GroteGnoom

Maybe we should speed them up? I saw GraphDB read one character at a time, which is very slow. Reading in a megabyte and processing it should probably not take...

Setting the cache because of igraph_i_simplify_sorted_int_adjacency_vector_in_place is surprisingly hard. It's just a few `if`s, but very easy to get wrong. `loops == IGRAPH_LOOPS_TWICE && multiple == IGRAPH_NO_MULTIPLE` is also a...

> I would not update the cache from the lazy adjlist/inclist functions. The chance of something going wrong is too great. Setting the cache in the lazy_adjlist_get_real is only done...

I thought I would just reduce some code duplication. I don't think increasing the API surface is a big deal. Even when internals would change, maybe this output would stay...

The same function already existed in the test utilities.

> A slight concern I have is that this duplicates a lot of the code, effectively creating many more tests than we had before. This will blow up the running...

> Optionally, the example can check the info value to see if we got convergence. With an input of 1, errors are generated when there's a problem, so this check...

> Docs could be explicit about what the "first" eigenvalue is (the largest, I presume) Yes, that would be nice. But I can't find it in the original docs either....

Now the dgeevx example does not show the general way to print out the eigenvectors, but to do it right we need something like arpack_unpack_complex.

> Do you mean that you would like to introduce a function that converts the "compressed" representation that LAPACK returns into a proper complex matrix? Yes, that would be nice,...