May I create a Template instance first and the use it to render context?

[childe]

In http://product.hubspot.com/blog/jinjava-a-jinja-for-your-java the author says speed of jinja2 in python is 1387.3108 runs/s which is a bit slower than jinjava.

But I compared the 2 projects:
jinja2 is much faster if a Template is prepared in advance.

python code:

from jinja2 import Template

n = 70000
event = {'name': "jia.liu"}
t = Template('Hello, {% if name is defined %} {{name}} {% else %} world {% endif %}')
msg = []
for i in range(n):
    msg.append(t.render(**event))

jinjava code:

Jinjava jinjava = new Jinjava();

Map<String, Object> context = new HashMap<>();      
context.put("name", "Jared");
String template = "Hello, {% if name is defined %} {{name}} {% else %} world {% endif %}";

for (int i = 0; i < 70000; i++) {

    jinjava.renderForResult(template, context);

}       
System.out.println(System.currentTimeMillis()-s);

I really hope i use jinjava in a wrong way. Is there any method I could first create a Template instance like jinja2?

[childe]

In my test running at macbook pro I5 2.6G, python code runs in 0.9s, and java 2.5s.

[jaredstehler]

You can create a parsed representation of a template using the interpreter instance, as follows:

    JinjavaInterpreter interpreter = new JinjavaInterpreter(jinjava, jinjava.getGlobalContext(), jinjava.getGlobalConfig());
    Node parsedTemplate = interpreter.parse(template);

    s = System.currentTimeMillis();

    for (int i = 0; i < 70000; i++) {
      interpreter.render(parsedTemplate, true);
    }

This actually runs in 1/3 of the time of your other java example, which is promising! 😄 I'm going to create a nicer factory method of getting a new interpreter instance from a Jinjava object.

[jaredstehler]

Results of updated benchmark code with precompiled (e6dbd9d):

Result: 148.433 ±(99.9%) 2.078 ops/s [Average]
  Statistics: (min, avg, max) = (77.945, 148.433, 158.897), stdev = 8.798
  Confidence interval (99.9%): [146.355, 150.511]


# Run complete. Total time: 00:27:09

Benchmark                                                Mode  Samples     Score     Error  Units
c.h.j.b.jinja2.Jinja2Benchmark.precompiledBenchmark     thrpt      200  9951.372 ± 203.122  ops/s
c.h.j.b.jinja2.Jinja2Benchmark.realWorldishBenchmark    thrpt      200  1160.294 ±  15.803  ops/s
c.h.j.b.liquid.LiquidBenchmark.parse                    thrpt      200  1657.029 ±  23.344  ops/s
c.h.j.b.liquid.LiquidBenchmark.parseAndRender           thrpt      200   148.433 ±   2.078  ops/s

[childe]

Yes, it is much faster than before~
But our java version is 1.7 unfortunately, so i have to use jinjava 1.0.9. Maybe that is why is runs 1/2 of the time of previous example.
Thank you all very much!

[childe]

before I could create Node , I need a interpreter , but context of the interpreter varies every time, so use below method is better:

        TokenParser t =  new TokenParser(null,template);
        Node parsedTemplate =TreeParser.parseTree(t);
        String s = interpreter.render(parsedTemplate);