Make the engine faster when nested for loops are used

[emilos]

todos: pure-engine x 6,225,365 ops/sec ±0.44% (87 runs sampled)
todos: underscore x 243,470 ops/sec ±0.29% (92 runs sampled)
todos: lodash x 285,776 ops/sec ±0.98% (91 runs sampled)
todos: handlebars x 793,441 ops/sec ±0.90% (90 runs sampled)
todos: mustache x 555,694 ops/sec ±0.48% (93 runs sampled)
Fastest is pure-engine
  ✔ benchmark: todos (30.7s)
friends: pure-engine x 1,683,728 ops/sec ±1.31% (88 runs sampled)
friends: underscore x 77,904 ops/sec ±0.74% (86 runs sampled)
friends: lodash x 98,544 ops/sec ±0.69% (89 runs sampled)
friends: handlebars x 803,258 ops/sec ±1.09% (90 runs sampled)
friends: mustache x 149,535 ops/sec ±1.22% (89 runs sampled)
Fastest is pure-engine
  ✔ benchmark: friends (30.5s)
if: pure-engine x 55,449,018 ops/sec ±1.55% (82 runs sampled)
if: underscore x 503,654 ops/sec ±0.95% (86 runs sampled)
if: lodash x 494,785 ops/sec ±1.22% (90 runs sampled)
if: handlebars x 841,275 ops/sec ±1.57% (83 runs sampled)
if: mustache x 737,372 ops/sec ±0.65% (92 runs sampled)
Fastest is pure-engine
  ✔ benchmark: if (30.4s)
projects: pure-engine x 1,873,031 ops/sec ±1.18% (84 runs sampled)
projects: underscore x 106,251 ops/sec ±0.31% (90 runs sampled)
projects: lodash x 141,294 ops/sec ±1.09% (91 runs sampled)
projects: handlebars x 721,065 ops/sec ±0.58% (87 runs sampled)
projects: mustache x 226,439 ops/sec ±0.98% (88 runs sampled)
Fastest is pure-engine
  ✔ benchmark: projects (30.8s)
search: pure-engine x 658,409 ops/sec ±1.37% (84 runs sampled)
search: underscore x 24,063 ops/sec ±0.96% (86 runs sampled)
search: lodash x 27,057 ops/sec ±1.00% (87 runs sampled)
search: handlebars x 767,058 ops/sec ±0.60% (85 runs sampled)
search: mustache x 103,704 ops/sec ±0.53% (92 runs sampled)
Fastest is handlebars
  ✔ benchmark: search (30.3s)

The engine is losing in the last benchmark, probably because we're not assigning variables in the best way (they're probably inlined right now, which is not ideal).

[zaiste]

@emilos I was just running this benchmark, trying to figure out the last test. What a coincidence! ;)

[emilos]

@zaiste 👍 the easiest way to debug this is to check what code is being generated by the compiler (basically console.log(template.toString())).

The output should be at least as good as if you were to type it by hand, if it's not then there's probably some performance bottleneck. Some other optimizations apply too of course, but in general less code is better.