feat: Avoid vertex and edge sequences in attributes

[krlmlr]

For #807.

[krlmlr]

This is branched off of v1.4.2, running revdepchecks now.

[krlmlr]

No problems at all!

https://github.com/igraph/rigraph/pull/808/files#diff-bbdda7f28391ee0ec001638e5603609582124345e0d65ebd4747cc592f9bd56dR1

I suggest we include this into the upcoming 1.4.3 release, to avoid problems like #807.

[szhorvat]

I suggest we include this into the upcoming 1.4.3 release, to avoid problems like #807.

Since 1.4.3 is published, how about a 1.4.4? If you do go ahead with a 1.4.4, can you please update the C core to the latest commit on this branch, to include an additional fix? The commit message explains the fix. https://github.com/igraph/igraph/tree/release/0.9

[szhorvat]

Ah, I see these commits were already included. I was confused by looking at the LTO checks on CRAN, but those are outdated, still showing information not even for 1.4.2, but for 1.4.1? Why would that be so @krlmlr, and is it something we should be concerned about?

https://www.stats.ox.ac.uk/pub/bdr/LTO/igraph.out

Notice this warning:

core/community/leading_eigenvector.c:156:12: warning: ‘igraph_i_community_leading_eigenvector2’ defined but not used [-Wunused-function]

It's clear from looking at line 156 of the mentioned source file that this warning is for version 1.4.1 or earlier.

[krlmlr]

CRAN keeps asking if LTO issues have been fixed, I keep telling them that we fix in igraph > 1.4.99 because of the C library update. Nothing to worry about for now.

[szhorvat]

These are not real issues. They are invalid warnings from GCC. As far as I'm concerned, this is a GCC bug.

I don't see any disadvantage-free way to make these GCC warnings go away.

[krlmlr]

I've seen CRAN in one instance accept LTO warnings as GCC bugs, we'll see.

[krlmlr]

Need to double-check if this is really needed or if perhaps there might be other ways to resolve this.

[szhorvat]

With igraph 0.10, which has been significantly cleaned up, it will be much easier to show that these are likely a GCC problem.

[krlmlr]

After reviewing `[.igraph.vs`(), I am now confident that this is a good solution.

@mbojan: We do support arbitrary vectors, but vertex and edge sets create an unhealthy back reference that our code isn't prepared to deal with.

[mbojan]

Thanks @krlmlr , that definitely makes sense to me. I already tried the version from the PR 👍