Uglifyjs: string template

[kenlaslo]

When string templates syntax exists in code, a bug has been found in the following part.

aaa bbb cccc;

The expected results isaaa\nbbb\nccc, however after compressing aaa\r\nbbb\r\nccc
as results, Two line breaks will occur.

thank you.

[alexlamsl]

Please provide all the information to reproduce the issue at hand, as detailed in the template when you open this report.
https://github.com/mishoo/UglifyJS2/blob/master/.github/ISSUE_TEMPLATE.md

[kzc]

repro:

$ cat abc.js
console.log(`a
b
c\nw\n\rx\r\ny\rz`);

$ cat abc.js | node | od -c
0000000    a  \n   b  \n   c  \n   w  \n  \r   x  \r  \n   y  \r   z  \n
0000020



md5-e9a98e66ef811e619d4a432a2a7e4483



$ cat abc.js | unix2dos | node | od -c
0000000    a  \n   b  \n   c  \n   w  \n  \r   x  \r  \n   y  \r   z  \n
0000020



md5-e9a98e66ef811e619d4a432a2a7e4483



$ cat abc.js | bin/uglifyjs -c | node | od -c
0000000    a  \n   b  \n   c  \n   w  \n  \r   x  \r  \n   y  \r   z  \n
0000020



md5-e9a98e66ef811e619d4a432a2a7e4483



$ cat abc.js | unix2dos | bin/uglifyjs -c | node | od -c
0000000    a  \r  \n   b  \r  \n   c  \n   w  \n  \r   x  \r  \n   y  \r
0000020    z  \n                                                        
0000022



md5-b8333ca2099f41ec408ebf8f9e53fcde



*.js    text eol=lf

[kzc]

Fix:

--- a/lib/parse.js
+++ b/lib/parse.js
@@ -519,7 +519,11 @@ function tokenizer($TEXT, filename, html5_comments, shebang) {
                 tok.end = false;
                 return tok;
             }
-
+            if (ch == "\r" && peek() == "\n") {
+                // treat a \r\n sequence as a single \n
+                ++S.pos;
+                ch = "\n";
+            }
             raw += ch;
             if (ch === "\\") {
                 var tmp = S.pos;

which is similar to:
https://github.com/mishoo/UglifyJS2/blob/harmony-v3.0.20/lib/parse.js#L295-L299

[alexlamsl]

@kzc thanks for the test case.

So is this line break conversion behaviour part of ES6 specification or just specific to Node.js?

[kzc]

It's an ES6 thing. acorn has the same behavior. As does babel:

$ cat abc.js | babel
"use strict";

console.log("a\nb\nc\nw\n\rx\r\ny\rz");

$ cat abc.js | unix2dos | babel
"use strict";

console.log("a\nb\nc\nw\n\rx\r\ny\rz");

[alexlamsl]

@kzc cool - I'll put that PR together later on today :+1:

[alexlamsl]

Some supporting text:
http://exploringjs.com/es6/ch_template-literals.html#_line-terminators-in-template-literals-are-always-lf-n

[alexlamsl]

But in that case, even \r alone will need to be converted to \n

[kenlaslo]

First, I'm sorry about late reply.
Thank you very much for your corrections.
If so, what should I do to enjoy the changes?
Reinstall or Edit the sources directly?
Note that, I am not familiar with node js and github here.
Thank you.

[kzc]

@kenlaslo You'll have to wait for a new 3.0.21 release for the fix. Perhaps in a couple of days.

[alexlamsl]

@kenlaslo as @kzc pointed out, 3.0.21 has not been released yet - I shall do so in a couple of hours.

As for how you may get this, it depends on how uglify-es is specified under your package.json. But in many cases npm update should get the job done.

[kenlaslo]

@kzc , @alexlamsl thank you so much :)