New CRAN release

[mattansb]

deal with #648

[mattansb]

@strengejacke @IndrajeetPatil Any idea why some tests are still failing when effectsize no longer uses the effectsize_type argument anywhere?

[IndrajeetPatil]

You need to use parameters 0.22.0, which has renamed this parameter. Bump the version in DESCRIPTION and hopefully it goes away.

[strengejacke]

It's indeed strange, we have this warning only since parameters 0.22.0?

[strengejacke]

You're calling, e.g. from .anova_es.aov():

params <- parameters::model_parameters(model, verbose = verbose, effects = "fixed", es_type = NULL)

the .aov method from model_parameters has no effects argument, and this partially matches with the deprecated, but not yet removed(!) argument effectsize_type. The only way to fix this is to remove the effects argument in your call in .anova_es.aov() (and in the other functions)

[IndrajeetPatil]

Oh boy. Okay, I have changed some options to make sure that we always get warnings if we are using partial matching anywhere.

You should see new warnings from now on about this issue.

[strengejacke]

Our goal was to remove that partial matching by renaming that argument in parameters, although there were no warnings, so Mattan can use both effects and es_type in his effectsize package - and now we actually made things worse (at least temporary...) 🙄

[strengejacke]

@mattansb Do we need to set effects = "fixed" in the anova methods at all? That argument doesn't exist in parameters::model_parameters() for anova objects.

[mattansb]

We do :/ some non-ANOVA objects get passed down there... Is there a verbose = FALSE option to suppress that partial matching warning?

[strengejacke]

No, there's not. 😞 Maybe we wrap the call into suppressWarnings+)?

[IndrajeetPatil]

You can also change the options generating warnings about partial matching using on.exit() for the scope of the function where this warning originates.

[mattansb]

But I want to capture other warnings...

I think I'll have to use:

foo <- function(x) {
  if (x == 1) {
    warning("something something effectsize_type...")
  } else {
    warning("some other warning...!")
  }
  x + 1
}

foo(1)
#> Warning in foo(1): something something effectsize_type...
#> [1] 2

foo(2)
#> Warning in foo(2): some other warning...!
#> [1] 3

withCallingHandlers({
  foo(1)
}, warning = function(w) {
  if (grepl("effectsize_type...", conditionMessage(w))) {
    invokeRestart("muffleWarning")
  }
})
#> [1] 2

withCallingHandlers({
  foo(2)
}, warning = function(w) {
  if (grepl("effectsize_type...", conditionMessage(w))) {
    invokeRestart("muffleWarning")
  }
})
#> Warning in foo(2): some other warning...!
#> [1] 3

^{Created on 2024-06-24 with reprex v2.1.0}

[mattansb]

Okay, @strengejacke 's fix seems to work (and all this lints! Wow!) - but I remember that argument was there for a reason. I'm sure it will come up eventually (:

Thanks!

[strengejacke]

I'm sure it will come up eventually (:

I'd say we quickly remove the deprecation warning in parameters, and add back the effects argument then. Maybe we add a "TODO" as comment to the code?

[strengejacke]

but I remember that argument was there for a reason

I think I also read an issue about this, but cannot quite remember when and where.

[strengejacke]

Was it related to the report package? Or because we could have parameter effect sizes for mixed models, and therefore needed to set effects = "fixed"?

[mattansb]

No idea 🤷‍♂️ We test mixed models, so it's not it.

Anyway, submitted.

[mattansb]

Thanks, on its way to CRAN.