Tamás Nepusz

It seems promising, but FEAST does not seem to work on macOS (Apple M1) yet; I tried this in the `src/` folder: ```sh $ make ARCH=x64 F90=gfortran MKL=no feast gfortran...

FWIW I'm probably going to go for `IGRAPH_LOOPS_ONCE` and `IGRAPH_LOOPS_TWICE` unless there are objections, for sake of readability. (I mean, if we have `NO_LOOPS` instead of `LOOPS_0`, we should probably...

Well, speaking about efficiency: I think it is actually _more_ efficient to modify the graph in-place (if it is done right internally, i.e. not by copying the graph internally and...

The key question is still whether we want to repurpose the `mode` argument to control the multiplicity of loops in undirected graphs. On one hand, it's ugly because we are...

If we really want to solve this on the C level, I'd keep the format string as is, and wrap the index variable in a macro (i.e. `IGRAPH_INDEX(x)`. `IGRAPH_INDEX(x)` could...

Note that if we go down the `IGRAPH_INDEX` macro route, then distros that unbundle the C core of igraph from higher level interfaces and link to it separately will run...

Idk; if you see a function named `igraph_set_base_index()`, I think it's a reasonable expectation that it would work not only for error messages; otherwise we should call it `igraph_set_display_index_base_for_error_messages()` :)

With the C attribute interface, vertex attributes have an implicit default value. For numeric attributes, this is NaN. For boolean attributes, this is `false`. For strings, I think the default...

> It produces a new graph instead of modifying the old one. This seems to be a recurring topic (I remember a similar discussion about vertex contraction). Here I wholeheartedly...

We would also need an argument that tells igraph what to do with the attributes of contracted edges. One thing that I can imagine people would want to do is...