Is caching not working as expected?

[IndrajeetPatil]

I was expecting that the second call here would be faster because it would use cache, but that's not what I see (at least locally).

library(lintr)

local_dir <- "/Users/indrajeetpatil/Documents/GitHub/lintr/R"

# the same behaviour is also seen with `lint_package()`
lint_dir(local_dir, cache = TRUE)
lint_dir(local_dir)                # I'd have expected this to be quick 

Either I am doing something wrong, or the caching is not working as expected.

[IndrajeetPatil]

> system.time(lintr::lint_dir("~/Documents/GitHub/lintr/R", cache = TRUE))
  |====================================================================================================================| 100%
   user  system elapsed 
 49.759   1.422  52.014 
> system.time(lintr::lint_dir("~/Documents/GitHub/lintr/R"))
  |====================================================================================================================| 100%
   user  system elapsed 
 58.436   1.401  61.078 

[AshesITR]

Does it work when specifying cache = TRUE both times?

[MichaelChirico]

This is WAI IMO, cache=FALSE should never try and read an existing cache (which may have been invalidated):

lintr/R/cache.R

Line 28 in f2da882

character()

Possibly, we could consider allowing cache=NULL to mean "use cache if found at default path", and further consider making that the default, but the current behavior for FALSE is correct.

[IndrajeetPatil]

Okay, I clearly have an incorrect mental model of {lintr}’s caching algorithm. 

As a user, here is how I thought it worked:

• step-1: run linter with cache = FALSE and realize it takes a while
• step-2: run linter with cache = TRUE and assume that unless the source code files, {lintr} package version, base-R version, etc. change, or I run cache_clear(), {lintr} is going to use this cache
• step-3: rerun linter without specifying the cache argument and expect it to run faster because it’s using previously generated cache 

Admittedly, this UX expectation comes from how caching works in {styler}, and I find it to be much more intuitive than what’s going on in {lintr}.

But I am happy to be challenged on that point :) 

[AshesITR]

I'm happy with revisiting the cache UX.

[MichaelChirico]

Noting: if cache = 'my_dir', I don't know that we have any way to recover that in the nest invocation. You'll have to keep calling with cache='my_dir'.

Anyone, +1 to revisiting. We don't have a cache_clear() function now, is important context for my reasoning. So if we add one, that changes things.