UNPKG

eslint

Version:

An AST-based pattern checker for JavaScript.

1,621 lines (1,436 loc) 71.1 kB
/** * @fileoverview A class to manage state of generating a code path. * @author Toru Nagashima */ "use strict"; //------------------------------------------------------------------------------ // Requirements //------------------------------------------------------------------------------ const CodePathSegment = require("./code-path-segment"), ForkContext = require("./fork-context"); //----------------------------------------------------------------------------- // Contexts //----------------------------------------------------------------------------- /** * Represents the context in which a `break` statement can be used. * * A `break` statement without a label is only valid in a few places in * JavaScript: any type of loop or a `switch` statement. Otherwise, `break` * without a label causes a syntax error. For these contexts, `breakable` is * set to `true` to indicate that a `break` without a label is valid. * * However, a `break` statement with a label is also valid inside of a labeled * statement. For example, this is valid: * * a : { * break a; * } * * The `breakable` property is set false for labeled statements to indicate * that `break` without a label is invalid. */ class BreakContext { /** * Creates a new instance. * @param {BreakContext} upperContext The previous `BreakContext`. * @param {boolean} breakable Indicates if we are inside a statement where * `break` without a label will exit the statement. * @param {string|null} label The label for the statement. * @param {ForkContext} forkContext The current fork context. */ constructor(upperContext, breakable, label, forkContext) { /** * The previous `BreakContext` * @type {BreakContext} */ this.upper = upperContext; /** * Indicates if we are inside a statement where `break` without a label * will exit the statement. * @type {boolean} */ this.breakable = breakable; /** * The label associated with the statement. * @type {string|null} */ this.label = label; /** * The fork context for the `break`. * @type {ForkContext} */ this.brokenForkContext = ForkContext.newEmpty(forkContext); } } /** * Represents the context for `ChainExpression` nodes. */ class ChainContext { /** * Creates a new instance. * @param {ChainContext} upperContext The previous `ChainContext`. */ constructor(upperContext) { /** * The previous `ChainContext` * @type {ChainContext} */ this.upper = upperContext; /** * The number of choice contexts inside of the `ChainContext`. * @type {number} */ this.choiceContextCount = 0; } } /** * Represents a choice in the code path. * * Choices are created by logical operators such as `&&`, loops, conditionals, * and `if` statements. This is the point at which the code path has a choice of * which direction to go. * * The result of a choice might be in the left (test) expression of another choice, * and in that case, may create a new fork. For example, `a || b` is a choice * but does not create a new fork because the result of the expression is * not used as the test expression in another expression. In this case, * `isForkingAsResult` is false. In the expression `a || b || c`, the `a || b` * expression appears as the test expression for `|| c`, so the * result of `a || b` creates a fork because execution may or may not * continue to `|| c`. `isForkingAsResult` for `a || b` in this case is true * while `isForkingAsResult` for `|| c` is false. (`isForkingAsResult` is always * false for `if` statements, conditional expressions, and loops.) * * All of the choices except one (`??`) operate on a true/false fork, meaning if * true go one way and if false go the other (tracked by `trueForkContext` and * `falseForkContext`). The `??` operator doesn't operate on true/false because * the left expression is evaluated to be nullish or not, so only if nullish do * we fork to the right expression (tracked by `nullishForkContext`). */ class ChoiceContext { /** * Creates a new instance. * @param {ChoiceContext} upperContext The previous `ChoiceContext`. * @param {string} kind The kind of choice. If it's a logical or assignment expression, this * is `"&&"` or `"||"` or `"??"`; if it's an `if` statement or * conditional expression, this is `"test"`; otherwise, this is `"loop"`. * @param {boolean} isForkingAsResult Indicates if the result of the choice * creates a fork. * @param {ForkContext} forkContext The containing `ForkContext`. */ constructor(upperContext, kind, isForkingAsResult, forkContext) { /** * The previous `ChoiceContext` * @type {ChoiceContext} */ this.upper = upperContext; /** * The kind of choice. If it's a logical or assignment expression, this * is `"&&"` or `"||"` or `"??"`; if it's an `if` statement or * conditional expression, this is `"test"`; otherwise, this is `"loop"`. * @type {string} */ this.kind = kind; /** * Indicates if the result of the choice forks the code path. * @type {boolean} */ this.isForkingAsResult = isForkingAsResult; /** * The fork context for the `true` path of the choice. * @type {ForkContext} */ this.trueForkContext = ForkContext.newEmpty(forkContext); /** * The fork context for the `false` path of the choice. * @type {ForkContext} */ this.falseForkContext = ForkContext.newEmpty(forkContext); /** * The fork context for when the choice result is `null` or `undefined`. * @type {ForkContext} */ this.nullishForkContext = ForkContext.newEmpty(forkContext); /** * Indicates if any of `trueForkContext`, `falseForkContext`, or * `nullishForkContext` have been updated with segments from a child context. * @type {boolean} */ this.processed = false; } } /** * Base class for all loop contexts. */ class LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string} type The AST node's `type` for the loop. