`undefined subclass` warning when `load_all`-ing under some conditions

brookslogan commented 1 year ago

Summary:

Trying to load epipredict locally gave me a weird warning:
```
Loading required package: parsnip
Warning message:
In .recacheSubclasses(def@className, def, env) :
undefined subclass "numericVector" of class "Mnumeric"; definition not updated
```
- With options(warn=2L), instead of warning, it asked to install quantreg.
- Installing quantreg appeared to address the warning. (Technically, it's the installation of its MatrixModels dependency that appeared to actually address the problem, but since epipredict imports from quantreg, might as well say to just install quantreg. [Except actually it was effectively an upgrade of MatrixModels from 0.5-0 to 0.5-1 that resolved the issue.])
- I can't reproduce after fixing, despite trying to rewind with renv. (--- Removing quantreg and MatrixModels, likely from system libraries rather than the renv, allows reproduction. I'm not sure if there's a way to make more self-contained renvs that wouldn't pull from system libraries.)

Comments:

Perhaps part of this could be due to some missed detail in the quantile regression engine or parsnip's design?
At least part of this seems due to confusing/suboptimal behavior of devtools and/or renv.
I don't know if this could also happen with user installs via devtools::install_github or not.
Maybe no action is required, but leaving this Issue around for others to be able to google could be helpful, to see that installing quantreg could fix. (Is "Discussions" or "Wiki" a more appropriate place?)

Exploration:

Trying to load epipredict gives a weird message:

loading local version with 0ae1e401e82a9d93f72e736cfb237aeab4ded395 checked out
plus a pre-existing renv (renv.lock.1.as.txt) from before quantreg dependency was added
maybe other stuff invading the renv from outside, e.g., a MatrixModels_0.5-0 from regular libraries (Matrix in the lockfile was 1.5-3 and on system was 1.4-1, so this doesn't seem related)

devtools::load_all(export_all = FALSE)
#> ℹ Loading epipredict
#> Loading required package: epiprocess
#> 
#> Attaching package: ‘epiprocess’
#> 
#> The following object is masked from ‘package:stats’:
#> 
#>     filter
#> 
#> Loading required package: parsnip
#> Warning message:
#> In .recacheSubclasses(def@className, def, env) :
#>   undefined subclass "numericVector" of class "Mnumeric"; definition not updated

Fresh session, same(!!) renv; somehow options(warn=2L) alters the operation!

devtools::load_all(export_all = FALSE)
#> ℹ Loading epipredict
#> Loading required package: epiprocess
#> 
#> Attaching package: ‘epiprocess’
#> 
#> The following object is masked from ‘package:stats’:
#> 
#>     filter
#> 
#> Loading required package: parsnip
#> ℹ The package `quantreg` is required.
#> ✖ Would you like to install it?
#> 
#> 1: Yes
#> 2: No

Fresh session, after saying Yes above (--> renv.lock.2.as.txt): success! No warning.

devtools::load_all(export_all = FALSE)
#> 
#> ℹ Loading epipredict
#> Loading required package: epiprocess
#> 
#> Attaching package: ‘epiprocess’
#> 
#> The following object is masked from ‘package:stats’:
#> 
#>     filter
#> 
#> Loading required package: parsnip

Fresh session, after reverting the renv.lock back to its initial state and renv::restore()-ing: also no warning somehow?! (though renv said something suspicious like * The library is already synchronized with the lockfile. when I tried to use it to do the reversion)

