blob: ddbbcc42ae1acd8f065e0ec7315bae0c4b686ee7 [file] [log] [blame]
/*
* Copyright (C) 2013 Apple Inc. All rights reserved.
* Copyright (C) 2015 Tobias Reiss <tobi+webkit@basecode.de>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS''
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
// In the inspector token types have been modified to include extra mode information
// after the actual token type. So we can't do token === "foo". So instead we do
// /\bfoo\b/.test(token).
CodeMirror.extendMode("javascript", {
shouldHaveSpaceBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (!token) {
if (content === "(") // Most keywords like "if (" but not "function(" or "typeof(".
return lastToken && /\bkeyword\b/.test(lastToken) && (lastContent !== "function" && lastContent !== "typeof" && lastContent !== "instanceof");
if (content === ":") // Ternary.
return state.lexical.type === "stat" || state.lexical.type === ")" || state.lexical.type === "]";
return false;
}
if (isComment)
return true;
if (/\boperator\b/.test(token)) {
if (!lastToken && (content === "+" || content === "-" || content === "~") && (lastContent !== ")" && lastContent !== "]")) // Possible Unary +/-.
return false;
if (content === "!") // Unary ! should not be confused with "!=".
return false;
return "+-/*%&&||!===+=-=>=<=?".indexOf(content) >= 0; // Operators.
}
if (/\bkeyword\b/.test(token)) { // Most keywords require spaces before them, unless a '}' can come before it.
if (content === "else" || content === "catch" || content === "finally")
return lastContent === "}";
if (content === "while" && lastContent === "}")
return state._jsPrettyPrint.lastContentBeforeBlock === "do";
return false;
}
return false;
},
shouldHaveSpaceAfterLastToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (lastToken && /\bkeyword\b/.test(lastToken)) { // Most keywords require spaces after them, unless a '{' or ';' can come after it.
if (lastContent === "else")
return true;
if (lastContent === "catch")
return true;
if (lastContent === "return")
return content !== ";";
if (lastContent === "throw")
return true;
if (lastContent === "try")
return true;
if (lastContent === "finally")
return true;
if (lastContent === "do")
return true;
return false;
}
if (lastToken && /\bcomment\b/.test(lastToken)) // Embedded /* comment */.
return true;
if (lastContent === ")") // "){".
return content === "{";
if (lastContent === ";") // In for loop.
return state.lexical.type === ")";
if (lastContent === "!") // Unary ! should not be confused with "!=".
return false;
// If this unary operator did not have a leading expression it is probably unary.
if ((lastContent === "+" || lastContent === "-" || lastContent === "~") && !state._jsPrettyPrint.unaryOperatorHadLeadingExpr)
return false;
return ",+-/*%&&||:!===+=-=>=<=?".indexOf(lastContent) >= 0; // Operators.
},
newlinesAfterToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (!token) {
if (content === ",") // In object literals, like in {a:1,b:2}, but not in param lists or vardef lists.
return state.lexical.type === "}" ? 1 : 0;
if (content === ";") // Everywhere except in for loop conditions.
return state.lexical.type !== ")" ? 1 : 0;
if (content === ":" && state.lexical.type === "}" && state.lexical.prev && state.lexical.prev.type === "form") // Switch case/default.
return 1;
return content.length === 1 && "{}".indexOf(content) >= 0 ? 1 : 0; // After braces.
}
if (isComment)
return 1;
return 0;
},
removeLastWhitespace: function(lastToken, lastContent, token, state, content, isComment)
{
return false;
},
removeLastNewline: function(lastToken, lastContent, token, state, content, isComment, firstTokenOnLine)
{
if (!token) {
if (content === "}") // "{}".
return lastContent === "{";
if (content === ";") // "x = {};" or ";;".
return "};".indexOf(lastContent) >= 0;
if (content === ":") // Ternary.
return lastContent === "}" && (state.lexical.type === "stat" || state.lexical.type === ")" || state.lexical.type === "]");
if (",().".indexOf(content) >= 0) // "})", "}.bind", "function() { ... }()", or "}, false)".
return lastContent === "}";
return false;
}
if (isComment) { // Comment after semicolon.
if (!firstTokenOnLine && lastContent === ";")
return true;
return false;
}
if (/\bkeyword\b/.test(token)) {
if (content === "else" || content === "catch" || content === "finally") // "} else", "} catch", "} finally"
return lastContent === "}";
if (content === "while" && lastContent === "}")
return state._jsPrettyPrint.lastContentBeforeBlock === "do";
return false;
}
return false;
},
indentAfterToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (content === "{")
return true;
if (content === "case" || content === "default")
return state.lexical.type === "}" && state.lexical.prev && state.lexical.prev.type === "form"; // Switch case/default.
