Package dependencies of modules

[mschubert]

There is two possibilities to use packages/package functions inside a module:

library(mypackage)
#or
mypackage::method()

Disregarding that the latter is preferred to not shadow objects in .GlobalEnv, either call occur somewhere in a given module.

The question is: is there a good way to track package dependencies of modules, without having to look for these calls in each module file?

Ideally, I would like to see something that provides:

• a way to track required packages in a module (roxygen annotations?)
• an option in modules to automatically install these packages upon module loading

[klmr]

Related: #45 — in the long run, neither library nor :: should be used in modules.

I’m not sure I see the use in annotating required packages, to be honest. In principle we could use Roxygen for this, though. The same goes for dependent modules, of course.

I’m also not a big fan of automatically installing dependencies when loading modules. But the deployment mechanism outlined in #11 would take care of this, don’t you think?

[mschubert]

Ok, let's split this into two parts:

• keeping track of dependencies in a machine-readable manner
• automatically satisfying those dependencies

For the first part, it shouldn't matter how external packages and modules are loaded. - Just that they are required in order for the module to function properly, and the module needs to somehow "know" about that.

For the 2nd you are right that satisfying them makes more sense during install time as opposed to runtime (however, if you git clone manually, you won't get this benefit - packages always do this when calling R CMD INSTALL).

Maybe you thought of handling both when writing #11, but there is no explicit mention of whether or how dependencies are to be resolved (within modules or including packages).

[klmr]

#11 doesn’t mention these issues yet because I haven’t thought about them in detail. In fact, a packaging system may be a separate project rather than being bundled with modules (these are related, but ultimately distinct concerns).

That said, I definitely see the need for this. I just need to find the time: I want to avoid a Pythonesque FUBAR, where we have umpteen competing package managers.

[klmr]

Package loading — #45 — is now implemented. This doesn’t solve tracking package dependencies yet. However, it may actually help in doing so because a package manager could conceivably parse the module code, resolve all calls to import_package and thus auto-generate a list of package dependencies (and the same for import and modules).

[klmr]

This issue is now tracked in two parts in #11, #28 and #71.