Inconsistent behavior on undefined object property reference

[mde]

Inconsistent behavior on undefined object property reference

[TimothyGu]

(BTW you don't need the eval())

Will fix.

[TimothyGu]

Technically this is a breaking change. @mde what do you think?

To fix this, I would need to make EJS output templates like one of these:

1. EJS 1 style:

   var buf = []
   [...]
   ;buf.push(data.test)
   [...]
   ;return buf.join('')

2. Ternary operator:

   var __output = ''
   [...]
   ;__output += (data.test) == null ? '' : data.test
   [...]
   ;return __output

I'll do some benchmark on these.

[TimothyGu]

ejs v1 comp x 24,390 ops/sec ±1.21% (95 runs sampled)
ejs v2 comp orig (__output) x 7,175 ops/sec ±0.57% (97 runs sampled)
ejs v2 comp buf x 7,111 ops/sec ±0.33% (101 runs sampled)
ejs v2 comp __output x 6,927 ops/sec ±0.39% (103 runs sampled)

ejs v1 x 344,302 ops/sec ±0.55% (100 runs sampled)
ejs v1 !_with x 1,560,845 ops/sec ±0.49% (96 runs sampled)
ejs v2 orig x 143,544 ops/sec ±0.52% (100 runs sampled)
ejs v2 orig !_with x 1,453,962 ops/sec ±0.67% (97 runs sampled)
ejs v2 buf x 192,506 ops/sec ±0.97% (100 runs sampled)
ejs v2 buf !_with x 1,328,785 ops/sec ±1.62% (92 runs sampled)
ejs v2 out x 131,761 ops/sec ±1.15% (94 runs sampled)
ejs v2 out !_with x 1,457,213 ops/sec ±3.15% (88 runs sampled)

  <% for (var i = 0; i < data.length; i++) { %>
    <% var objects = data[i] %>
    <h1><%= data[i].title %></h1>
    <% for (var j = 0; j < objects.length; j++) { %>
      <% var obj = objects[j] %>
      <% if (obj.passed) { %>
        <p>Passed: <%= obj.name %></p>
        <p>Value: <%= obj.value %></p>
      <% } else { %>
        <p>Failed: <%= obj.name %></p>
        <p>Value: <%= obj.value %></p>
      <% } %>
    <% } %>
  <% } %>

    { data: [
      {
        title: "Test title",
        objects: [
          {
            name: 'this',
            value: 21032,
            passed: false
          }, {
            name: 'test',
            value: 0,
            passed: true
          }
        ]
      }, {
        title: "Another data",
        objects: [
          {
            name: 'that',
            value: 1012021,
            passed: true
          }, {
            name: 'tset',
            value: -2,
            passed: false
          }
        ]
      }
    ]}

According to the principle of the least slow I think if we are going to make this change it should use var buf = [] (the original EJS <=1 approach). I am still surprised how much slower EJS 2 is compared to EJS 1. It might be because of EJS 2's length compared to EJS 1 (~550 vs ~370).

[TimothyGu]

Now I read more and more EJS 1 code, I really think it's fairly reusable as EJS v2. The OO stuff and expensive regex matching is really dragging the compilation speed down. EJS v1 depends purely on indexOf which, believe it or not, is fast.

[mde]

I seem to remember fixing this for escaped output a long time ago in my original codebase. We can do some rewriting for perf, but it's an easy fix for unescaped output too. I'll push a fix, and we can talk about rewrite and refactor in another GH issue or PR.

[mde]

Fixed in master in a09d026.