r.js 2.1.3 (and greater) is four times slower than 2.1.2

[mdaniel]

$ wget -O r-2.1.2.js https://github.com/jrburke/r.js/raw/2.1.2/dist/r.js
$ time /usr/bin/node r-2.1.2.js -o target/js-config/app.build.js
real    1m9.357s
user    1m8.717s
sys 0m0.498s

$ wget -O r-2.1.3.js https://github.com/jrburke/r.js/raw/2.1.3/dist/r.js
$ time /usr/bin/node r-2.1.3.js -o target/js-config/app.build.js
real    4m31.063s
user    4m29.710s
sys 0m0.876s

https://raw.github.com/jrburke/r.js/2.1.4/dist/r.js
real    4m27.894s
user    4m26.718s
sys 0m0.743s

https://raw.github.com/jrburke/r.js/2.1.5/dist/r.js
real    4m27.136s
user    4m26.043s
sys 0m0.698s

https://raw.github.com/jrburke/r.js/2.1.6/dist/r.js
real    4m40.676s
user    4m39.208s
sys 0m0.895s

This is with nodejs v0.10.7, Linux amd64.

app.build.js is {
"optimize": "uglify2"
, "optimizeCss": "standard"
, "generateSourceMaps":true
, "preserveLicenseComments":false
, "useStrict": true
} // and some other parts that I doubt are relevant; let me know if you need the specifics

[jrburke]

My first guess is that it is the source map work, if you set generateSourceMaps: false, does that perform better?

[mdaniel]

It doesn't seem to help.
with 2.1.6:
real 4m28.787s
user 4m27.493s
sys 0m0.748s

with 2.1.2:
real 1m4.882s
user 1m4.347s
sys 0m0.393s

[jrburke]

Looking at the things changed for 2.1.3:
https://github.com/jrburke/r.js/issues?milestone=21&state=closed

There was a fix for uglify2js, I was inlining it incorrectly so it was missing some content for some WHITESPACE_CHARS, fixed in #323, perhaps that would explain it? Another candidate may be #245 since it does some extra AST work, but that should only come into play if namespace: something is in the build config.

So, what may be interesting to try is to use optimize: 'uglify' which uses uglifyjs1, and if with that setting the times are similar between 2.1.2 and 2.1.3, then it seems like the culprit is just uglify2. 'uglify' does not support source maps so you will need to turn that option off too for that test.

[mdaniel]

Yup, I can confirm that optimize: 'uglify' moves the execution times back to 0m50 for all the versions.

Do I need to file a subsequent issue with Uglify2, citing this issue?

[jrburke]

Ideally, create a JS file you want to run through uglify-js. Try it with the stand-alone uglify-js and record the times between uglify1 and uglify2, and hopefully pass that JS file to the uglify folks to ask what accounts for the slow down.

By generating the numbers with standalone uglifyjs, that will hopefully not confuse them by thinking it is an issue with r.js.

Going to close this as it is something uglify will need to address itself.

[gamelover42]

I'm not sure if this is related to the problem described above but it is worth mentioning that there is a known performance issue with uglify2 and pre-minified files mishoo/UglifyJS#321