Fix exception caused by non-ascii paths

doc/CHANGELOG.rst

@@ -23,6 +23,8 @@ Latest Changes:
 
   - Support of np.float16 conversion with arrays.
 
+  - Fixed a problem that caused ``dir(jpype.JPackage("mypackage"))`` to fail if
+    the class path contained non-ascii characters. See issue #1194.
 
 - **1.5.0 - 2023-04-03**
 

jpype/_core.py

@@ -19,6 +19,7 @@
 
 import atexit
 import os
+from pathlib import Path
 import sys
 import typing
 
@@ -31,6 +32,8 @@
 
 from ._jvmfinder import *
 
+from jpype._classpath import addClassPath
+
 # This import is required to bootstrap importlib, _jpype uses importlib.util
 # but on some systems it may not load properly from C.  To make sure it gets
 # loaded properly we are going to force the issue even through we don't
@@ -115,12 +118,15 @@ def _hasClassPath(args) -> bool:
     return False
 
 
-def _handleClassPath(classpath) -> str:
+def _handleClassPath(
+    classpath: typing.Union[typing.Sequence[_PathOrStr], _PathOrStr, None] = None,
+) -> typing.Sequence[str]:
     """
     Return a classpath which represents the given tuple of classpath specifications
     """
     out = []
-
+    if classpath is None:
+        return out
     if isinstance(classpath, (str, os.PathLike)):
         classpath = (classpath,)
     try:
@@ -145,7 +151,7 @@ def _handleClassPath(classpath) -> str:
             out.extend(glob.glob(pth + '.jar'))
         else:
             out.append(pth)
-    return _classpath._SEP.join(out)
+    return out
 
 
 _JVM_started = False
@@ -221,8 +227,6 @@ def startJVM(
         # Allow the path to be a PathLike.
         jvmpath = os.fspath(jvmpath)
 
-    extra_jvm_args: typing.Tuple[str, ...] = tuple()
-
     # Classpath handling
     if _hasClassPath(jvmargs):
         # Old style, specified in the arguments
@@ -233,18 +237,14 @@ def startJVM(
         # Not specified at all, use the default classpath.
         classpath = _classpath.getClassPath()
 
-    # Handle strings and list of strings.
-    if classpath:
-        extra_jvm_args += (f'-Djava.class.path={_handleClassPath(classpath)}', )
-
     try:
         import locale
         # Gather a list of locale settings that Java may override (excluding LC_ALL)
         categories = [getattr(locale, i) for i in dir(locale) if i.startswith('LC_') and i != 'LC_ALL']
         # Keep the current locale settings, else Java will replace them.
         prior = [locale.getlocale(i) for i in categories]
         # Start the JVM
-        _jpype.startup(jvmpath, jvmargs + extra_jvm_args,
+        _jpype.startup(jvmpath, jvmargs,
                        ignoreUnrecognized, convertStrings, interrupt)
         # Collect required resources for operation
         initializeResources()
@@ -264,6 +264,21 @@ def startJVM(
                 raise RuntimeError(f"{jvmpath} is older than required Java version{version}") from ex
         raise
 
+    """Prior versions of JPype used the jvmargs to setup the class paths via 
+    JNI (Java Native Interface) option strings:
+    i.e -Djava.class.path=... 
+    See: https://docs.oracle.com/javase/7/docs/technotes/guides/jni/spec/invocation.html
+    
+    Unfortunately, unicode is unsupported by this interface on windows, since
+    windows uses wide-byte (16bit) character encoding.
+    See: https://stackoverflow.com/questions/20052455/jni-start-jvm-with-unicode-support
+
+    To resolve this issue we add the classpath after initialization since this
+    does support unicode via java strings.
+    """
+    for cp in _handleClassPath(classpath):
+        addClassPath(Path.cwd() / Path(cp).resolve())
+
 
 def initializeResources():
     global _JVM_started
@@ -347,7 +362,7 @@ def initializeResources():
     _jpype.JPypeContext = _jpype.JClass('org.jpype.JPypeContext').getInstance()
     _jpype.JPypeClassLoader = _jpype.JPypeContext.getClassLoader()
 
