jhsx/gotemplatebenchmark

comparing the performance of different template engines

goTemplateBenchmark

comparing the performance of different template engines

full featured template engines

• Ace
• Amber
• Go
• Handlebars
• Kasia
• Mustache
• Pongo2
• Soy

precompilation to Go code

• ego
• egon
• egonslinso
• ftmpl
• Gorazor
• Quicktemplate

transpiling to HTML

• Damsel

Why?

Just for fun. Go Templates work nice out of the box and should be used for rendering from a security point of view. If you really care about performance you should cache the rendered output.

on second thought: I have some templates that cannot be cached in my production code, thats why I'm interested in performant HTML generation using templates. After trying the code generation based projects I liked ego most, but some features where missing and generated code could be optimized further. That's why I created a fork and included the results in this benchmark.

Results

Changed the environment to my local dev laptop: i7-6700T 16GB Mem Golang: 1.7rc6

full featured template engines

go test -bench "k(Ace|Amber|Golang|Handlebars|Kasia|Mustache|Pongo2|Soy)$" -benchmem -benchtime=3s

Name	Runs	µs/op	B/op	allocations/op
Ace	300,000	16.232	5,608	77
Amber	1,000,000	5.537	1,929	39
Golang	1,000,000	5.392	1,904	38
Handlebars	500,000	13.094	4,260	90
Kasia	1,000,000	3.159	2,147	26
Mustache	1,000,000	4.148	1,569	28
Pongo2	1,000,000	4.289	3,302	46
Soy	1,000,000	3.083	1,863	26

precompilation to Go code

go test -bench "k(Ego|Egon|EgonSlinso|Quicktemplate|Ftmpl|Gorazor)$" -benchmem -benchtime=3s

Name	Runs	µs/op	B/op	allocations/op
Ego	5,000,000	1.032	914	8
Egon	2,000,000	2.274	827	22
EgonSlinso	10,000,000	0.541	828	0
Ftmpl	3,000,000	1.779	1,142	12
Gorazor	3,000,000	1.655	613	11
Quicktemplate	10,000,000	0.446	799	0

transpiling to HTML

I removed Damsel, because transpilation should just happen once at startup. If you cache the transpilation result, which is recommended, you would have the same performance numbers as html/template for rendering.

more complex test with template inheritance (if possible)

go test . -bench="Complex" -benchmem -benchtime=3s

Name	Runs	µs/op	B/op	allocations/op
ComplexEgo	1,000,000	4.899	2,561	41
ComplexEgoSlinso	2,000,000	2.338	2,070	7
ComplexEgon	500,000	10.148	3,521	101
ComplexFtmpl	500,000	8.639	5,300	48
ComplexFtmplInclude	500,000	8.269	5,300	48
ComplexGolang	100,000	43.787	12,415	295
ComplexGorazor	300,000	12.573	8,327	73
ComplexMustache	200,000	26.876	7,856	166
ComplexQuicktemplate	2,000,000	2.442	1,892	0

Security

All packages assume that template authors are trusted. If you allow custom templates you have to sanitize your user input e.g. bluemonday. Generally speaking I would suggest to sanitize every input not just HTML-input.

Framework	Security	Comment
Ace	No
amber	No
Damsel	Yes, if html/template is used for executing	Damsel transpiles to HTML
ego	Partial (html.EscapeString)	only HTML, others need to be called manually
egon	Partial (html.EscapeString)	only HTML, others need to be called manually
egonslinso	Partial (html.EscapeString)	only HTML, others need to be called manually
ftmpl	Partial (html.EscapeString)	only HTML, others need to be called manually
Go	Yes	contextual escaping html/template Security Model
Gorazor	Partial (template.HTMLEscapeString)	only HTML, others need to be called manually
Handlebars	Partial (raymond.escape)	only HTML
Kasia	Partial (kasia.WriteEscapedHtml)	only HTML
Mustache	Partial (template.HTMLEscape)	only HTML
Pongo2	Partial (pongo2.filterEscape, pongo2.filterEscapejs)	autoescape only escapes HTML, others could be implemented as pongo filters
Quicktemplate	Partial (html.EscapeString)	only HTML, others need to be called manually
Soy	Partial (template.HTMLEscapeString, url.QueryEscape, template.JSEscapeString)	autoescape only escapes HTML, contextual escaping is defined as a project goal