Failure to build package shouldn't stop other packages from being installed?

[hadley]

✓ Cached copy of BrailleR-deps 0.31.3 (source) is the latest build
✓ Got extrafontdb 1.0 (macos) (6.80 kB)                                         
✓ Got extrafont 0.17 (macos) (50.06 kB)                                         
✓ Got Rttf2pt1 1.3.8 (macos) (106.26 kB)                                        
✓ Got roloc 0.1-1 (macos) (123.31 kB)                                           
✓ Got nortest 1.0-4 (macos) (36.32 kB)                                          
✓ Got devtools 2.4.0 (macos) (392.41 kB)                                        
✓ Got moments 0.14 (macos) (54.27 kB)                                           
✓ Got tidyselect 1.1.1 (macos) (198.80 kB)                                      
✓ Got mathjaxr 1.4-0 (macos) (817.56 kB)                                        
✓ Got lattice 0.20-44 (source) (399.26 kB)                                      
✓ Got gridSVG 1.7-2 (macos) (1.20 MB)                                           
✓ Got installr 0.22.0 (source) (146.54 kB)                                      
✓ Got MASS 7.3-54 (source) (506.25 kB)                                          
✓ Got reticulate 1.20 (source) (832.95 kB)                                      
✓ Got colorspace 2.0-1 (source) (2.19 MB)                                       
✓ Got mgcv 1.8-35 (macos) (3.51 MB)                                             
✓ Got XML 3.99-0.6 (macos) (3.75 MB)                                            
✓ Got rolocISCCNBS 0.1 (macos) (4.95 MB)                                        
✓ Downloaded 18 packages (19.27 MB)in 4.7s                                    
ℹ Building colorspace 2.0-1
ℹ Building installr 0.22.0
ℹ Building lattice 0.20-44
ℹ Building MASS 7.3-54
x Failed to build installr 0.22.0   

And none of the other packages are installed — maybe it's ok for suggested packages to fail to install?

[jimhester]

Maybe as an option, but I don't think it would be a good idea to have non-failure be the default.

[gaborcsardi]

Yeah, it would have to be optional in general. OTOH, when installing a windows-only package on Unix fails, that should just work by default, maybe with a message.

[gaborcsardi]

The plan is to use the ? notation to tell pak to ignore a failure in the dependencies, e.g.

pak::pak(c(".", "badpackage?ignorefailure"))

?ignorefailure ignores that package if it fails to download, install, etc.

There will be also ?ignore, ignores it completely, does not even try to download, etc.