test/ejs.js fails with ejs.0.8.4

[hpaulj]

test/ejs.js gives an error

1) class EjsEngine < Template Asset [.jst.ejs] #compile() "before all" hook:
 Uncaught TypeError: Object #<Object> has no method 'asset_data_uri'

This is with a newly installed mincer. It runs ok if I install the previous version of ejs.

npm install ejs@0.8.3

So evidently there is some incompatibility with the latest ejs (0.8.4).

[hpaulj]

As best I can tell this error is because the new ejs wraps the template buf in a function evaluation

var buf = [];
with (locals || {}) {
 buf.push('');__stack.lineno=1; this.id = 1 ; buf.push('\n<header id="', escape((__stack.lineno=2,  "header" +     this.id )), '">\n    <p>', escape((__stack.lineno=3,  this.text )), '\n  <img src="', escape((__stack.lineno=4,   this.asset_data_uri("header.jpg") )), '" title="beautiful header" />\n</header>');
}

is now

var buf = [];
with (locals || {}) { function(){
 buf.push('');__stack.lineno=1; this.id = 1 ; buf.push('\n<header id="', escape((__stack.lineno=2,  "header" + this.id )), '">\n    <p>', escape((__stack.lineno=3,  this.text )), '\n  <img src="', escape((__stack.lineno=4,  this.asset_data_uri("header.jpg") )), '" title="beautiful header" />\n</header>');
})();
}

It's the result of tj/ejs#79 which is supposed to "Allow named function in views"

[ixti]

Yeah, the problem came out of this change. But I have a strong believe that it is caused by lazySource not being playing nice with this change. So I'm gonna drop lazySource to avoid this. lazySource were introduced to allow engines generate different things (JST vs really pre-processing) but I have a feeling it can be resolved much easier.

[puzrin]

We decided, that lazy-loading feature implementation is not good, and makes too many problems with archtecture. So, it will be removed. I've disabled appropriate tests.

PS. First pass still failing, second is ok. To reproduce - remove ./test/assets folder before runing tests.

[hpaulj]

The problem isn't just the lazy-loading. A simple EJS.render(source, options) fails if the template is using this.
I can get this test to work by changing ejs/lib/ejs.js to produce

var buf = [];
with (locals || {}) {(function(){
 buf.push('');__stack.lineno=1; this.id = 1 ; buf.push('\n<header id="', escape((__stack.lineno=2,  "header" + this.id )), '">\n      <p>', escape((__stack.lineno=3,  this.text )), '\n  <img src="', escape((__stack.lineno=4,  this.asset_data_uri("header.jpg") )), '" title="beautiful header" />\n</header>');
}).call(this);
}

In other words, pass the scope through this new function layer.

I submitted tj/ejs#126 for this issue (6/24)

[hpaulj]

With this async compile, how are (or were) local variables supposed to be passed to the template? For example, if in this test, this.text was replaced with fig.text, how would fig get passed through? Or is this a moot issue with the demise of lazy-loading?

So far the best I can do is use a helper function.

[ixti]

Thanks in advance. Although lazySources are adding it's own problems, so I was able to remove some problems with cleaning them out - I have a better plan how to re-introduce ability to use EJS for both server-side preprocessing and for JST.

[ixti]

Hmm. I tend to think you're right. We pass something like this to EJS: { scope: context, locals: locals }. Locals are holding only helper functions, so if you'll add a helper named foo you'll be able to use it as foo() from the templates. Variables can be assigned to context: env.ContextClass.prototype.foobar = 123, but as far as I can see EJS use scope as this of rendered function...

[ixti]

Probably it worth to rework the way we prepare locals to respect properties, not only helper functions...

[puzrin]

Solved in master, because lazy-stuff removed