return false;
},
newlineBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (state._jsPrettyPrint.shouldIndent)
return true;
return content === "}" && lastContent !== "{"; // "{}"
},
indentBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (state._jsPrettyPrint.shouldIndent)
return true;
return false;
},
dedentsBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
var dedent = 0;
if (state._jsPrettyPrint.shouldDedent)
dedent += state._jsPrettyPrint.dedentSize;
if (!token && content === "}")
dedent += 1;
else if (token && /\bkeyword\b/.test(token) && (content === "case" || content === "default"))
dedent += 1;
return dedent;
},
modifyStateForTokenPre: function(lastToken, lastContent, token, state, content, isComment)
{
if (!state._jsPrettyPrint) {
state._jsPrettyPrint = {
indentCount: 0, // How far have we indented because of single statement blocks.
shouldIndent: false, // Signal we should indent on entering a single statement block.
shouldDedent: false, // Signal we should dedent on leaving a single statement block.
dedentSize: 0, // How far we should dedent when leaving a single statement block.
lastIfIndentCount: 0, // Keep track of the indent the last time we saw an if without braces.
openBraceStartMarkers: [], // Keep track of non-single statement blocks.
openBraceTrackingCount: -1, // Keep track of "{" and "}" in non-single statement blocks.
unaryOperatorHadLeadingExpr: false, // Try to detect if a unary operator had a leading expression and therefore may be binary.
lastContentBeforeBlock: undefined, // Used to detect if this was a do/while.
};
}
// - Entering:
// - Preconditions:
// - last lexical was a "form" we haven't encountered before
// - last content was ")", "else", or "do"
// - current lexical is not ")" (in an expression or condition)
// - Cases:
// 1. "{"
// - indent +0
// - save this indent size so when we encounter the "}" we know how far to dedent
// 2. "else if"
// - indent +0 and do not signal to add a newline and indent
// - mark the last if location so when we encounter an "else" we know how far to dedent
// - mark the lexical state so we know we are inside a single statement block
// 3. Token without brace.
// - indent +1 and signal to add a newline and indent
// - mark the last if location so when we encounter an "else" we know how far to dedent
// - mark the lexical state so we know we are inside a single statement block
if (!isComment && state.lexical.prev && state.lexical.prev.type === "form" && !state.lexical.prev._jsPrettyPrintMarker && (lastContent === ")" || lastContent === "else" || lastContent === "do") && (state.lexical.type !== ")")) {
if (content === "{") {
// Save the state at the opening brace so we can return to it when we see "}".
var savedState = {indentCount: state._jsPrettyPrint.indentCount, openBraceTrackingCount: state._jsPrettyPrint.openBraceTrackingCount, lastContentBeforeBlock: lastContent};
state._jsPrettyPrint.openBraceStartMarkers.push(savedState);
state._jsPrettyPrint.openBraceTrackingCount = 1;
} else if (state.lexical.type !== "}") {
// Increase the indent count. Signal for a newline and indent if needed.
if (!(lastContent === "else" && content === "if")) {
state._jsPrettyPrint.indentCount++;
state._jsPrettyPrint.shouldIndent = true;
}
state.lexical.prev._jsPrettyPrintMarker = true;
if (state._jsPrettyPrint.enteringIf)
state._jsPrettyPrint.lastIfIndentCount = state._jsPrettyPrint.indentCount - 1;
}
}
// - Leaving:
// - Preconditions:
// - ignore ";", wait for the next token instead.
// - Cases:
// 1. "else"
// - dedent to the last "if"
// 2. "}" and all braces we saw are balanced
// - dedent to the last "{"
// 3. Token without a marker on the stack
// - dedent all the way
else {
console.assert(!state._jsPrettyPrint.shouldDedent);
console.assert(!state._jsPrettyPrint.dedentSize);
// Track "{" and "}" to know when the "}" is really closing a block.
if (!isComment) {
if (content === "{")
state._jsPrettyPrint.openBraceTrackingCount++;
else if (content === "}")
state._jsPrettyPrint.openBraceTrackingCount--;
}
if (content === ";") {
// Ignore.
} else if (content === "else") {
// Dedent to the last "if".
if (lastContent !== "}") {
state._jsPrettyPrint.shouldDedent = true;
state._jsPrettyPrint.dedentSize = state._jsPrettyPrint.indentCount - state._jsPrettyPrint.lastIfIndentCount;
state._jsPrettyPrint.lastIfIndentCount = 0;
}
} else if (content === "}" && !state._jsPrettyPrint.openBraceTrackingCount && state._jsPrettyPrint.openBraceStartMarkers.length) {
// Dedent to the last "{".
var savedState = state._jsPrettyPrint.openBraceStartMarkers.pop();
state._jsPrettyPrint.shouldDedent = true;
state._jsPrettyPrint.dedentSize = state._jsPrettyPrint.indentCount - savedState.indentCount;
state._jsPrettyPrint.openBraceTrackingCount = savedState.openBraceTrackingCount;
state._jsPrettyPrint.lastContentBeforeBlock = savedState.lastContentBeforeBlock;
} else {
// Dedent all the way.
