check_if_installed() issue on R old-release?

[strengejacke]

@rempsyc @etiennebacher @IndrajeetPatil Not sure what's going on, it seems we have no issues on old-R for insight, however, in other packages, we have:

https://github.com/easystats/parameters/actions/runs/6151829558/job/16692707814

Error in mapply(FUN = f, ..., SIMPLIFY = FALSE): zero-length inputs cannot be mixed with those of non-zero length

The affected code seems to be this part:

insight/R/check_if_installed.R

Lines 73 to 80 in 7ee2d31

	} else if (!is.null(minimum_version)) {
	needs_update <- unlist(Map(function(p, m) {
	if (is.na(m)) {
	FALSE
	} else {
	utils::packageVersion(p) < package_version(m)
	}
	}, package, minimum_version))

My first guess was that package_version() causes issues with old-R, but then it would fail for insight as well, which is not happening. Another guess is that the function that checks for dependency-versions has some issues, but I'm not sure why?

insight/R/check_if_installed.R

Lines 154 to 167 in 7ee2d31

	.get_dep_version <- function(dep, pkg = utils::packageName()) {
	suggests_field <- utils::packageDescription(pkg, fields = "Suggests")
	suggests_list <- unlist(strsplit(suggests_field, ",", fixed = TRUE))
	out <- lapply(dep, function(x) {
	dep_string <- grep(x, suggests_list, value = TRUE, fixed = TRUE)
	dep_string <- dep_string[which.min(nchar(dep_string))]
	dep_string <- unlist(strsplit(dep_string, ">", fixed = TRUE))
	gsub("[^0-9.]+", "", dep_string[2])
	})
	if (all(is.na(out))) {
	out <- NULL
	}
	unlist(out)
	}

Any ideas?

[rempsyc]

Probably .get_dep_version is the issue, I'll look into it

[strengejacke]

See current changes I made. It seems that if no version was given in the description, an element of length 0 was returned, and Map() got package with a string element, and minimum_version was of length 0, that's why Map() failed.

This should be fixed now, but I wonder why tests did not fail for insight?

[rempsyc]

Weird thing is that it seems to work OK on macOS, Windows, and ubuntu release and oldrel, but not ubuntu oldrel-2 and oldrel-3. I'd like to troubleshoot this but I only have access to a Windows machine... What's the recommended approach in these scenarios?

In any case, it seems like the Map call that fails on line 74 uses the minimum_version object created through .get_dep_version, so it must not return a valid result in some situations like for parameters. The Map call checks whether m (the minimum version) is missing through is.na(). But if m is NULL, then length(is.na(NULL)) = 0 for some but not others for lists, which could explain zero-length inputs cannot be mixed with those of non-zero length?

[rempsyc]

Thanks, that should settle it. The only potential issue is that by setting minimum_version to NULL for old R releases, the DESCRIPTION file won't be checked and people may not have the correct version installed. However, before this automatic check, it was also the case that many packages with a minimum version in DESCRIPTION were not specified with a minimum version in check_if_installed, so for those edge cases it does not change much.

[strengejacke]

I think I already removed the check for R version, see https://github.com/easystats/insight/blob/main/R/check_if_installed.R

[strengejacke]

I guess this should be fixes (e.g., performance no longer fails). @easystats/maintainers should we submit a "hotfix" of insight to CRAN? Last update of that package was just a few days ago.

[IndrajeetPatil]

Yes, but it's a critical enough fix, CRAN won't complain at all. We just need to mention this in CRAN comments doc.

[strengejacke]

Which R versions are oldrel-2 and oldrel-3 (and oldrel)?

[IndrajeetPatil]

• oldrel-2: 4.1.3 (https://api.r-hub.io/rversions/r-oldrel/2)
• oldrel-3: 4.0.5 (https://api.r-hub.io/rversions/r-oldrel/3)

[strengejacke]

hm, it's weird that there's an issue in R 4.1, but not in R 4.2 for Map() when one vector is empty (this is what the error msg was indicating):

See https://github.com/easystats/parameters/actions/runs/6158994652/job/16712908066:

mapply(rep, times = 1:4, x = 4:1)
#> [[1]]
#> [1] 4
#> 
#> [[2]]
#> [1] 3 3
#> 
#> [[3]]
#> [1] 2 2 2
#> 
#> [[4]]
#> [1] 1 1 1 1
mapply(rep, times = 1:4, x = c())
#> list()

^{Created on 2023-09-13 with reprex v2.0.2}

See

Error in `mapply(FUN = f, ..., SIMPLIFY = FALSE)`: zero-length inputs cannot be mixed with those of non-zero length
Backtrace:
    ▆
 1. ├─parameters::standard_error(x) at test-posterior.R:47:2
 2. └─parameters:::standard_error.draws(x) at parameters/R/4_standard_error.R:63:2
 3.   ├─parameters:::.posterior_draws_to_df(model) at parameters/R/methods_posterior.R:49:2
 4.   └─parameters:::.posterior_draws_to_df.draws_list(model) at parameters/R/methods_posterior.R:75:2
 5.     └─insight::check_if_installed("posterior") at parameters/R/methods_posterior.R:87:2
 6.       ├─base::unlist(...) at insight/R/check_if_installed.R:79:4
 7.       └─base::Map(...) at insight/R/check_if_installed.R:79:4
 8.         └─base::mapply(FUN = f, ..., SIMPLIFY = FALSE)

Now, after making changes to check_if_installed(), we no longer see this issue? But I can't believe there was a change in mapply() between R 4.1 and R 4.2, so I wonder why this error only occurred for "old-rel".

[etiennebacher]

Could be due to https://bugs.r-project.org/show_bug.cgi?id=18164 (implemented in R 4.1.2)? Random guess, it's the only thing I saw in the NEWS of R