Speed up

[lulco]

• Create separate dir for each run (we need this because separate phpstan process would check all templates in phpstan-latte dir, also the old one) (All compiled templates from one run will be stored in one directory within tmpDir #427)
• Compile templates first, then analyse them (Prepared for another way of analysis #428)
• Try to use external phpstan process (first as feature flag - without it we will still use actual processing, I don’t know how to cover all use cases so this should be some string where user can define path to phpstan with config etc, we will add path to compiled templates dir and output format - json) (Added feature phpstanCommand to analyse compiled templates via separated command #429)
• Add fast check method for node visitors to check if they should run (e.g. no form / input in latte means we don't need to run form node visitors)
• Try to use PHPStan cache for repeated runs

[MartinMystikJonas]

It would be ideal to have same dir for following runs for same project to make use of phpstan results cache. But we need to figure out how to prevent overwriting identical templates (so phostan would not thinknit changed) and how to clean up old templates.

Maybe first compile to some tmp dir and then compare this tmp dir with analysed dir?

[lulco]

Complete? Great :)

I'm ready to sacrifice our internal compiled template cache mechanism for the sake of speed up. I'm cleaning the phpstan latte dir before almost each run anyway

[lulco]

Question is: “what is slowest part of this extension?”
My candidates:

• non-paralell analysis (phpstan a analyse in one thread)
• compiling templates
• Adjusting compiled templates (node visitors)

[MartinMystikJonas]

Sorry I put unlocked phone to my pocket 😀

My main candidate is latte context collection for template resolvers. But it is also hardest thing to speed up significantly.

[lulco]

I did some simple measurements:
Getting context is slow but not slowest part. Two slowest parts are: "compiling + adjusting templates" and "analysing files".

Sometimes analysing files is the slowest, sometimes adjusting templates is the slowest. I'm not sure if it depends on size of template or something else.

We can look at node visitors, but there is not much things to approve if we want to have testable compiled templates, but we can try to run phpstan via "exec"-like command. We will see

[MartinMystikJonas]

Idea for adjusting templates: Maybe visitors can specify some fast checks (like template/comoiled template contains given string/regex) and run only if it is true.

For example it does not make sense to run forms related visitors for templates without any form/input tags.

[lulco]

Idea for adjusting templates: Maybe visitors can specify some fast checks (like template/comoiled template contains given string/regex) and run only if it is true.

For example it does not make sense to run forms related visitors for templates without any form/input tags.

I like the idea for fast checks, just don't like the implementation with regexes, but maybe it is best thing we can do :)

[lulco]

I've achieved 12% speed up on my test app with running phpstan on compiled templates via shell_exec. I have to admit it is lower then my expectations :D Tomorrow I'll push it and you can try it too.