PathMatchingResourcePatternResolver cannot load resources with a '#' in their file name within JARs

[tofdragon]

Affects: spring version 5.1.5

Description

Given file /validation/api#test.json within a.jar which is in the classpath...

a project dependent a.jar, in project code

ResourcePatternResolver resourceLoader = new PathMatchingResourcePatternResolver();
Resource[] resources22 = resourceLoader.getResources("classpath*:/validation/**/api#test.json");
InputStream inputStream = resources[0].getInputStream();

error:

Exception in thread "main" java.io.FileNotFoundException: JAR entry validation/api not found in a.jar
	at sun.net.www.protocol.jar.JarURLConnection.connect(JarURLConnection.java:142)
	at sun.net.www.protocol.jar.JarURLConnection.getInputStream(JarURLConnection.java:150)
	at java.net.URL.openStream(URL.java:1038)

Analysis

The URL created in org.springframework.core.io.UrlResource.createRelative(String) treats everything after # as the URL fragment, causing the file path to be incorrect when the file is last read.

URL object:

• file: a.jar!/validation/api
• path: a.jar!/validation/api
• ref: test.json

Solutions

Solution 1

extends PathMatchingResourcePatternResolver

@Override
protected Set<Resource> doFindPathMatchingJarResources() {
  ....
    
      result.add(rootDirResource.createRelative(ParseUtil.encodePath(relativePath)))
  ...
}

Solution 2

 URL url = resources[0].getURL();
            if (ResourceUtils.isJarURL(url) || ResourceUtils.isJarFileURL(url)) {
                String parent = url.toExternalForm().substring(0, url.toExternalForm().indexOf("!/validation/") + 13);
                String encodeFileName = url.toExternalForm().substring(url.toExternalForm().indexOf("!/validation/") + 13);
                String encodeFile = ParseUtil.encodePath(encodeFileName);
                URL url21 = new URL(parent + encodeFile);
                inputStream = url21.openStream();
            } else {
                inputStream = resources[0].getInputStream();
            }

Question

Is there any other better solution, or is there a problem with my use?

[sbrannen]

Thanks for raising the issue.

As an interim solution, is it possible for you to rename the JSON files so that they do not contain hash tags (i.e., #) in their file names?

Also, where does ParseUtil.encodePath(String) come from? Is that a utility that you wrote, and if so, what type of encoding does it perform?

After a bit of research, I assume you're referring to sun.net.www.ParseUtil, but Spring cannot use that class. So, perhaps java.net.URLEncoder.encode(String, String) is an alternative.

[sbrannen]

Is there any other better solution, or is there a problem with my use?

Are your JSON files always in JAR files?

If so, have you tried the following?

Resource[] resources = resourceLoader.getResources("classpath*:/validation/**/api%23test.json");

[sbrannen]

I've edited this issue to improve the formatting. You might want to check out this Mastering Markdown guide for future reference.

[sbrannen]

FYI: b173a93 demonstrates that PathMatchingResourcePatternResolver does not have a problem loading resources with a # in their file name from the filesystem.

[tofdragon]

@sbrannen

1. where does ParseUtil.encodePath(String) come from?

<1-1>ParseUtil is sun.net.www.ParseUtil. java.net.URLEncoder.encode() is OK!

<1-2>java.net.JarURLConnection

private void parseSpecs(URL url) throws MalformedURLException {
       ....
            // *****  use ParseUtil ******
            entryName = ParseUtil.decode (entryName);
       ...
    }

<1-3>I found that the spring boot also has a JarUrlConnection, and there is also a decoding function in the guess.

2. Resource[] resources = resourceLoader.getResources("classpath*:/validation/**/api%23test.json");

That does not work

PathMatchingResourcePatternResolver method

protected Set<Resource> doFindPathMatchingJarResources()

···
for (Enumeration<JarEntry> entries = jarFile.entries(); entries.hasMoreElements();) {
				JarEntry entry = entries.nextElement();
				String entryPath = entry.getName();
				if (entryPath.startsWith(rootEntryPath)) {
					String relativePath = entryPath.substring(rootEntryPath.length()); 
                                        // ******  Unable to match escaped  *****
					if (getPathMatcher().match(subPattern, relativePath)) {
						result.add(rootDirResource.createRelative(relativePath));
					}
				}
			}
···

3. PathMatchingResourcePatternResolver does not have a problem loading resources with a # in their file name from the filesystem

It is OK when in the file system

the processing logic of jar and file is not the same. When dealing with url, I did not carefully look at the source code to analyze the difference between jar and file

4. It is best to say that the file name does not contain #. Because of the project, there is no way to make a large rewrite, and the decoding has been considered in JarUrlConnection. Does it mean that the contract of Url Jar is considered to support encoding?

[lgxbslgx]

@tofdragon Could you give us a demo code which can reproduce the bug?

[tofdragon]

@lgxbslgx Please see the problem description