-    # Everything successed so started is now true.
+    # Everything succeeded so started is now true.
     _JVM_started = True
 
 

native/java/org/jpype/pkg/JPypePackageManager.java

@@ -19,6 +19,7 @@
 import java.net.URI;
 import java.net.URISyntaxException;
 import java.net.URL;
+import java.net.URLDecoder;
 import java.nio.file.FileSystem;
 import java.nio.file.FileSystemNotFoundException;
 import java.nio.file.FileSystems;
@@ -458,13 +459,33 @@ private static void collectContents(Map<String, URI> out, Path path2)
     }
   }
 
-  // Java 8 windows bug https://bugs.java.com/bugdatabase/view_bug.do?bug_id=8131067
   private static URI toURI(Path path)
   {
     URI uri = path.toUri();
-    if (uri.getScheme().equals("jar") && uri.toString().contains("%2520"))
-      uri = URI.create("jar:" + uri.getRawSchemeSpecificPart().replaceAll("%25", "%"));
-    return uri;
+
+    try {
+      // Java 8 bug https://bugs.java.com/bugdatabase/view_bug.do?bug_id=8131067
+      // Zip file system provider returns doubly % encoded URIs. We resolve this
+      // by re-encoding the URI after decoding it.
+      uri = new URI(
+        uri.getScheme(),
+        URLDecoder.decode(uri.getSchemeSpecificPart()),
+        uri.getFragment()
+      );
+
+    // `toASCIIString` ensures the URI is URL encoded with only ascii
+    // characters. This avoids issues in `sun.nio.fs.UnixUriUtils.fromUri` that
+    // naively uses `uri.getRawPath()` despite the possibility that it contains
+    // non-ascii characters that will cause errors. By using `toASCIIString` and
+    // re-wrapping it in a URI object we ensure that the URI is properly
+    // encoded. See: https://github.com/jpype-project/jpype/issues/1194
+      return new URI(uri.toASCIIString());
+    } catch (Exception e) {
+      // This exception *should* never occur as we are re-encoding a valid URI.
+      // Throwing a runtime exception avoids java exception handling boilerplate
+      // for a situation that *should* never occur.
+      throw new RuntimeException("Failed to encode URI: " + uri, e);
+    }
   }
 //</editor-fold>
 }

project/jars/unicode_à😎/README

@@ -0,0 +1,2 @@
+Test to verify that JAR's with unicode characters in the path are loaded correctly.
+The built jar should be placed in ``test/jar/unicode_à😎/sample_package.jar``.

project/jars/unicode_à😎/build.sh

@@ -0,0 +1,2 @@
+javac --release 8 -d classes src/org/jpype/sample_package/*.java
+jar --create --file sample_package.jar -C classes .

project/jars/unicode_à😎/src/org/jpype/sample_package/A.java

@@ -0,0 +1,5 @@
+package org.jpype.sample_package;
+
+public class A
+{
+}

project/jars/unicode_à😎/src/org/jpype/sample_package/B.java

@@ -0,0 +1,5 @@
+package org.jpype.sample_package;
+
+public class B
+{
+}

test/jpypetest/test_startup.py

@@ -17,7 +17,6 @@
 # *****************************************************************************
 import jpype
 import subrun
-import functools
 import os
 from pathlib import Path
 import unittest
@@ -146,3 +145,10 @@ def testPathTwice(self):
     def testBadKeyword(self):
         with self.assertRaises(TypeError):
             jpype.startJVM(invalid=True)  # type: ignore
+
+    def testNonASCIIPath(self):
+        """Test that paths with non-ASCII characters are handled correctly.
+        Regression test for https://github.com/jpype-project/jpype/issues/1194
+        """
+        jpype.startJVM(jvmpath=Path(self.jvmpath), classpath="test/jar/unicode_à😎/sample_package.jar")
+        assert dir(jpype.JPackage('org.jpype.sample_package')) == ['A', 'B']