Describe the bug, including details regarding any error messages, version, and platform.
Format.asString builds the parameterised format strings with String.format("+w:%d", ...) and no explicit Locale. java.util.Formatter localises the digits of %d with Locale.getDefault(Locale.Category.FORMAT), so on a JVM whose default locale uses digits outside ASCII (ar-EG, fa-IR, bn-BD, mr-IN, 98 of the 1069 locales available on JDK 21) a FixedSizeList(8) is exported as +w:٨ instead of +w:8. The same applies to FixedSizeBinary and Decimal.
|
return String.format("d:%d,%d", type.getPrecision(), type.getScale()); |
|
return String.format( |
|
"d:%d,%d,%d", type.getPrecision(), type.getScale(), type.getBitWidth()); |
|
return String.format("+w:%d", type.getListSize()); |
|
return String.format("w:%d", type.getByteWidth()); |
Other implementations parse the parameter as an ASCII integer and reject the string. arrow-rs fails at https://github.com/apache/arrow-rs/blob/58.0.0/arrow-schema/src/ffi.rs#L494-L503 with C Data interface error: The FixedSizeList type requires an integer parameter representing number of elements per list. The Java importer reads the parameter with Integer.parseInt (https://github.com/apache/arrow-java/blob/v19.0.0/c/src/main/java/org/apache/arrow/c/Format.java#L352-L354), which accepts Unicode digits, so a Java to Java round trip does not expose the problem.
Reproduction (arrow-java 19.0.0 and 18.1.0, Temurin 21.0.6, run with --add-opens=java.base/java.nio=ALL-UNNAMED):
import java.nio.charset.StandardCharsets;
import java.util.HexFormat;
import java.util.List;
import java.util.Locale;
import org.apache.arrow.c.ArrowSchema;
import org.apache.arrow.c.Data;
import org.apache.arrow.c.NativeUtil;
import org.apache.arrow.memory.BufferAllocator;
import org.apache.arrow.memory.RootAllocator;
import org.apache.arrow.vector.types.FloatingPointPrecision;
import org.apache.arrow.vector.types.pojo.ArrowType;
import org.apache.arrow.vector.types.pojo.Field;
import org.apache.arrow.vector.types.pojo.FieldType;
public class FormatLocale {
public static void main(String[] args) {
Locale.setDefault(Locale.forLanguageTag("ar-EG"));
Field item = new Field("item",
FieldType.nullable(new ArrowType.FloatingPoint(FloatingPointPrecision.SINGLE)), null);
Field vec = new Field("vec", FieldType.nullable(new ArrowType.FixedSizeList(8)), List.of(item));
try (BufferAllocator allocator = new RootAllocator()) {
ArrowSchema schema = ArrowSchema.allocateNew(allocator);
Data.exportField(allocator, vec, null, schema);
String format = NativeUtil.toJavaString(schema.snapshot().format);
System.out.println(format + " "
+ HexFormat.ofDelimiter(" ").formatHex(format.getBytes(StandardCharsets.UTF_8)));
System.out.println("Java import: " + Data.importField(allocator, schema, null).getType());
}
}
}
+w:٨ 2b 77 3a d9 a8
Java import: FixedSizeList(8)
With Locale.US the first line is +w:8 2b 77 3a 38.
Describe the bug, including details regarding any error messages, version, and platform.
Format.asStringbuilds the parameterised format strings withString.format("+w:%d", ...)and no explicitLocale.java.util.Formatterlocalises the digits of%dwithLocale.getDefault(Locale.Category.FORMAT), so on a JVM whose default locale uses digits outside ASCII (ar-EG,fa-IR,bn-BD,mr-IN, 98 of the 1069 locales available on JDK 21) aFixedSizeList(8)is exported as+w:٨instead of+w:8. The same applies toFixedSizeBinaryandDecimal.arrow-java/c/src/main/java/org/apache/arrow/c/Format.java
Line 66 in b410fb2
arrow-java/c/src/main/java/org/apache/arrow/c/Format.java
Lines 68 to 69 in b410fb2
arrow-java/c/src/main/java/org/apache/arrow/c/Format.java
Line 96 in b410fb2
arrow-java/c/src/main/java/org/apache/arrow/c/Format.java
Line 91 in b410fb2
Other implementations parse the parameter as an ASCII integer and reject the string. arrow-rs fails at https://github.com/apache/arrow-rs/blob/58.0.0/arrow-schema/src/ffi.rs#L494-L503 with
C Data interface error: The FixedSizeList type requires an integer parameter representing number of elements per list. The Java importer reads the parameter withInteger.parseInt(https://github.com/apache/arrow-java/blob/v19.0.0/c/src/main/java/org/apache/arrow/c/Format.java#L352-L354), which accepts Unicode digits, so a Java to Java round trip does not expose the problem.Reproduction (arrow-java 19.0.0 and 18.1.0, Temurin 21.0.6, run with
--add-opens=java.base/java.nio=ALL-UNNAMED):With
Locale.USthe first line is+w:8 2b 77 3a 38.