'maven-javadoc-plugin error javadoc: error - cannot read Input length = 1 with non-ASCII characters in directory name

I'm using OpenJDK 11 on Windows 10. I have a very simple POM, for a single Java file, that generates Javadocs. Here is an extract:

<properties>
  <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
  <maven.compiler.source>11</maven.compiler.source>
  <maven.compiler.target>11</maven.compiler.target>
</properties>

<build>
  <plugins>
    <plugin>
      <groupId>org.apache.maven.plugins</groupId>
      <artifactId>maven-javadoc-plugin</artifactId>
      <version>3.0.1</version>
      <executions>
        <execution>
          <goals>
            <goal>jar</goal>
          </goals>
        </execution>
      </executions>
    </plugin>
  </plugins>
</build>

Strangely just running mvn clean package causes an error:

[ERROR] Failed to execute goal org.apache.maven.plugins:maven-javadoc-plugin:3.0.1:jar (default) on project foobar: MavenReportException: Error while generating Javadoc:
[ERROR] Exit code: 1 - javadoc: error - cannot read Input length = 1
[ERROR]
[ERROR] Command line was: C:\bin\jdk-11\bin\javadoc.exe @options @packages

In target/apidocs there are only three files: javadoc.bat, options, and packages. The options file is the most interesting. It explicitly says UTF-8 everywhere, as it should. But look at these lines:

-sourcepath
C:/projects/li��o 1/src/main/java

This project is in C:\projects\lição 1. It appears that somewhere along the chain Java or Maven or the Javadoc plugin didn't correctly convert the directory name to UTF-8.

Sure enough; when I renamed the directories in Windows to remove non-ASCII characters, mvn clean package worked just fine.

This would seem like a blatant bug; once Maven starts, everything should be UTF-8 throughout. Is it a problem with the Javadoc plugin? Anyone have an idea where this originates? Where should I file a bug ticket? Or am I doing something wrong?

javamavenmaven-javadoc-plugin


Solution 1:[1]

As you say, this looks like the encoding used to write the files to target/apidocs.

Looking through the source for the maven-javadoc-plugin, it is just using the platform encoding when writing these files - e.g. this line.

Directly setting the encoding while calling Maven repaired the example above for me:

mvn clean package -Dfile.encoding=UTF-8

This feels more like a workaround than a good fix though - it needs to assume there is nothing else depending on the platform encoding in the Maven build.

I think the cause is a change in the main JDK between 8 and 9. The bit of code (actually under javac) that parses the argument files (e.g. @options in the javadoc command line) has switched from using the platform encoding here to calling Files.newBufferedReader() here. Files.newBufferedReader(Path) states that it uses UTF-8 if the encoding is not specified. This means argument files, in both javac and javadoc, must now be encoded in UTF-8.

Solution 2:[2]

This maven build error was caused by adding some illegal argument in the plugin definition. For example, I used a "filter" element and it turned out that the Spring Boot version I was using did not recognize this argument resulting in an obscure "Length=1" error as you posted in the headline. By removing the illegal argument and problem was cleared.

Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution Source
Solution 1
Solution 2 William

Related Questions