comments.js 11.1 KB
var assert = require("assert");
var types = require("./types");
var n = types.namedTypes;
var isArray = types.builtInTypes.array;
var isObject = types.builtInTypes.object;
var linesModule = require("./lines");
var fromString = linesModule.fromString;
var Lines = linesModule.Lines;
var concat = linesModule.concat;
var util = require("./util");
var comparePos = util.comparePos;
var childNodesCacheKey = require("private").makeUniqueKey();

// TODO Move a non-caching implementation of this function into ast-types,
// and implement a caching wrapper function here.
function getSortedChildNodes(node, resultArray) {
    if (!node) {
        return;
    }

    // The .loc checks below are sensitive to some of the problems that
    // are fixed by this utility function. Specifically, if it decides to
    // set node.loc to null, indicating that the node's .loc information
    // is unreliable, then we don't want to add node to the resultArray.
    util.fixFaultyLocations(node);

    if (resultArray) {
        if (n.Node.check(node) &&
            n.SourceLocation.check(node.loc)) {
            // This reverse insertion sort almost always takes constant
            // time because we almost always (maybe always?) append the
            // nodes in order anyway.
            for (var i = resultArray.length - 1; i >= 0; --i) {
                if (comparePos(resultArray[i].loc.end,
                               node.loc.start) <= 0) {
                    break;
                }
            }
            resultArray.splice(i + 1, 0, node);
            return;
        }
    } else if (node[childNodesCacheKey]) {
        return node[childNodesCacheKey];
    }

    var names;
    if (isArray.check(node)) {
        names = Object.keys(node);
    } else if (isObject.check(node)) {
        names = types.getFieldNames(node);
    } else {
        return;
    }

    if (!resultArray) {
        Object.defineProperty(node, childNodesCacheKey, {
            value: resultArray = [],
            enumerable: false
        });
    }

    for (var i = 0, nameCount = names.length; i < nameCount; ++i) {
        getSortedChildNodes(node[names[i]], resultArray);
    }

    return resultArray;
}

// As efficiently as possible, decorate the comment object with
// .precedingNode, .enclosingNode, and/or .followingNode properties, at
// least one of which is guaranteed to be defined.
function decorateComment(node, comment) {
    var childNodes = getSortedChildNodes(node);

    // Time to dust off the old binary search robes and wizard hat.
    var left = 0, right = childNodes.length;
    while (left < right) {
        var middle = (left + right) >> 1;
        var child = childNodes[middle];

        if (comparePos(child.loc.start, comment.loc.start) <= 0 &&
            comparePos(comment.loc.end, child.loc.end) <= 0) {
            // The comment is completely contained by this child node.
            decorateComment(comment.enclosingNode = child, comment);
            return; // Abandon the binary search at this level.
        }

        if (comparePos(child.loc.end, comment.loc.start) <= 0) {
            // This child node falls completely before the comment.
            // Because we will never consider this node or any nodes
            // before it again, this node must be the closest preceding
            // node we have encountered so far.
            var precedingNode = child;
            left = middle + 1;
            continue;
        }

        if (comparePos(comment.loc.end, child.loc.start) <= 0) {
            // This child node falls completely after the comment.
            // Because we will never consider this node or any nodes after
            // it again, this node must be the closest following node we
            // have encountered so far.
            var followingNode = child;
            right = middle;
            continue;
        }

        throw new Error("Comment location overlaps with node location");
    }

    if (precedingNode) {
        comment.precedingNode = precedingNode;
    }

    if (followingNode) {
        comment.followingNode = followingNode;
    }
}

exports.attach = function(comments, ast, lines) {
    if (!isArray.check(comments)) {
        return;
    }

    var tiesToBreak = [];

    comments.forEach(function(comment) {
        comment.loc.lines = lines;
        decorateComment(ast, comment);

        var pn = comment.precedingNode;
        var en = comment.enclosingNode;
        var fn = comment.followingNode;

        if (pn && fn) {
            var tieCount = tiesToBreak.length;
            if (tieCount > 0) {
                var lastTie = tiesToBreak[tieCount - 1];

                assert.strictEqual(
                    lastTie.precedingNode === comment.precedingNode,
                    lastTie.followingNode === comment.followingNode
                );

                if (lastTie.followingNode !== comment.followingNode) {
                    breakTies(tiesToBreak, lines);
                }
            }

            tiesToBreak.push(comment);

        } else if (pn) {
            // No contest: we have a trailing comment.
            breakTies(tiesToBreak, lines);
            addTrailingComment(pn, comment);

        } else if (fn) {
            // No contest: we have a leading comment.
            breakTies(tiesToBreak, lines);
            addLeadingComment(fn, comment);

        } else if (en) {
            // The enclosing node has no child nodes at all, so what we
            // have here is a dangling comment, e.g. [/* crickets */].
            breakTies(tiesToBreak, lines);
            addDanglingComment(en, comment);