devtools::load_all('~/files/tooling/epipredict')
#> ℹ Loading epipredict
#> Loading required package: epiprocess
#> 
#> Attaching package: ‘epiprocess’
#> 
#> The following object is masked from ‘package:stats’:
#> 
#>     filter
#> 
#> Loading required package: parsnip
sessionInfo() # a bit too late as I can no longer reproduce
#> R version 4.1.3 (2022-03-10)
#> Platform: x86_64-redhat-linux-gnu (64-bit)
#> Running under: Fedora Linux 36 (Workstation Edition)
#> 
#> Matrix products: default
#> BLAS/LAPACK: /usr/lib64/libflexiblas.so.3.2
#> 
#> locale:
#>  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
#>  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
#>  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices datasets  utils     methods   base     
#> 
#> other attached packages:
#> [1] epipredict_0.0.3.9999 testthat_3.1.4        parsnip_1.0.3        
#> [4] epiprocess_0.5.0.9999
#> 
#> loaded via a namespace (and not attached):
#>  [1] fs_1.5.2             usethis_2.1.6        lubridate_1.9.0     
#>  [4] devtools_2.4.4       rprojroot_2.0.3      tools_4.1.3         
#>  [7] profvis_0.3.7        utf8_1.2.2           R6_2.5.1            
#> [10] rpart_4.1.16         DBI_1.1.3            colorspace_2.0-3    
#> [13] nnet_7.3-17          urlchecker_1.0.1     withr_2.5.0         
#> [16] tidyselect_1.2.0     prettyunits_1.1.1    processx_3.7.0      
#> [19] compiler_4.1.3       cli_3.5.0            quantreg_5.94       
#> [22] SparseM_1.81         desc_1.4.2           scales_1.2.1        
#> [25] callr_3.7.1          stringr_1.4.1        digest_0.6.30       
#> [28] tsibble_1.1.1        pkgconfig_2.0.3      htmltools_0.5.3     
#> [31] parallelly_1.32.1    sessioninfo_1.2.2    fastmap_1.1.0       
#> [34] htmlwidgets_1.5.4    rlang_1.0.6          rstudioapi_0.13     
#> [37] shiny_1.7.2          farver_2.1.1         generics_0.1.3      
#> [40] dplyr_1.0.10         distributional_0.3.0 magrittr_2.0.3      
#> [43] Matrix_1.5-3         Rcpp_1.0.9           munsell_0.5.0       
#> [46] fansi_1.0.3          lifecycle_1.0.3      stringi_1.7.8       
#> [49] brio_1.1.3           MASS_7.3-58.1        pkgbuild_1.3.1      
#> [52] recipes_1.0.1        grid_4.1.3           parallel_4.1.3      
#> [55] listenv_0.8.0        promises_1.2.0.1     crayon_1.5.2        
#> [58] miniUI_0.1.1.1       lattice_0.20-45      splines_4.1.3       
#> [61] anytime_0.3.9        ps_1.7.1             pillar_1.8.1        
#> [64] future.apply_1.9.0   codetools_0.2-18     pkgload_1.3.1       
#> [67] glue_1.6.2           data.table_1.14.6    remotes_2.4.2       
#> [70] renv_0.15.5          vctrs_0.5.1          httpuv_1.6.5        
#> [73] MatrixModels_0.5-1   gtable_0.3.1         purrr_1.0.0         
#> [76] tidyr_1.2.1          future_1.27.0        assertthat_0.2.1    
#> [79] cachem_1.0.6         ggplot2_3.4.0        gower_1.0.0         
#> [82] mime_0.12            prodlim_2019.11.13   xtable_1.8-4        
#> [85] later_1.3.0          class_7.3-20         survival_3.2-13     
#> [88] timeDate_4021.104    tibble_3.1.8         memoise_2.0.1       
#> [91] hardhat_1.2.0        lava_1.6.10          workflows_1.0.0     
#> [94] timechange_0.1.1     globals_0.16.2       ellipsis_0.3.2      
#> [97] ipred_0.9-13

[Erroneously copy-pasted content removed.]

dajmcdon commented 1 year ago

Possible fixes may be updating R or parsnip. Both are behind mine by a bit, and I've never seen this warning. Maybe a linux thing?

brookslogan commented 1 year ago

Upgrading parsnip did not help. I didn't try upgrading R. Key problem seems to have been using some MatrixModels version 0.5-0 tucked away somewhere; upgrading to version 0.5-1 resolves. I think there's probably some devtools and/or renv and/or quantreg/parsnip DESCRIPTION suboptimalities somewhere in the mix here. We can prevent with a simple update to epipredict's DESCRIPTION. Will PR.

cmu-delphi / epipredict

`undefined subclass` warning when `load_all`-ing under some conditions #157