Improved Rendering Performance

[tdotclare]

This release substantially improves the rendering performance of Leaf on typical calls where the template is fully resolved and cached.

Performance Comparison

• Linear test - 10 flat templates, 1 million render calls against them
• Random test - 10 flat templates (layer 3), 20 templates referencing one of the flat templates (layer 2), 100 templates referencing two random layer 2 templates (layer 1). 1 million render calls against a random one of the 130 total templates.
• Each test was run 50 times on an 4GHz i7 Retina iMac

Linear

Branch	Min	Avg	Max	Avg Baseline	Avg CPU Time	CPU Baseline
1.0.0rc-1.13	4.27s	4.72s	5.14s	44.83%	1m 14s	40.72%
1.0.0rc-1.12	10.4s	10.52s	11.4s	100%	3m 1s	100%

Random

Branch	Min	Avg	Max	Avg Baseline	Avg CPU Time	CPU Baseline
1.0.0rc-1.13	4.43s	4.82s	5.17s	49.9%	1m 18s	45.3%
1.0.0rc-1.12	9s	9.66s	10.61s	100%	2m 51s	100%

NOTE this is purely a pipeline measurement - the templates used are lightweight and require near-zero time to serialize

[tdotclare]

Detailed Change Summary

This release substantially improves the typical rendering performance of Leaf on normal render calls where the AST storage adheres to SynchronousLeafCache

Internal Changes

SynchronousLeafCache adherents guarantee a non-future cache retrieval method if appropriate (eg, in-memory storage of the cache). This is currently an internal-only protocol that provides func retrieve(documentName: String) -> LeafAST? as a blocking retrieval method as an alternative to an EventLoopFuture<LeafAST?> return

In the typical use case where a call to LeafRenderer.render() references a template that is already present in the AST cache, is fully resolved, and is flat, this substantially improves render speed by skipping future handling.

Comparison to 1.0.0.rc-1.12 Performance

• Linear test - 10 flat templates, 1 million render calls against them
• Random test - 10 flat templates (layer 3), 20 templates referencing one of the flat templates (layer 2), 100 templates referencing two random layer 2 templates (layer 1). 1 million render calls against a random one of the 130 total templates.
• Each test was run 50 times on an 4GHz i7 Retina iMac
• Branches - 1.0.0rc-1.12 (current) and BlockingLeafCache

Linear

Branch	Min	Avg	Max	Avg Baseline	Avg CPU Time	CPU Baseline
1.0.0rc-1.12	10.4s	10.52s	11.4s	100%	3m 1s	100%
BlockingLeafCache	4.27s	4.72s	5.14s	44.83%	1m 14s	40.72%

Random

Branch	Min	Avg	Max	Avg Baseline	Avg CPU Time	CPU Baseline
1.0.0rc-1.12	9s	9.66s	10.61s	100%	2m 51s	100%
BlockingLeafCache	4.43s	4.82s	5.17s	49.9%	1m 18s	45.3%

NOTE this is purely a pipeline measurement - the templates themselves are lightweight and require next to no time to serialize

Other Notes

• LeafCache.load() has been renamed LeafCache.retrieve() for clarity.
• Certain default implementations and deprecated interfaces resulting from changes to LeafAST storage on LeafCache have been removed.

[tanner0101]

These changes are now available in 1.0.0-rc.1.13