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. */ constructor(upperContext, type, label, breakContext) { /** * The previous `LoopContext`. * @type {LoopContext} */ this.upper = upperContext; /** * The AST node's `type` for the loop. * @type {string} */ this.type = type; /** * The label for the loop from an enclosing `LabeledStatement`. * @type {string|null} */ this.label = label; /** * The fork context for when `break` is encountered. * @type {ForkContext} */ this.brokenForkContext = breakContext.brokenForkContext; } } /** * Represents the context for a `while` loop. */ class WhileLoopContext extends LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. */ constructor(upperContext, label, breakContext) { super(upperContext, "WhileStatement", label, breakContext); /** * The hardcoded literal boolean test condition for * the loop. Used to catch infinite or skipped loops. * @type {boolean|undefined} */ this.test = void 0; /** * The segments representing the test condition where `continue` will * jump to. The test condition will typically have just one segment but * it's possible for there to be more than one. * @type {Array<CodePathSegment>|null} */ this.continueDestSegments = null; } } /** * Represents the context for a `do-while` loop. */ class DoWhileLoopContext extends LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. * @param {ForkContext} forkContext The enclosing fork context. */ constructor(upperContext, label, breakContext, forkContext) { super(upperContext, "DoWhileStatement", label, breakContext); /** * The hardcoded literal boolean test condition for * the loop. Used to catch infinite or skipped loops. * @type {boolean|undefined} */ this.test = void 0; /** * The segments at the start of the loop body. This is the only loop * where the test comes at the end, so the first iteration always * happens and we need a reference to the first statements. * @type {Array<CodePathSegment>|null} */ this.entrySegments = null; /** * The fork context to follow when a `continue` is found. * @type {ForkContext} */ this.continueForkContext = ForkContext.newEmpty(forkContext); } } /** * Represents the context for a `for` loop. */ class ForLoopContext extends LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. */ constructor(upperContext, label, breakContext) { super(upperContext, "ForStatement", label, breakContext); /** * The hardcoded literal boolean test condition for * the loop. Used to catch infinite or skipped loops. * @type {boolean|undefined} */ this.test = void 0; /** * The end of the init expression. This may change during the lifetime * of the instance as we traverse the loop because some loops don't have * an init expression. * @type {Array<CodePathSegment>|null} */ this.endOfInitSegments = null; /** * The start of the test expression. This may change during the lifetime * of the instance as we traverse the loop because some loops don't have * a test expression. * @type {Array<CodePathSegment>|null} */ this.testSegments = null; /** * The end of the test expression. This may change during the lifetime * of the instance as we traverse the loop because some loops don't have * a test expression. * @type {Array<CodePathSegment>|null} */ this.endOfTestSegments = null; /** * The start of the update expression. This may change during the lifetime * of the instance as we traverse the loop because some loops don't have * an update expression. * @type {Array<CodePathSegment>|null} */ this.updateSegments = null; /** * The end of the update expresion. This may change during the lifetime * of the instance as we traverse the loop because some loops don't have * an update expression. * @type {Array<CodePathSegment>|null} */ this.endOfUpdateSegments = null; /** * The segments representing the test condition where `continue` will * jump to. The test condition will typically have just one segment but * it's possible for there to be more than one. This may change during the * lifetime of the instance as we traverse the loop because some loops * don't have an update expression. When there is an update expression, this * will end up pointing to that expression; otherwise it will end up pointing * to the test expression. * @type {Array<CodePathSegment>|null} */ this.continueDestSegments = null; } } /** * Represents the context for a `for-in` loop. * * Terminology: * - "left" means the part of the loop to the left of the `in` keyword. For * example, in `for (var x in y)`, the left is `var x`. * - "right" means the part of the loop to the right of the `in` keyword. For * example, in `for (var x in y)`, the right is `y`. */ class ForInLoopContext extends LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. */ constructor(upperContext, label, breakContext) { super(upperContext, "ForInStatement", label, breakContext); /** * The segments that came immediately before the start of the loop. * This allows you to traverse backwards out of the loop into the * surrounding code. This is necessary to evaluate the right expression * correctly, as it must be evaluated in the same way as the left * expression, but the pointer to these segments would otherwise be * lost if not stored on the instance. Once the right expression has * been evaluated, this property is no longer used. * @type {Array<CodePathSegment>|null} */ this.prevSegments = null; /** * Segments representing the start of everything to the left of the * `in` keyword. This can be used to move forward towards * `endOfLeftSegments`. `leftSegments` and `endOfLeftSegments` are * effectively the head and tail of a doubly-linked