Tony-X commented on code in PR #12688: URL: https://github.com/apache/lucene/pull/12688#discussion_r1381994228
########## lucene/sandbox/src/java/org/apache/lucene/sandbox/codecs/lucene90/randomaccess/TermsIndexBuilder.java: ########## @@ -0,0 +1,70 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.lucene.sandbox.codecs.lucene90.randomaccess; + +import java.io.IOException; +import java.util.Arrays; +import org.apache.lucene.util.BytesRef; +import org.apache.lucene.util.IntsRefBuilder; +import org.apache.lucene.util.fst.FST; +import org.apache.lucene.util.fst.FSTCompiler; +import org.apache.lucene.util.fst.PositiveIntOutputs; +import org.apache.lucene.util.fst.Util; + +/** + * Builds a term index for a given field. Logically this is a map: term -> (type, ord) where the + * ordinals are scoped to type (not global). + */ +final class TermsIndexBuilder { + private static long MAX_ORD = (1L << 60) - 1; + + private final long[] countPerType = new long[TermType.NUM_TOTAL_TYPES]; + private final FSTCompiler<Long> fstCompiler = + new FSTCompiler<>(FST.INPUT_TYPE.BYTE1, PositiveIntOutputs.getSingleton()); + + TermsIndexBuilder() { + Arrays.fill(countPerType, -1); + } + + public void addTerm(BytesRef term, TermType termType) throws IOException { + IntsRefBuilder scratchInts = new IntsRefBuilder(); + long ord = ++countPerType[termType.getId()]; + fstCompiler.add(Util.toIntsRef(term, scratchInts), encode(ord, termType)); + } + + public TermsIndex build() throws IOException { + return new TermsIndex(fstCompiler.compile()); + } + + private long encode(long ord, TermType termType) { + // use a single long to encode `ord` and `termType` + // also consider the special value of `PositiveIntOutputs.NO_OUTPUT == 0` + // so it looks like this |... ord ...| termType| ... hasOutput ...| + // where termType takes 3 bit and hasOutput takes the lowest bit. The rest is taken by ord + if (ord < 0) { + throw new IllegalArgumentException("can't encode negative ord"); + } + if (ord > MAX_ORD) { + throw new IllegalArgumentException( + "Input ord is too large for TermType: " Review Comment: Did you mean to spell out the input ord? ########## lucene/sandbox/src/java/org/apache/lucene/sandbox/codecs/lucene90/randomaccess/Lucene90RandomAccessDictionaryPostingsFormat.java: ########## @@ -0,0 +1,82 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.lucene.sandbox.codecs.lucene90.randomaccess; + +import java.io.IOException; +import org.apache.lucene.codecs.FieldsConsumer; +import org.apache.lucene.codecs.FieldsProducer; +import org.apache.lucene.codecs.PostingsFormat; +import org.apache.lucene.codecs.PostingsReaderBase; +import org.apache.lucene.codecs.PostingsWriterBase; +import org.apache.lucene.codecs.lucene90.Lucene90PostingsFormat; +import org.apache.lucene.codecs.lucene90.Lucene90PostingsReader; +import org.apache.lucene.codecs.lucene90.Lucene90PostingsWriter; +import org.apache.lucene.index.SegmentReadState; +import org.apache.lucene.index.SegmentWriteState; +import org.apache.lucene.util.IOUtils; + +/** + * Similar to {@link Lucene90PostingsFormat} but with a different term dictionary implementation. Review Comment: > BTW I think FST will pack better if you put the 3 "term type" bits at the end (lsb) of the long instead of msb, because then prefixes can be shared across all term types, instead of just within each term type. But, this may not be a win because it may take more vLong bytes to write each delta, not sure :) I believe this is what it is currently doing :) it will not create full 64bit long this way. To be specific -- * bit 0: indicate it is not `NO_OUTPUT` always set to 1 for valid (termType, ord) * bit [1, 4): termType * bit [4, 64): ord value bit index is from low to high, bit 0 is least significant bit. https://github.com/Tony-X/lucene/blob/ramdon_access_term_dict/lucene/sandbox/src/java/org/apache/lucene/sandbox/codecs/lucene90/randomaccess/TermsIndexBuilder.java#L68 [code](https://github.com/Tony-X/lucene/blob/ramdon_access_term_dict/lucene/sandbox/src/java/org/apache/lucene/sandbox/codecs/lucene90/randomaccess/TermsIndexBuilder.java#L68) ```java return (ord << 4) | ((long) termType.getId() << 1) | 1L; ``` ########## lucene/sandbox/src/java/org/apache/lucene/sandbox/codecs/lucene90/randomaccess/bitpacking/BitPacker.java: ########## Review Comment: Hey Knize, thanks for reviewing. This is not test-only. It is used by the `TestTermStateCodecImpl`. I'm in the process of building the real compact bit packer. > I'm also curious every time I see a new bit packer as we do this a lot throughout the code. Is there some reuse from another class impl maybe? PackedInts? DataOutput#writeVLong? Do we need it? I did search through the code base and couldn't find something I can use. The goal here is to pack a sequence of values that have different bitwidths . We can't use `PackedInts` as it requires values to have same bitwidth. We can't use VLong either since we aim to write fixed size record, so that we can do random access. More detailed discussion can be found in this email thread: https://lists.apache.org/list?d...@lucene.apache.org:lte=1M:packedInts -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
