kristoffSC commented on code in PR #21393: URL: https://github.com/apache/flink/pull/21393#discussion_r1068750626
########## flink-table/flink-table-code-splitter/src/main/java/org/apache/flink/table/codesplit/BlockStatementSplitter.java: ########## @@ -0,0 +1,417 @@ +/* + * 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.flink.table.codesplit; + +import org.apache.flink.annotation.Internal; +import org.apache.flink.table.codesplit.JavaParser.BlockStatementContext; +import org.apache.flink.table.codesplit.JavaParser.StatementContext; + +import org.antlr.v4.runtime.CharStreams; +import org.antlr.v4.runtime.CommonTokenStream; +import org.antlr.v4.runtime.ParserRuleContext; +import org.antlr.v4.runtime.TokenStreamRewriter; +import org.antlr.v4.runtime.atn.PredictionMode; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.Map.Entry; +import java.util.stream.Collectors; + +/** + * Extract statements from IFs, ELSEs and WILEs blocks making them smaller. + * + * <p><i>Before</i> + * + * <pre><code> + * while (counter > 0) { + * int localA = a + 1000; + * System.out.println(localA); + * if (a > 0) { + * b = a * 2; + * c = b * 2; + * System.out.println(b); + * } else { + * b = a * 3; + * System.out.println(b); + * } + * counter--; + * } + * + * </code></pre> + * + * <p><i>After</i> + * + * <pre><code> + * while (counter > 0) { + * myFun_whileBody0_0(int a); + * if (a > 0) { + * myFun_whileBody0_0_ifBody0(int a); + * } else { + * myFun_whileBody0_0_ifBody1(int a); + * } + * counter--; + * } + * </code></pre> + * + * <p>Where bodies of extracted "methods" are: + * + * <pre><code> + * myFun_whileBody0_0(int a) -> + * int localA = a + 1000; + * System.out.println(localA); + * </code></pre> + * + * <pre><code> + * myFun_whileBody0_0_ifBody0(int a) -> + * b = a * 2; + * c = b * 2; + * System.out.println(b); + * </code></pre> + * + * <pre><code> + * myFun_whileBody0_0_ifBody1(int a) -> + * b = a * 3; + * System.out.println(b); + * </code></pre> + */ +@Internal +public class BlockStatementSplitter { + + private final String code; + + private final List<BlockStatementSplitter> children = new ArrayList<>(); + + private final String parameters; + + private BlockStatementVisitor visitor; + + /** + * Initialize new BlockStatementSplitter. + * + * @param code a code block that should be rewritten. + * @param parameters parameters definition that should be used for extracted methods. + */ + public BlockStatementSplitter(String code, String parameters) { + this.code = code; + this.parameters = parameters; + } + + /** + * This method extracts statements from IFs, ELSE's and WHILE blocks from block code used during + * initialization of this object. Every entry of returned map can be seen as new method nam (map + * key) and method's body. The block names will be prefixed with provided context. + * + * @param context prefix for extracted blocks. + * @return a map of block name to block statements mappings. The key can be interpreted as name + * of extracted block/method and corresponding List represents individual statements (block' + * lines) for this block. + */ + public Map<String, List<String>> extractBlocks(String context) { + + this.visitor = new BlockStatementVisitor(code, context, parameters); + JavaParser javaParser = new JavaParser(visitor.tokenStream); + javaParser.getInterpreter().setPredictionMode(PredictionMode.SLL); + visitor.visitStatement(javaParser.statement()); + + Map<String, List<String>> extractedBlocks = new HashMap<>(); + for (BlockStatementVisitor child : visitor.children) { + int counter = 0; + for (ContextTextPair extractedBlock : child.extractedSingleBlocks) { + ParserRuleContext parserRuleContext = extractedBlock.parserRuleContext; + if (parserRuleContext instanceof BlockStatementContext) { + StatementContext statement = + ((BlockStatementContext) parserRuleContext).statement(); + + if (statement != null + && (statement.IF() != null + || statement.ELSE() != null + || statement.WHILE() != null)) { + + BlockStatementSplitter splitter = + new BlockStatementSplitter( + extractedBlock.ruleText, this.parameters); + Map<String, List<String>> rewrite = + splitter.extractBlocks(child.context + "_" + counter++); + this.children.add(splitter); + + mergeBlocks(rewrite, extractedBlocks); + } + } + } + } + + Map<String, List<String>> localBlocks = visitor.getAllBlocks(); + mergeBlocks(localBlocks, extractedBlocks); + + return extractedBlocks; + } + + /** + * Rewrite code block that was used for this object initialization. + * + * @return a map which key represent rewritten block name and value represents rewritten code + * block, including calls to extracted methods + */ + public Map<String, String> rewriteBlock() { + + for (BlockStatementSplitter child : children) { + child.visitor.rewrite(); + } + visitor.rewrite(); + + Map<String, String> rewriteBlocks = new HashMap<>(); + for (BlockStatementSplitter child : children) { + Map<String, String> childRewriteBlocks = child.rewriteBlock(); + for (Entry<String, String> entry : childRewriteBlocks.entrySet()) { + rewriteBlocks.merge( + entry.getKey(), + entry.getValue(), + (s, s2) -> { + throw new RuntimeException( + String.format( + "Override rewritten block for key %s. Blocks are %s -> %s", + entry.getKey(), s, s2)); + }); + } + } + + rewriteBlocks.put(visitor.context, visitor.rewriter.getText()); + return rewriteBlocks; + } + + private static void mergeBlocks( + Map<String, List<String>> mergeA, Map<String, List<String>> mergeB) { + + for (Entry<String, List<String>> entry : mergeA.entrySet()) { + mergeB.merge( + entry.getKey(), + entry.getValue(), + (aBlocks, bBlocks) -> { + List<String> merge = new ArrayList<>(aBlocks.size() + bBlocks.size()); + merge.addAll(aBlocks); + merge.addAll(bBlocks); + return merge; + }); + } + } + + private static class BlockStatementVisitor extends JavaParserBaseVisitor<Void> { + + private final List<BlockStatementVisitor> children = new ArrayList<>(); + + private final List<ContextTextPair> extractedSingleBlocks = new ArrayList<>(); + + private final String context; + + private final CommonTokenStream tokenStream; + + private final TokenStreamRewriter rewriter; + + private final String parameters; + + private int counter = 0; + + private int bscCounter = 0; + + private BlockStatementVisitor(String code, String context, String parameters) { + this.tokenStream = new CommonTokenStream(new JavaLexer(CharStreams.fromString(code))); + this.rewriter = new TokenStreamRewriter(tokenStream); + this.context = context; + this.parameters = parameters; + } + + private BlockStatementVisitor( + TokenStreamRewriter rewriter, String context, String parameters) { + this.tokenStream = null; + this.rewriter = rewriter; + this.context = context; + this.parameters = parameters; + } + + @Override + public Void visitStatement(StatementContext ctx) { + + if (ctx.getChildCount() == 0) { + return null; + } + + if (ctx.WHILE() != null) { + for (StatementContext statement : ctx.statement()) { + BlockStatementVisitor visitor = + new BlockStatementVisitor( + rewriter, context + "_whileBody" + counter++, parameters); + children.add(visitor); + visitor.visitStatement(statement); + } + } else if (ctx.IF() != null) { + for (StatementContext sc : ctx.statement()) { + BlockStatementVisitor visitor = + new BlockStatementVisitor( + rewriter, context + "_ifBody" + counter++, parameters); + children.add(visitor); + visitor.visitStatement(sc); + } + } else { Review Comment: Hi @tsreaper In fact we don't need `ctx.ELSE()` branch at all. I will try to explain why. The initial input statement for `BlockStatementSplitter` can be two things: - a statement that is represented like this: ``` if (a[2] == 0) { a[2] = 1; } else { a[2] = b[2]; } ``` - a block, that will be represented as this: ``` { if (a[2] == 0) { a[2] = 1; } else { a[2] = b[2]; } } ``` Mind that block has additional brackets `{ }` at the beginning and at the end. When `BlockStatementSplitter` is called from `BlockStatementRewriter` it will always be called by passing the statement, so without brackets `{ }`. The statement will never start from `else`, it will start from IF or WHILE (this is what we detect for now) or it will be what I call a "single line statement", meaning operation on a field, for example `a[2] = 1;`. Lets look at the example. The `BlockStatementRewriter` called `BlockStatementSplitter::visitStatement` with below statement: ``` if (a[0] == 0) { if (a[2] == 0) { a[2] = 1; } else { a[2] = b[2]; } } else { a[0] = b[0]; a[1] = b[1]; } ``` For this, the condition `if (ctx.WHILE() != null)` will be true. In this branch we will iterate through every `ctx.statement()` The first item would be: ``` { if (a[2] == 0) { a[2] = 1; } else { a[2] = b[2]; } } ``` So the `TRUE` branch of the first level IF/ELSE expression. We will create a new context to indicate that we will process an IF(possibly also ELSE) statement and we will create new `BlockStatementVisitor` and call visitStatement method to process that block, since it can contain more nested structures we would like to rewrite. Mind that now the input argument is a "block", it starts and ends with brackets `{ }` The second item in the iteration routine will be this: ``` { a[0] = b[0]; a[1] = b[1]; } ``` So the `FALSE` branch of the first level IF/ELSE expression. Same as for TRUE, we will create new context and `BlockStatementVisitor` to process that block since it can contain more nested structures we would like to rewrite. In this case the input also would be the "block" since it starts and ends with brackets `{ }`, same as for TRUE branch. In both new Visitors, conditions `ctx.WHILE() != null` and `ctx.IF() != null` will be false since now we have a "block", it starts from `{`. So we will land at `if (ctx.block() != null) ` line ~224 of `BlockStatementSplitter.java`. Since this is a block now, we need to iterate over all its statements hence `for (BlockStatementContext bsc : ctx.block().blockStatement())` For `TRUE` branch from the initial statement it will be only one item: ``` if (a[2] == 0) { a[2] = 1; } else { a[2] = b[2]; } ``` We will create new Visitor and process this statement further. For `FALSE` branch, from the initial statement it will contain two elements. Those will be "single line elements": - a[0] = b[0]; - a[1] = b[1]; Those will be added to `extractedSingleBlocks` blocks and extracted to separate method. I hope that I managed to explain the logic here, let me know please. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