list. * @type {Array<CodePathSegment>|null} */ this.leftSegments = null; /** * Segments representing the end of everything to the left of the * `in` keyword. This can be used to move backward towards `leftSegments`. * `leftSegments` and `endOfLeftSegments` are effectively the head * and tail of a doubly-linked list. * @type {Array<CodePathSegment>|null} */ this.endOfLeftSegments = null; /** * The segments representing the left expression where `continue` will * jump to. In `for-in` loops, `continue` must always re-execute the * left expression each time through the loop. This contains the same * segments as `leftSegments`, but is duplicated here so each loop * context has the same property pointing to where `continue` should * end up. * @type {Array<CodePathSegment>|null} */ this.continueDestSegments = null; } } /** * Represents the context for a `for-of` loop. */ class ForOfLoopContext extends LoopContextBase { /** * Creates a new instance. * @param {LoopContext|null} upperContext The previous `LoopContext`. * @param {string|null} label The label for the loop from an enclosing `LabeledStatement`. * @param {BreakContext} breakContext The context for breaking the loop. */ constructor(upperContext, label, breakContext) { super(upperContext, "ForOfStatement", label, breakContext); /** * The segments that came immediately before the start of the loop. * This allows you to traverse backwards out of the loop into the * surrounding code. This is necessary to evaluate the right expression * correctly, as it must be evaluated in the same way as the left * expression, but the pointer to these segments would otherwise be * lost if not stored on the instance. Once the right expression has * been evaluated, this property is no longer used. * @type {Array<CodePathSegment>|null} */ this.prevSegments = null; /** * Segments representing the start of everything to the left of the * `of` keyword. This can be used to move forward towards * `endOfLeftSegments`. `leftSegments` and `endOfLeftSegments` are * effectively the head and tail of a doubly-linked list. * @type {Array<CodePathSegment>|null} */ this.leftSegments = null; /** * Segments representing the end of everything to the left of the * `of` keyword. This can be used to move backward towards `leftSegments`. * `leftSegments` and `endOfLeftSegments` are effectively the head * and tail of a doubly-linked list. * @type {Array<CodePathSegment>|null} */ this.endOfLeftSegments = null; /** * The segments representing the left expression where `continue` will * jump to. In `for-in` loops, `continue` must always re-execute the * left expression each time through the loop. This contains the same * segments as `leftSegments`, but is duplicated here so each loop * context has the same property pointing to where `continue` should * end up. * @type {Array<CodePathSegment>|null} */ this.continueDestSegments = null; } } /** * Represents the context for any loop. * @typedef {WhileLoopContext|DoWhileLoopContext|ForLoopContext|ForInLoopContext|ForOfLoopContext} LoopContext */ /** * Represents the context for a `switch` statement. */ class SwitchContext { /** * Creates a new instance. * @param {SwitchContext} upperContext The previous context. * @param {boolean} hasCase Indicates if there is at least one `case` statement. * `default` doesn't count. */ constructor(upperContext, hasCase) { /** * The previous context. * @type {SwitchContext} */ this.upper = upperContext; /** * Indicates if there is at least one `case` statement. `default` doesn't count. * @type {boolean} */ this.hasCase = hasCase; /** * The `default` keyword. * @type {Array<CodePathSegment>|null} */ this.defaultSegments = null; /** * The default case body starting segments. * @type {Array<CodePathSegment>|null} */ this.defaultBodySegments = null; /** * Indicates if a `default` case and is empty exists. * @type {boolean} */ this.foundEmptyDefault = false; /** * Indicates that a `default` exists and is the last case. * @type {boolean} */ this.lastIsDefault = false; /** * The number of fork contexts created. This is equivalent to the * number of `case` statements plus a `default` statement (if present). * @type {number} */ this.forkCount = 0; } } /** * Represents the context for a `try` statement. */ class TryContext { /** * Creates a new instance. * @param {TryContext} upperContext The previous context. * @param {boolean} hasFinalizer Indicates if the `try` statement has a * `finally` block. * @param {ForkContext} forkContext The enclosing fork context. */ constructor(upperContext, hasFinalizer, forkContext) { /** * The previous context. * @type {TryContext} */ this.upper = upperContext; /** * Indicates if the `try` statement has a `finally` block. * @type {boolean} */ this.hasFinalizer = hasFinalizer; /** * Tracks the traversal position inside of the `try` statement. This is * used to help determine the context necessary to create paths because * a `try` statement may or may not have `catch` or `finally` blocks, * and code paths behave differently in those blocks. * @type {"try"|"catch"|"finally"} */ this.position = "try"; /** * If the `try` statement has a `finally` block, this affects how a * `return` statement behaves in the `try` block. Without `finally`, * `return` behaves as usual and doesn't require a fork; with `finally`, * `return` forks into the `finally` block, so we need a fork context * to track it. * @type {ForkContext|null} */ this.returnedForkContext = hasFinalizer ? ForkContext.newEmpty(forkContext) : null; /** * When a `throw` occurs inside of a `try` block, the code path forks * into the `catch` or `finally` blocks, and this fork context tracks * that path. * @type {ForkContext} */ this.thrownForkContext = ForkContext.newEmpty(forkContext); /** * Indicates if the last segment in the `try` block is reachable. * @type {boolean} */ this.lastOfTryIsReachable = false; /** * Indicates if the last segment in the `catch` block is reachable. * @type {boolean} */ this.lastOfCatchIsReachable = false; } } //------------------------------------------------------------------------------ // Helpers //------------------------------------------------------------------------------ /** * Adds given segments into the `dest` array. * If the `others` array does not include the given segments, adds to the `all` * array as well. * * This adds only reachable and used segments. * @param {CodePathSegment[]} dest A destination array (`returnedSegments` or `thrownSegments`). * @param {CodePathSegment[]} others Another destination array (`returnedSegments` or `thrownSegments`). * @param {CodePathSegment[]} all The unified destination array (`finalSegments`). * @param {CodePathSegment[]} segments Segments to add. * @returns {void} */ function addToReturnedOrThrown(dest, others, all, segments) { for (let i = 0; i < segments.length; ++i) { const segment = segments[i]; dest.push(segment); if (!others.includes(segment)) { all.push(segment); } } } /** * Gets a loop context for a `continue` statement based on a given label. * @param {CodePathState} state The state to search within. * @param {string|null} label The label of a `continue` statement. * @returns {LoopContext} A loop-context for a `continue` statement. */ function getContinueContext(state, label) { if (!label) { return state.loopContext; } let context = state.loopContext; while (context) { if (context.label === label) { return context; } context = context.upper; } /* c8 ignore next */ return null; } /** * Gets a context for a `break` statement. * @param {CodePathState} state The state to search within. * @param {string|null} label The label of a `break` statement. * @returns {BreakContext} A context for a `break` statement. */ function getBreakContext(state, label) { let context = state.breakContext; while (context) { if (label ? context.label === label : context.breakable) { return context; } context = context.upper; } /* c8 ignore next */ return null; } /** * Gets a context for a `return` statement. There is just one special case: * if there is a `try` statement with a `finally` block, because that alters * how `return` behaves; otherwise, this just passes through the given state. * @param {CodePathState} state The state to search within * @returns {TryContext|CodePathState} A context for a `return` statement. */ function getReturnContext(state) { let context = state.tryContext; while (context) { if (context.hasFinalizer && context.position !== "finally") { return context; } context = context.upper; } return state; } /** * Gets a context for a `throw` statement. There is just one special case: * if there is a `try` statement with a `finally` block and we are inside of * a `catch` because that changes how `throw` behaves; otherwise, this just * passes through the given state. * @param {CodePathState} state The state to search within. * @returns {TryContext|CodePathState} A context for a `throw` statement. */ function getThrowContext(state) { let context = state.tryContext; while (context) { if ( context.position === "try" || (context.hasFinalizer && context.position === "catch") ) { return context; } context = context.upper; } return state; } /** * Removes a given value from a given array. * @param {any[]} elements An array to remove the specific element. * @param {any} value The value to be removed. * @returns {void} */ function removeFromArray(elements, value) { elements.splice(elements.indexOf(value), 1); } /** * Disconnect given segments. * * This is used in a process for switch statements. * If there is the "default" chunk before other cases, the order is different * between node's and running's. * @param {CodePathSegment[]} prevSegments Forward segments to disconnect. * @param {CodePathSegment[]} nextSegments Backward segments to disconnect. * @returns {void} */ function disconnectSegments(prevSegments, nextSegments) { for (let i = 0; i < prevSegments.length; ++i) { const prevSegment = prevSegments[i]; const nextSegment = nextSegments[i]; removeFromArray(prevSegment.nextSegments, nextSegment); removeFromArray(prevSegment.allNextSegments, nextSegment); removeFromArray(nextSegment.prevSegments, prevSegment); removeFromArray(nextSegment.allPrevSegments, prevSegment); } } /** * Creates looping path between two arrays of segments, ensuring that there are * paths going between matching segments in the arrays. * @param {CodePathState} state The state to operate on. * @param {CodePathSegment[]} unflattenedFromSegments Segments which are source. * @param {CodePathSegment[]} unflattenedToSegments Segments which are destination. * @returns {void} */ function makeLooped(state, unflattenedFromSegments, unflattenedToSegments) { const fromSegments = CodePathSegment.flattenUnusedSegments( unflattenedFromSegments, ); const toSegments = CodePathSegment.flattenUnusedSegments( unflattenedToSegments, ); const end = Math.min(fromSegments.length, toSegments.length); /* * This loop effectively updates a doubly-linked list between two collections * of segments making sure that segments in the same array indices are * combined to create a path. */ for (let i = 0; i < end; ++i) { // get the segments in matching array indices const fromSegment = fromSegments[i]; const toSegment = toSegments[i]; /* * If the destination segment is reachable, then create a path from the * source segment to the destination segment. */ if (toSegment.reachable) { fromSegment.nextSegments.push(toSegment); } /* * If the source segment is reachable, then create a path from the * destination segment back to the source segment. */ if (fromSegment.reachable) { toSegment.prevSegments.push(fromSegment); } /* * Also update the arrays that don't care if the segments are reachable * or not. This should always happen regardless of anything else. */ fromSegment.allNextSegments.push(toSegment); toSegment.allPrevSegments.push(fromSegment); /* * If the destination segment has at least two previous segments in its * path then that means there was one previous segment before this iteration * of the loop was executed. So, we need to mark the source segment as * looped. */ if (toSegment.allPrevSegments.length >= 2) { CodePathSegment.markPrevSegmentAsLooped(toSegment, fromSegment); } // let the code path analyzer know that there's been a loop created state.notifyLooped(fromSegment, toSegment); } } /** * Finalizes segments of `test` chunk of a ForStatement. * * - Adds `false` paths to paths which are leaving from the loop. * - Sets `true` paths to paths which go to the body. * @param {LoopContext} context A loop context to modify. * @param {ChoiceContext} choiceContext A choice context of this loop. * @param {CodePathSegment[]} head The current head paths. * @returns {void} */ function finalizeTestSegmentsOfFor(context, choiceContext, head) { /* * If this choice context doesn't already contain paths from a * child context, then add the current head to each potential path. */ if (!choiceContext.processed) { choiceContext.trueForkContext.add(head); choiceContext.falseForkContext.add(head); choiceContext.nullishForkContext.add(head); } /* * If the test condition isn't a hardcoded truthy value, then `break` * must follow the same path as if the test condition is false. To represent * that, we append the path for when the loop test is false (represented by * `falseForkContext`) to the `brokenForkContext`. */ if (context.test !== true) { context.brokenForkContext.addAll(choiceContext.falseForkContext); } context.endOfTestSegments = choiceContext.trueForkContext.makeNext(0, -1); } //------------------------------------------------------------------------------ // Public Interface //------------------------------------------------------------------------------ /** * A class which manages state to analyze code paths. */ class CodePathState { /** * Creates a new instance. * @param {IdGenerator} idGenerator An id generator to generate id for code * path segments. * @param {Function} onLooped A callback function to notify looping. */ constructor(idGenerator, onLooped) { /** * The ID generator to use when creating new segments. * @type {IdGenerator} */ this.idGenerator = idGenerator; /** * A callback function to call when there is a loop. * @type {Function} */ this.notifyLooped = onLooped; /** * The root fork context for this state. * @type {ForkContext} */ this.forkContext = ForkContext.newRoot(idGenerator); /** * Context for logical expressions, conditional expressions, `if` statements, * and loops. * @type {ChoiceContext} */ this.choiceContext = null; /** * Context for `switch` statements. * @type {SwitchContext} */ this.switchContext = null; /** * Context for `try` statements. * @type {TryContext} */ this.tryContext = null; /** * Context for loop statements. * @type {LoopContext} */ this.loopContext = null; /** * Context for `break` statements. * @type {BreakContext} */ this.breakContext = null; /** * Context for `ChainExpression` nodes. * @type {ChainContext} */ this.chainContext = null; /** * An array that tracks the current segments in the state. The array * starts empty and segments are added with each `onCodePathSegmentStart` * event and removed with each `onCodePathSegmentEnd` event. Effectively, * this is tracking the code path segment traversal as the state is * modified. * @type {Array<CodePathSegment>} */ this.currentSegments = []; /** * Tracks the starting segment for this path. This value never changes. * @type {CodePathSegment} */ this.initialSegment = this.forkContext.head[0]; /** * The final segments of the code path which are either `return` or `throw`. * This is a union of the segments in `returnedForkContext` and `thrownForkContext`. * @type {Array<CodePathSegment>} */ this.finalSegments = []; /** * The final segments of the code path which are `return`. These * segments are also contained in `finalSegments`. * @type {Array<CodePathSegment>} */ this.returnedForkContext = []; /** * The final segments of the code path which are `throw`. These * segments are also contained in `finalSegments`. * @type {Array<CodePathSegment>} */ this.thrownForkContext = []; /* * We add an `add` method so that these look more like fork contexts and * can be used interchangeably when a fork context is needed to add more * segments to a path. * * Ultimately, we want anything added to `returned` or `thrown` to also * be added to `final`. We only add reachable and used segments to these * arrays. */ const final = this.finalSegments; const returned = this.returnedForkContext; const thrown = this.thrownForkContext; returned.add = addToReturnedOrThrown.bind( null, returned, thrown, final, ); thrown.add = addToReturnedOrThrown.bind(null, thrown, returned, final); } /** * A passthrough property exposing the current pointer as part of the API. * @type {CodePathSegment[]} */ get headSegments() { return this.forkContext.head; } /** * The parent forking context. * This is used for the root of new forks. * @type {ForkContext} */ get parentForkContext() { const current = this.forkContext; return current && current.upper; } /** * Creates and stacks new forking context. * @param {boolean} forkLeavingPath A flag which shows being in a * "finally" block. * @returns {ForkContext} The created context. */ pushForkContext(forkLeavingPath) { this.forkContext = ForkContext.newEmpty( this.forkContext, forkLeavingPath, ); return this.forkContext; } /** * Pops and merges the last forking context. * @returns {ForkContext} The last context. */ popForkContext() { const lastContext = this.forkContext; this.forkContext = lastContext.upper; this.forkContext.replaceHead(lastContext.makeNext(0, -1)); return lastContext; } /** * Creates a new path. * @returns {void} */ forkPath() { this.forkContext.add(this.parentForkContext.makeNext(-1, -1)); } /** * Creates a bypass path. * This is used for such as IfStatement which does not have "else" chunk. * @returns {void} */ forkBypassPath() { this.forkContext.add(this.parentForkContext.head); } //-------------------------------------------------------------------------- // ConditionalExpression, LogicalExpression, IfStatement //-------------------------------------------------------------------------- /** * Creates a context for ConditionalExpression, LogicalExpression, AssignmentExpression (logical assignments only), * IfStatement, WhileStatement, DoWhileStatement, or ForStatement. * * LogicalExpressions have cases that it goes different paths between the * `true` case and the `false` case. * * For Example: * * if (a || b) { * foo(); * } else { * bar(); * } * * In this case, `b` is evaluated always in the code path of the `else` * block, but it's not so in the code path of the `if` block. * So there are 3 paths. * * a -> foo(); * a -> b -> foo(); * a -> b -> bar(); * @param {string} kind A kind string. * If the new context is LogicalExpression's or AssignmentExpression's, this is `"&&"` or `"||"` or `"??"`. * If it's IfStatement's or ConditionalExpression's, this is `"test"`. * Otherwise, this is `"loop"`. * @param {boolean} isForkingAsResult Indicates if the result of the choice * creates a fork. * @returns {void} */ pushChoiceContext(kind, isForkingAsResult) { this.choiceContext = new ChoiceContext( this.choiceContext, kind, isForkingAsResult, this.forkContext, ); } /** * Pops the last choice context and finalizes it. * This is called upon leaving a node that represents a choice. * @throws {Error} (Unreachable.) * @returns {ChoiceContext} The popped context. */ popChoiceContext() { const poppedChoiceContext = this.choiceContext; const forkContext = this.forkContext; const head = forkContext.head; this.choiceContext = poppedChoiceContext.upper; switch (poppedChoiceContext.kind) { case "&&": case "||": case "??": /* * The `head` are the path of the right-hand operand. * If we haven't previously added segments from child contexts, * then we add these segments to all possible forks. */ if (!poppedChoiceContext.processed) { poppedChoiceContext.trueForkContext.add(head); poppedChoiceContext.falseForkContext.add(head); poppedChoiceContext.nullishForkContext.add(head); } /* * If this context is the left (test) expression for another choice * context, such as `a || b` in the expression `a || b || c`, * then we take the segments for this context and move them up * to the parent context. */ if (poppedChoiceContext.isForkingAsResult) { const parentContext = this.choiceContext; parentContext.trueForkContext.addAll( poppedChoiceContext.trueForkContext, ); parentContext.falseForkContext.addAll( poppedChoiceContext.falseForkContext, ); parentContext.nullishForkContext.addAll( poppedChoiceContext.nullishForkContext, ); parentContext.processed = true; // Exit early so we don't collapse all paths into one. return poppedChoiceContext; } break; case "test": if (!poppedChoiceContext.processed) { /* * The head segments are the path of the `if` block here. * Updates the `true` path with the end of the `if` block. */ poppedChoiceContext.trueForkContext.clear(); poppedChoiceContext.trueForkContext.add(head); } else { /* * The head segments are the path of the `else` block here. * Updates the `false` path with the end of the `else` * block. */ poppedChoiceContext.falseForkContext.clear(); poppedChoiceContext.falseForkContext.add(head); } break; case "loop": /* * Loops are addressed in `popLoopContext()` so just return * the context without modification. */ return poppedChoiceContext; /* c8 ignore next */ default: throw new Error("unreachable"); } /* * Merge the true path with the false path to create a single path. */ const combinedForkContext = poppedChoiceContext.trueForkContext; combinedForkContext.addAll(poppedChoiceContext.falseForkContext); forkContext.replaceHead(combinedForkContext.makeNext(0, -1)); return poppedChoiceContext; } /** * Creates a code path segment to represent right-hand operand of a logical * expression. * This is called in the preprocessing phase when entering a node. * @throws {Error} (Unreachable.) * @returns {void} */ makeLogicalRight() { const currentChoiceContext = this.choiceContext; const forkContext = this.forkContext; if (currentChoiceContext.processed) { /* * This context was already assigned segments from a child * choice context. In this case, we are concerned only about * the path that does not short-circuit and so ends up on the * right-hand operand of the logical expression. */ let prevForkContext; switch (currentChoiceContext.kind) { case "&&": // if true then go to the right-hand side. prevForkContext = currentChoiceContext.trueForkContext; break; case "||": // if false then go to the right-hand side. prevForkContext = currentChoiceContext.falseForkContext; break; case "??": // Both true/false can short-circuit, so needs the third path to go to the right-hand side. That's nullishForkContext. prevForkContext = currentChoiceContext.nullishForkContext; break; default: throw new Error("unreachable"); } /* * Create the segment for the right-hand operand of the logical expression * and adjust the fork context pointer to point there. The right-hand segment * is added at the end of all segments in `prevForkContext`. */ forkContext.replaceHead(prevForkContext.makeNext(0, -1)); /* * We no longer need this list of segments. * * Reset `processed` because we've removed the segments from the child * choice context. This allows `popChoiceContext()` to continue adding * segments later. */ prevForkContext.clear(); currentChoiceContext.processed = false; } else { /* * This choice context was not assigned segments from a child * choice context, which means that it's a terminal logical * expression. * * `head` is the segments for the left-hand operand of the * logical expression. * * Each of the fork contexts below are empty at this point. We choose * the path(s) that will short-circuit and add the segment for the * left-hand operand to it. Ultimately, this will be the only segment * in that path due to the short-circuting, so we are just seeding * these paths to start. */ switch (currentChoiceContext.kind) { case "&&": /* * In most contexts, when a && expression evaluates to false, * it short circuits, so we need to account for that by setting * the `falseForkContext` to the left operand. * * When a && expression is the left-hand operand for a ?? * expression, such as `(a && b) ?? c`, a nullish value will * also short-circuit in a different way than a false value, * so we also set the `nullishForkContext` to the left operand. * This path is only used with a ?? expression and is thrown * away for any other type of logical expression, so it's safe * to always add. */ currentChoiceContext.falseForkContext.add(forkContext.head); currentChoiceContext.nullishForkContext.add( forkContext.head, ); break; case "||": // the true path can short-circuit. currentChoiceContext.trueForkContext.add(forkContext.head); break; case "??": // both can short-circuit. currentChoiceContext.trueForkContext.add(forkContext.head); currentChoiceContext.falseForkContext.add(forkContext.head); break; default: throw new Error("unreachable"); } /* * Create the segment for the right-hand operand of the logical expression * and adjust the fork context pointer to point there. */ forkContext.replaceHead(forkContext.makeNext(-1, -1)); } } /** * Makes a code path segment of the `if` block. * @returns {void} */ makeIfConsequent() { const context = this.choiceContext; const forkContext = this.forkContext; /* * If any result were not transferred from child contexts, * this sets the head segments to both cases. * The head segments are the path of the test expression. */ if (!context.processed) { context.trueForkContext.add(forkContext.head); context.falseForkContext.add(forkContext.head); context.nullishForkContext.add(forkContext.head); } context.processed = false; // Creates new path from the `true` case. forkContext.replaceHead(context.trueForkContext.makeNext(0, -1)); } /** * Makes a code path segment of the `else` block. * @returns {void} */ makeIfAlternate() { const context = this.choiceContext; const forkContext = this.forkContext; /* * The head segments are the path of the `if` block. * Updates the `true` path with the end of the `if` block. */ context.trueForkContext.clear(); context.trueForkContext.add(forkContext.head); context.processed = true; // Creates new path from the `false` case. forkContext.replaceHead(context.falseForkContext.makeNext(0, -1)); } //-------------------------------------------------------------------------- // ChainExpression //-------------------------------------------------------------------------- /** * Pushes a new `ChainExpression` context to the stack. This method is * called when entering a `ChainExpression` node. A chain context is used to * count forking in the optional chain then merge them on the exiting from the * `ChainExpression` node. * @returns {void} */ pushChainContext() { this.chainContext = new ChainContext(this.chainContext); } /** * Pop a `ChainExpression` context from the stack. This method is called on * exiting from each `ChainExpression` node. This merges all forks of the * last optional chaining. * @returns {void} */ popChainContext() { const context = this.chainContext; this.chainContext = context.upper; // pop all choice contexts of this. for (let i = context.choiceContextCount; i > 0; --i) { this.popChoiceContext(); } } /** * Create a choice context for optional access. * This method is called on entering to each `(Call|Member)Expression[optional=true]` node. * This creates a choice context as similar to `LogicalExpression[operator="??"]` node. * @returns {void} */ makeOptionalNode() { if (this.chainContext) { this.chainContext.choiceContextCount += 1; this.pushChoiceContext("??", false); } } /** * Create a fork. * This method is called on entering to the `arguments|property` property of each `(Call|Member)Expression` node. * @returns {void} */ makeOptionalRight() { if (this.chainContext) { this.makeLogicalRight(); } } //-------------------------------------------------------------------------- // SwitchStatement //-------------------------------------------------------------------------- /** * Creates a context object of SwitchStatement and stacks it. * @param {boolean} hasCase `true` if the switch statement has one or more * case parts. * @param {string|null} label The label text. * @returns {void} */ pushSwitchContext(hasCase, label) { this.switchContext = new SwitchContext(this.switchContext, hasCase); this.pushBreakContext(true, label); } /** * Pops the last context of