phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-4237) Allow sorting on (Java) collation keys for non-English locales
Date Fri, 13 Oct 2017 07:39:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-4237?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16203159#comment-16203159
] 

ASF GitHub Bot commented on PHOENIX-4237:
-----------------------------------------

Github user snakhoda-sfdc commented on a diff in the pull request:

    https://github.com/apache/phoenix/pull/275#discussion_r144483837
  
    --- Diff: phoenix-core/src/main/java/org/apache/phoenix/expression/function/CollationKeyFunction.java
---
    @@ -0,0 +1,233 @@
    +package org.apache.phoenix.expression.function;
    +
    +import java.sql.SQLException;
    +import java.text.Collator;
    +import java.util.Arrays;
    +import java.util.List;
    +import java.util.Locale;
    +
    +import org.apache.commons.lang.BooleanUtils;
    +import org.apache.commons.logging.Log;
    +import org.apache.commons.logging.LogFactory;
    +import org.apache.hadoop.hbase.io.ImmutableBytesWritable;
    +import org.apache.phoenix.expression.Expression;
    +import org.apache.phoenix.parse.FunctionParseNode;
    +import org.apache.phoenix.schema.tuple.Tuple;
    +import org.apache.phoenix.schema.types.PBoolean;
    +import org.apache.phoenix.schema.types.PDataType;
    +import org.apache.phoenix.schema.types.PInteger;
    +import org.apache.phoenix.schema.types.PIntegerArray;
    +import org.apache.phoenix.schema.types.PUnsignedIntArray;
    +import org.apache.phoenix.schema.types.PVarbinary;
    +import org.apache.phoenix.schema.types.PVarchar;
    +import org.apache.phoenix.schema.types.PhoenixArray;
    +
    +import com.force.db.i18n.LinguisticSort;
    +import com.force.i18n.LocaleUtils;
    +
    +import com.ibm.icu.impl.jdkadapter.CollatorICU;
    +import com.ibm.icu.util.ULocale;
    +
    +/**
    + * A Phoenix Function that calculates a collation key for an input string based
    + * on a caller-provided locale and collator strength and decomposition settings.
    + * 
    + * It uses the open-source grammaticus and i18n packages to obtain the collators
    + * it needs.
    + * 
    + * @author snakhoda
    + *
    + */
    +@FunctionParseNode.BuiltInFunction(name = CollationKeyFunction.NAME, args = {
    +		// input string
    +		@FunctionParseNode.Argument(allowedTypes = { PVarchar.class }),
    +		// ISO Code for Locale
    +		@FunctionParseNode.Argument(allowedTypes = { PVarchar.class }, isConstant = true),
    +		// whether to use special upper case collator
    +		@FunctionParseNode.Argument(allowedTypes = { PBoolean.class }, defaultValue = "false",
isConstant = true),
    +		// collator strength
    +		@FunctionParseNode.Argument(allowedTypes = { PInteger.class }, defaultValue = "null",
isConstant = true),
    +		// collator decomposition
    +		@FunctionParseNode.Argument(allowedTypes = { PInteger.class }, defaultValue = "null",
isConstant = true) })
    +public class CollationKeyFunction extends ScalarFunction {
    +
    +	private static final Log LOG = LogFactory.getLog(CollationKeyFunction.class);
    +
    +	public static final String NAME = "COLLKEY";
    +
    +	public CollationKeyFunction() {
    +	}
    +
    +	public CollationKeyFunction(List<Expression> children) throws SQLException {
    +		super(children);
    +	}
    +
    +	@Override
    +	public boolean evaluate(Tuple tuple, ImmutableBytesWritable ptr) {
    +		try {
    +			String inputValue = getInputValue(tuple, ptr);
    +			String localeISOCode = getLocaleISOCode(tuple, ptr);
    +			Boolean useSpecialUpperCaseCollator = getUseSpecialUpperCaseCollator(tuple, ptr);
    +			Integer collatorStrength = getCollatorStrength(tuple, ptr);
    +			Integer collatorDecomposition = getCollatorDecomposition(tuple, ptr);
    +
    +			Locale locale = LocaleUtils.get().getLocaleByIsoCode(localeISOCode);
    +			
    +			if(LOG.isDebugEnabled()) {
    +				LOG.debug(String.format("Locale: " + locale.toLanguageTag()));
    +			}
    +			
    +			LinguisticSort linguisticSort = LinguisticSort.get(locale);
    +
    +			Collator collator = BooleanUtils.isTrue(useSpecialUpperCaseCollator)
    +					? linguisticSort.getUpperCaseCollator(false) : linguisticSort.getCollator();
    +
    +			if (collatorStrength != null) {
    +				collator.setStrength(collatorStrength);
    +			}
    +
    +			if (collatorDecomposition != null) {
    +				collator.setDecomposition(collatorDecomposition);
    +			}
    +
    +			if(LOG.isDebugEnabled()) {
    +				LOG.debug(String.format("Collator: [strength: %d, decomposition: %d], Special-Upper-Case:
%s",
    +					collator.getStrength(), collator.getDecomposition(), BooleanUtils.isTrue(useSpecialUpperCaseCollator)));
    +			}
    +			
    +			byte[] collationKeyByteArray = collator.getCollationKey(inputValue).toByteArray();
    +
    +			if(LOG.isDebugEnabled()) {
    +				LOG.debug("Collation key bytes:" + Arrays.toString(collationKeyByteArray));
    +			}
    +			
    +			// byte is signed in Java, but we need unsigned values for comparison
    +			// https://www.programcreek.com/java-api-examples/index.php?api=java.text.CollationKey
    +			// Byte.toUnsignedInt will convert a byte value between [-128,127] to an int value
    +			// between [0,255]
    --- End diff --
    
