adding .onAttach for smooth data loading

[benyamindsmith]

When igraphdata is presently loaded. The data does not look nice when the library is called without igraph. I added a zzz.R file and an .onAttach function.

Without .onAttach

With .onAttach

This is a fix:

Let me know if this is helpful or not.

[benyamindsmith]

I will update and resubmit.

[benyamindsmith]

The issue appears to be with macos. Not sure what needs to be done here.

[benyamindsmith]

@krlmlr - a gentle reminder about this :)

Curious if this is going to be merged or not.

[maelle]

Hello @benyamindsmith! Nice idea. Do you think users often load igraphdata without loading igraph first?

In any case this reminds me of #16

[maelle]

Curious to hear what @schochastics @krlmlr think

[schochastics]

I think it is save to assume that if you use igraphdata, you will also have igraph loaded. However, I still think this is reasonable to merge since there might be use cases I do not see atm

[benyamindsmith]

@maelle

I personally have been doing development/scripting in R with libname::funcname sort of syntax. So when I was running code before with igraph I often did not write library(igraph) on the top of my scripts and would just use igraph::funcname.

This is what motivated my PR.

Beyond my personal reason, I can't see why one would leave this out as using .onAttach ensures that the datasets are presented as intended.

[szhorvat]

I do not feel qualified to state an opinion here since I am an R novice.

But out of curiosity I tried to find other packages that do this and I could not. I looked here:

• https://www.bioconductor.org/packages/3.20/BiocViews.html#___ExperimentData
• https://cran.r-project.org/web/packages/available_packages_by_name.html

I do wonder why.

Is it not a convenience during interactive work to have the autoload? I often end up loading igraphdata without loading igraph first, just out of forgetfulness. But it isn't really an issue as it seems perfectly fine to load igraph after loading the data and still use the datasets.

[schochastics]

There are other packages that do this, just not with igraph (at least my quick search turned out empty). You can check on the github repo of CRAN:
https://github.com/search?q=org%3Acran%20%22.onAttach%20%3C-%20function(libname%2C%20pkgname)%22%20%20%22requireName%22&type=code

[krlmlr]

Thanks. I now see what the problem is -- igraph is imported but the import isn't actually used when igraphdata is loaded. One way to make sure loading happens is to add

#' @importFrom igraph V
NULL

or similar. This adds an entry to NAMESPACE which in turn ensures that igraph is loaded as soon as igraphdata is loaded. I suspect there is little need for using the igraphdata package without igraph.

[benyamindsmith]

@krlmlr could we get a sanity check on that to confirm that the @importFrom specification makes .onAttach irrelevant? A screenshot would be good enough.

[krlmlr]

.onAttach() is weaker than the namespace declaration. With the latter, the following code will work as expected in a fresh R session:

igraphdata::karate