var shouldDedent = true;
var lexical = state.lexical.prev;
while (lexical) {
if (lexical._jsPrettyPrintMarker) {
shouldDedent = false;
break;
}
lexical = lexical.prev;
}
if (shouldDedent) {
state._jsPrettyPrint.shouldDedent = true;
state._jsPrettyPrint.dedentSize = state._jsPrettyPrint.indentCount;
}
}
}
// Signal for when we will be entering an if.
if (token && state.lexical.type === "form" && state.lexical.prev && state.lexical.prev !== "form" && /\bkeyword\b/.test(token))
state._jsPrettyPrint.enteringIf = content === "if";
},
modifyStateForTokenPost: function(lastToken, lastContent, token, state, content, isComment)
{
if (state._jsPrettyPrint.shouldIndent)
state._jsPrettyPrint.shouldIndent = false;
if (state._jsPrettyPrint.shouldDedent) {
state._jsPrettyPrint.indentCount -= state._jsPrettyPrint.dedentSize;
state._jsPrettyPrint.dedentSize = 0;
state._jsPrettyPrint.shouldDedent = false;
}
if ((content === "+" || content === "-" || content === "~") && (lastContent === ")" || lastContent === "]" || /\b(?:variable|number)\b/.test(lastToken)))
state._jsPrettyPrint.unaryOperatorHadLeadingExpr = true;
else
state._jsPrettyPrint.unaryOperatorHadLeadingExpr = false;
}
});
CodeMirror.extendMode("css", {
shouldHaveSpaceBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (!token) {
if (content === "{")
return true;
return ">+~-*/".indexOf(content) >= 0; // calc() expression or child/sibling selectors
}
if (isComment)
return true;
if (/\bkeyword\b/.test(token)) {
if (content.charAt(0) === "!") // "!important".
return true;
return false;
}
return false;
},
shouldHaveSpaceAfterLastToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (!lastToken) {
if (lastContent === ",")
return true;
if (lastContent === ":") // Space in "prop: value" but not in a selectors "a:link" or "div::after" or media queries "(max-device-width:480px)".
return state.state === "prop";
if (lastContent === ")" && (content !== ")" && content !== ",")) {
if (/\bnumber\b/.test(token)) // linear-gradient(rgb(...)0%,rgb(...)100%)
return true;
if (state.state === "prop") // -webkit-transform:rotate(...)translate(...);
return true;
if (state.state === "media" || state.state === "atBlock_parens") // Space in "not(foo)and" but not at the end of "not(not(foo))"
return true;
return false; // color: rgb(...);
}
return ">+~-*/".indexOf(lastContent) >= 0; // calc() expression or child/sibling selectors
}
if (/\bcomment\b/.test(lastToken))
return true;
if (/\bkeyword\b/.test(lastToken)) // media-query keywords
return state.state === "media" || (state.state === "atBlock_parens" && content !== ")");
return false;
},
newlinesAfterToken: function(lastToken, lastContent, token, state, content, isComment)
{
if (!token) {
if (content === ";")
return 1;
if (content === ",") { // "a,b,c,...,z{}" rule list at top level or in @media top level and only if the line length will be large.
if ((state.state === "top" || state.state === "media") && state._cssPrettyPrint.lineLength > 60) {
state._cssPrettyPrint.lineLength = 0;
return 1;
}
return 0;
}
if (content === "{")
return 1;
if (content === "}") // 2 newlines between rule declarations.
return 2;
return 0;
}
if (isComment)
return 1;
return 0;
},
removeLastWhitespace: function(lastToken, lastContent, token, state, content, isComment)
{
return false;
},
removeLastNewline: function(lastToken, lastContent, token, state, content, isComment, firstTokenOnLine)
{
if (isComment) { // Comment after semicolon.
if (!firstTokenOnLine && lastContent === ";")
return true;
return false;
}
return content === "}" && (lastContent === "{" || lastContent === "}"); // "{}" and "}\n}" when closing @media.
},
indentAfterToken: function(lastToken, lastContent, token, state, content, isComment)
{
return content === "{";
},
newlineBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
return content === "}" && (lastContent !== "{" && lastContent !== "}"); // "{}" and "}\n}" when closing @media.
},
indentBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
return false;
},
dedentsBeforeToken: function(lastToken, lastContent, token, state, content, isComment)
{
return content === "}" ? 1 : 0;
},
modifyStateForTokenPost: function(lastToken, lastContent, token, state, content, isComment)
{
if (!state._cssPrettyPrint)
state._cssPrettyPrint = {lineLength: 0};
// In order insert newlines in selector lists we need keep track of the length of the current line.
// This isn't exact line length, only the builder knows that, but it is good enough to get an idea.
// If we are at a top level, keep track of the current line length, otherwise we reset to 0.
if (!isComment && (state.state === "top" || state.state === "media" || state.state === "pseudo"))
state._cssPrettyPrint.lineLength += content.length;
else
state._cssPrettyPrint.lineLength = 0;
}
});