Why is BiocManager emitting messages on package load?

[LTLA]

The change in 1fed528 is causing emission of annoying messages in all packages that Import BiocManager. For example, BiocStyle imports BiocManager, and this is causing all my vignettes and reports generated in an interactive session to contain this message:

Bioconductor version 3.13 (BiocManager 1.30.12), ?BiocManager::install for help

in every location where I have library(BiocStyle). I didn't bother to suppressPackageStartupMessages() because BiocStyle was not emitting any messages previously. One could argue that I should suppress these messages, but then I would say that I shouldn't have to see this message in the first place, especially if I'm not using BiocManager directly. Why the chattiness?

[hpages]

I just noticed that too earlier today after doing library(BiocStyle) on one of the build machines. Looks like the .onAttach hook was replaced by an .onLoad hook (see commit 1fed528). Not sure why.

As a general rule I don't think .onLoad hooks should print messages.

[LiNk-NY]

Any insights, Martin? @mtmorgan Thanks!

[mtmorgan]

The banner can be useful in bug reports, and the 'best practice' recommendation for use (e.g., BiocManager::version()) doesn't attach the package. So it made sense to have the message in .onLoad rather than .onAttach.

I'm not sure why a package would import BiocManager: package installation shouldn't be something a package attempts to 'do' for the user.

[LTLA]

BiocStyle imports BiocManager to get version(), so that Biocpkg() and friends can point to the correct URL.

[hpages]

Maybe best practice should be to explicitly library() the package before using it interactively? That's kind of the standard thing to do, for any package.

[mtmorgan]

I chose not to emphasize using library() because I wanted commands that the user sees to be fully resolved and therefore not subject to namespace collisions -- the natural interpretation of install(), version(), etc resolved with BiocManager::...() seem like a better choice than some name mangling bm_installl() etc.

I will revisit this so that version() etc returns an object that is a version and hence computable, but prints with information about package version etc. I will remove the startup message.

[HenrikBengtsson]

You could condition the message based on how the package was loaded by looking at sys.calls()[1]. For instance, show it if user called it via a BiocManager::..., library(BiocManager), ...

[mtmorgan]

Thanks @HenrikBengtsson; I shy away from that level of complexity.