Performance Optimizations

[Shaun Patterson]

Currently working on optimizing some templates on a large website.  I found some interesting things.

I'm using precompiled templates and jinja2 compat mode

Some observations / questions 

1. Variables in loops not cached

{% for entry in entries %}

         {{ some.other.object.value }}

{% endfor %}    

 

is much slower than 

{% set value = some.other.object.value }}

{% for entry in entries %}

    {{ value }}

{% endfor %}

I've only just begun looking into the Nunjucks code but I suspect there is not much optimization going on in the compile step.   Any object.key.value access seems to be a bit slow. 

2.  How are {% includes %} optimized?

Our codebase is heavily partialized, so we use a lot of {% includes... %} throughout the codebase.  We had a few macros like:

{% macro blah() %}

   {% include 'test.html' %}

{% endmacro %}

and I noticed that, if this macro was used in a loop for instance, that replacing the include with the contents of the file made the macro 25% faster or so.  I would imagine {% includes %} on a compiled template are just a function call to the compiled template but one function call doesn't seem to justify a 25% slowdown

3.  Try/Catch in the compiled code.  

In compiler.js (https://github.com/mozilla/nunjucks/blob/master/src/compiler.js)  emitFuncBegin and emitFuncEnd emit a try/catch block.  Doesn't this force the whole template to go into "non-optimized" mode in v8?  I'll be testing the speed with and without the try/catch block in the next few days when I have some time.  But has there been any work to investigate some of the try/catch strategies out there on the web?  Or is this non-issue? 

Any other well-known gotchas when writing templates?  The template rendering on a few of our pages is greater than 1 second.  After some simple optimizations (#1, #2) I've reduced the time by around 40% but I'm interested in other optimization tips.

Thanks

--

Shaun