        } else {
            throw new Error("AST contains no nodes at all?");
        }
    });

    breakTies(tiesToBreak, lines);

    comments.forEach(function(comment) {
        // These node references were useful for breaking ties, but we
        // don't need them anymore, and they create cycles in the AST that
        // may lead to infinite recursion if we don't delete them here.
        delete comment.precedingNode;
        delete comment.enclosingNode;
        delete comment.followingNode;
    });
};

function breakTies(tiesToBreak, lines) {
    var tieCount = tiesToBreak.length;
    if (tieCount === 0) {
        return;
    }

    var pn = tiesToBreak[0].precedingNode;
    var fn = tiesToBreak[0].followingNode;
    var gapEndPos = fn.loc.start;

    // Iterate backwards through tiesToBreak, examining the gaps
    // between the tied comments. In order to qualify as leading, a
    // comment must be separated from fn by an unbroken series of
    // whitespace-only gaps (or other comments).
    for (var indexOfFirstLeadingComment = tieCount;
         indexOfFirstLeadingComment > 0;
         --indexOfFirstLeadingComment) {
        var comment = tiesToBreak[indexOfFirstLeadingComment - 1];
        assert.strictEqual(comment.precedingNode, pn);
        assert.strictEqual(comment.followingNode, fn);

        var gap = lines.sliceString(comment.loc.end, gapEndPos);
        if (/\S/.test(gap)) {
            // The gap string contained something other than whitespace.
            break;
        }

        gapEndPos = comment.loc.start;
    }

    while (indexOfFirstLeadingComment <= tieCount &&
           (comment = tiesToBreak[indexOfFirstLeadingComment]) &&
           // If the comment is a //-style comment and indented more
           // deeply than the node itself, reconsider it as trailing.
           comment.type === "Line" &&
           comment.loc.start.column > fn.loc.start.column) {
        ++indexOfFirstLeadingComment;
    }

    tiesToBreak.forEach(function(comment, i) {
        if (i < indexOfFirstLeadingComment) {
            addTrailingComment(pn, comment);
        } else {
            addLeadingComment(fn, comment);
        }
    });

    tiesToBreak.length = 0;
}

function addCommentHelper(node, comment) {
    var comments = node.comments || (node.comments = []);
    comments.push(comment);
}

function addLeadingComment(node, comment) {
    comment.leading = true;
    comment.trailing = false;
    addCommentHelper(node, comment);
}

function addDanglingComment(node, comment) {
    comment.leading = false;
    comment.trailing = false;
    addCommentHelper(node, comment);
}

function addTrailingComment(node, comment) {
    comment.leading = false;
    comment.trailing = true;
    addCommentHelper(node, comment);
}

function printLeadingComment(commentPath, print) {
    var comment = commentPath.getValue();
    n.Comment.assert(comment);

    var loc = comment.loc;
    var lines = loc && loc.lines;
    var parts = [print(commentPath)];

    if (comment.trailing) {
        // When we print trailing comments as leading comments, we don't
        // want to bring any trailing spaces along.
        parts.push("\n");

    } else if (lines instanceof Lines) {
        var trailingSpace = lines.slice(
            loc.end,
            lines.skipSpaces(loc.end)
        );

        if (trailingSpace.length === 1) {
            // If the trailing space contains no newlines, then we want to
            // preserve it exactly as we found it.
            parts.push(trailingSpace);
        } else {
            // If the trailing space contains newlines, then replace it
            // with just that many newlines, with all other spaces removed.
            parts.push(new Array(trailingSpace.length).join("\n"));
        }

    } else {
        parts.push("\n");
    }

    return concat(parts);
}

function printTrailingComment(commentPath, print) {
    var comment = commentPath.getValue(commentPath);
    n.Comment.assert(comment);

    var loc = comment.loc;
    var lines = loc && loc.lines;
    var parts = [];

    if (lines instanceof Lines) {
        var fromPos = lines.skipSpaces(loc.start, true) || lines.firstPos();
        var leadingSpace = lines.slice(fromPos, loc.start);

        if (leadingSpace.length === 1) {
            // If the leading space contains no newlines, then we want to
            // preserve it exactly as we found it.
            parts.push(leadingSpace);
        } else {
            // If the leading space contains newlines, then replace it
            // with just that many newlines, sans all other spaces.
            parts.push(new Array(leadingSpace.length).join("\n"));
        }
    }

    parts.push(print(commentPath));

    return concat(parts);
}

exports.printComments = function(path, print) {
    var value = path.getValue();
    var innerLines = print(path);
    var comments = n.Node.check(value) &&
        types.getFieldValue(value, "comments");

    if (!comments || comments.length === 0) {
        return innerLines;
    }

    var leadingParts = [];
    var trailingParts = [innerLines];

    path.each(function(commentPath) {
        var comment = commentPath.getValue();
        var leading = types.getFieldValue(comment, "leading");
        var trailing = types.getFieldValue(comment, "trailing");

        if (leading || (trailing && comment.type !== "Block")) {
            leadingParts.push(printLeadingComment(commentPath, print));
        } else if (trailing) {
            assert.strictEqual(comment.type, "Block");
            trailingParts.push(printTrailingComment(commentPath, print));
        }
    }, "comments");

    leadingParts.push.apply(leadingParts, trailingParts);
    return concat(leadingParts);
};