Benchmarks results using jade 1.10.0 and ejs 2.3.1

Please note that compileDebug was disabled (to be expected in production), but also that self(jade) was enabled/with(ejs) was disabled. This is less likely to happen in production but otherwise ejs becomes incredibly slow because it relies on JavaScript's own with as opposed to jade's usage of the highly optimized https://github.com/ForbesLindesay/with. Therefore, ejs can show all of its strength and is expected to outperform Jade because

1. Jade's for loop adds sugar to handle both arrays and objects -> additional runtime cost
2. Jade's for loop gets its own closure and variable scope

JSPerf numbers (http://jsperf.com/html-compilers) show that, even in the most synthetic of benchmarks, this should cause only a 10% - 15% drop and less on a 'real' website.

Jade

• mixin-escape: 44930ms
• mixin-no-escape: 20945ms
• no-mixin-escape: 28157ms
• no-mixin-no-escape: 6409ms
• simple: 404ms

Jade (alubbe optimized branch)

• mixin-escape: 7457ms
• mixin-no-escape: 2305ms
• no-mixin-escape: 7884ms
• no-mixin-no-escape: 1313ms
• simple: 259ms

EJS

• mixin-escape: 6608ms
• mixin-no-escape: 1653ms
• no-mixin-escape: 7710ms
• no-mixin-no-escape: 1456ms
• simple: 310ms

Identified reasons for performance differences

• jade's escape function is slower than EJS' (20-30%, fixed with pugjs/pug#1976)
• Array.join is slower than +=, especially for a small number of items (15-50%, fixed with pugjs/pug#1977)
• jade's each has to perform runtime checks whether to iterate of an array or object, making it a bit slower than two separate iteration implementations (~5%)
• jade's each runs in an IIFE for variable scoping, losing quite a bit of performance (~20%, fixed by pugjs/pug#1983)
• jade's mixin implementation is really, really slow. It runs about 400% slower than EJS' functions. (fixed by pugjs/pug#1982 - turned out to be a v8 bug)

See pugjs/pug#1975