    I started out with a straight byte array (VARBINARY), but realized the issue is related
to 2s complement binary representation.
    
    Here's what I was able to gather:
    
    The collator's collation key produces a byte array that holds values that are supposed
to be treated as unsigned. However, since byte is a signed (2s complement) datatype in Java,
doing a Java comparison on them with regular operators (which is presumably what Phoenix does)
produces incorrect results.
    
    For example, 00000001 represents 1 and 11111111 represents -1 in 2s complement. If we
simply compare those bytes using standard operators then 11111111 is less than 00000001. But
a collator key byte array that returns these two bytes would like them to be treated as "1"
and "255" respectively and would need 11111111 to be greater than 00000001.
    
    Getting this effect out of a byte array basically means taking each byte and widening
it (to an int) which results in the lower order bytes representing the unsigned positive number
that was intended by the collation key.
    
    See this example usage of toByteArray, which does the same thing: http://icu-project.org/apiref/icu4j59m1/com/ibm/icu/text/CollationKey.html#toByteArray--
    
    BTW, Byte.toUnsignedInt in Java 8 is basically a bitwise AND with 0xFF.


> Allow sorting on (Java) collation keys for non-English locales
> --------------------------------------------------------------
>
>                 Key: PHOENIX-4237
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4237
>             Project: Phoenix
>          Issue Type: Improvement
>            Reporter: Shehzaad Nakhoda
>             Fix For: 4.12.0
>
>
> Strings stored via Phoenix can be composed from a subset of the entire set of Unicode
characters. The natural sort order for strings for different languages often differs from
the order dictated by the binary representation of the characters of these strings. Java provides
the idea of a Collator which given an input string and a (language) locale can generate a
Collation Key which can then be used to compare strings in that natural order.
> Salesforce has recently open-sourced grammaticus. IBM has open-sourced ICU4J some time
ago. These technologies can be combined to provide a robust new Phoenix function that can
be used in an ORDER BY clause to sort strings according to the user's locale.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message