseURL)
|
|
}
|
|
}
|
|
|
|
func TestRemoveStyleScript(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<title>Test</title>
|
|
<script src="tololo.js"></script>
|
|
</head>
|
|
<body>
|
|
<script src="tololo.js"></script>
|
|
<style>
|
|
h1 {color:red;}
|
|
p {color:blue;}
|
|
</style>
|
|
<article>Some content</article>
|
|
</body>
|
|
</html>`
|
|
want := `<div><div><article>Somecontent</article></div></div>`
|
|
|
|
_, content, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
content = strings.ReplaceAll(content, "\n", "")
|
|
content = strings.ReplaceAll(content, " ", "")
|
|
content = strings.ReplaceAll(content, "\t", "")
|
|
|
|
if content != want {
|
|
t.Errorf(`Invalid content, got %s instead of %s`, content, want)
|
|
}
|
|
}
|
|
|
|
func TestRemoveBlacklist(t *testing.T) {
|
|
html := `
|
|
<html>
|
|
<head>
|
|
<title>Test</title>
|
|
</head>
|
|
<body>
|
|
<article class="super-ad">Some content</article>
|
|
<article class="g-plus-crap">Some other thing</article>
|
|
<article class="stuff popupbody">And more</article>
|
|
<article class="legit">Valid!</article>
|
|
</body>
|
|
</html>`
|
|
want := `<div><div><articleclass="legit">Valid!</article></div></div>`
|
|
|
|
_, content, err := ExtractContent(strings.NewReader(html))
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
|
|
content = strings.ReplaceAll(content, "\n", "")
|
|
content = strings.ReplaceAll(content, " ", "")
|
|
content = strings.ReplaceAll(content, "\t", "")
|
|
|
|
if content != want {
|
|
t.Errorf(`Invalid content, got %s instead of %s`, content, want)
|
|
}
|
|
}
|