SwitchStatement and finalizes it. * * - Disposes all forking stack for `case` and `default`. * - Creates the next code path segment from `context.brokenForkContext`. * - If the last `SwitchCase` node is not a `default` part, creates a path * to the `default` body. * @returns {void} */ popSwitchContext() { const context = this.switchContext; this.switchContext = context.upper; const forkContext = this.forkContext; const brokenForkContext = this.popBreakContext().brokenForkContext; if (context.forkCount === 0) { /* * When there is only one `default` chunk and there is one or more * `break` statements, even if forks are nothing, it needs to merge * those. */ if (!brokenForkContext.empty) { brokenForkContext.add(forkContext.makeNext(-1, -1)); forkContext.replaceHead(brokenForkContext.makeNext(0, -1)); } return; } const lastSegments = forkContext.head; this.forkBypassPath(); const lastCaseSegments = forkContext.head; /* * `brokenForkContext` is used to make the next segment. * It must add the last segment into `brokenForkContext`. */ brokenForkContext.add(lastSegments); /* * Any value that doesn't match a `case` test should flow to the default * case. That happens normally when the default case is last in the `switch`, * but if it's not, we need to rewire some of the paths to be correct. */ if (!context.lastIsDefault) { if (context.defaultBodySegments) { /* * There is a non-empty default case, so remove the path from the `default` * label to its body for an accurate representation. */ disconnectSegments( context.defaultSegments, context.defaultBodySegments, ); /* * Connect the path from the last non-default case to the body of the * default case. */ makeLooped(this, lastCaseSegments, context.defaultBodySegments); } else { /* * There is no default case, so we treat this as if the last case * had a `break` in it. */ brokenForkContext.add(lastCaseSegments); } } // Traverse up to the original fork context for the `switch` statement for (let i = 0; i < context.forkCount; ++i) { this.forkContext = this.forkContext.upper; } /* * Creates a path from all `brokenForkContext` paths. * This is a path after `switch` statement. */ this.forkContext.replaceHead(brokenForkContext.makeNext(0, -1)); } /** * Makes a code path segment for a `SwitchCase` node. * @param {boolean} isCaseBodyEmpty `true` if the body is empty. * @param {boolean} isDefaultCase `true` if the body is the default case. * @returns {void} */ makeSwitchCaseBody(isCaseBodyEmpty, isDefaultCase) { const context = this.switchContext; if (!context.hasCase) { return; } /* * Merge forks. * The parent fork context has two segments. * Those are from the current `case` and the body of the previous case. */ const parentForkContext = this.forkContext; const forkContext = this.pushForkContext(); forkContext.add(parentForkContext.makeNext(0, -1)); /* * Add information about the default case. * * The purpose of this is to identify the starting segments for the * default case to make sure there is a path there. */ if (isDefaultCase) { /* * This is the default case in the `switch`. * * We first save the current pointer as `defaultSegments` to point * to the `default` keyword. */ context.defaultSegments = parentForkContext.head; /* * If the body of the case is empty then we just set * `foundEmptyDefault` to true; otherwise, we save a reference * to the current pointer as `defaultBodySegments`. */ if (isCaseBodyEmpty) { context.foundEmptyDefault = true; } else { context.defaultBodySegments = forkContext.head; } } else { /* * This is not the default case in the `switch`. * * If it's not empty and there is already an empty default case found, * that means the default case actually comes before this case, * and that it will fall through to this case. So, we can now * ignore the previous default case (reset `foundEmptyDefault` to false) * and set `defaultBodySegments` to the current segments because this is * effectively the new default case. */ if (!isCaseBodyEmpty && context.foundEmptyDefault) { context.foundEmptyDefault = false; context.defaultBodySegments = forkContext.head; } } // keep track if the default case ends up last context.lastIsDefault = isDefaultCase; context.forkCount += 1; } //-------------------------------------------------------------------------- // TryStatement //-------------------------------------------------------------------------- /** * Creates a context object of TryStatement and stacks it. * @param {boolean} hasFinalizer `true` if the try statement has a * `finally` block. * @returns {void} */ pushTryContext(hasFinalizer) { this.tryContext = new TryContext( this.tryContext, hasFinalizer, this.forkContext, ); } /** * Pops the last context of TryStatement and finalizes it. * @returns {void} */ popTryContext() { const context = this.tryContext; this.tryContext = context.upper; /* * If we're inside the `catch` block, that means there is no `finally`, * so we can process the `try` and `catch` blocks the simple way and * merge their two paths. */ if (context.position === "catch") { this.popForkContext(); return; } /* * The following process is executed only when there is a `finally` * block. */ const originalReturnedForkContext = context.returnedForkContext; const originalThrownForkContext = context.thrownForkContext; // no `return` or `throw` in `try` or `catch` so there's nothing left to do if ( originalReturnedForkContext.empty && originalThrownForkContext.empty ) { return; } /* * The following process is executed only when there is a `finally` * block and there was a `return` or `throw` in the `try` or `catch` * blocks. */ // Separate head to normal paths and leaving paths. const headSegments = this.forkContext.head; this.forkCon