Understanding `storage` in `spectrum()`

[maelle]

From #2058 the failure is

Error in `eigen_adjacency_impl(graph, algorithm = algorithm, which = which, 
    options = options)`: argument "storage" is missing, with no default
Backtrace:
    ▆
 1. └─igraph::spectrum(g, which = list(howmany = 4, pos = "LA")) at [test-centrality.R:690:3](vscode-file://vscode-app/usr/share/positron/resources/app/out/vs/code/electron-browser/workbench/workbench.html#)
 2.   └─igraph:::eigen_adjacency_impl(...) at [rigraph/R/centrality.R:1247:3](vscode-file://vscode-app/usr/share/positron/resources/app/out/vs/code/electron-browser/workbench/workbench.html#)

eigen_adjacency_impl() has a storage argument. However, R_igraph_eigen_adjacency doesn't!

The YAML config from src has

igraph_eigen_adjacency:
    PARAMS: |-
        GRAPH graph, EIGENALGO algorithm=ARPACK,
        EIGENWHICH which=Default,
        INOUT ARPACK_OPTIONS options=ARPACK_DEFAULTS,
        INOUT ARPACK_STORAGE storage, OUT VECTOR values, OUT MATRIX vectors,
        OUT VECTOR_COMPLEX cmplxvalues, OUT MATRIX_COMPLEX cmplxvectors

I have two questions @szhorvat

• What is the storage argument of eigen adjacency? From the changelog, related to storage I only see "igraph_matrix_copy_to() gained an igraph_matrix_storage_t storage parameter that specifies whether the data should be written in column-major or row-major format."
• What could have gone wrong in the autogeneration for R_igraph_eigen_adjacency to not have the storage argument?

[szhorvat]

It has nothing to do with the storage argument of igraph_matrix_copy_to.

I expect igraph_eigen_adjacency just passes storage to igraph_arpack_rnsolve / igraph_arpack_rssolve. You can check the documentation of storage there. In short, this is for C memory allocation and is irrelevant for R. Just pass NULL in C.

[maelle]

Thanks! Do you mean I should somehow edit

igraph_eigen_adjacency:
    PARAMS: |-
        GRAPH graph, EIGENALGO algorithm=ARPACK,
        EIGENWHICH which=Default,
        INOUT ARPACK_OPTIONS options=ARPACK_DEFAULTS,
        INOUT ARPACK_STORAGE storage, OUT VECTOR values, OUT MATRIX vectors,
        OUT VECTOR_COMPLEX cmplxvalues, OUT MATRIX_COMPLEX cmplxvectors

so that the C code would pass NULL? I'm a bit confused as you can see.

[szhorvat]

Can you please try either OPTIONAL INOUT ARPACK_STORAGE storage or INOUT ARPACK_STORAGE storage=NULL? This is INOUT, not OUT, and I am not sure if this will work, and can't take the time to test right now ... got to run

[maelle]

I don't understand which YAML I should edit. The one that rigraph has from the core, or one of the rigraph specific ones?

[szhorvat]

The rigraph-specific one makes it possible to override the one from the C core on a per function basis. Please do not modify what you have in the vendored C core—those changes should be done upstream once we know what to do. Can you please try to make use of the override (copy and paste the spec from the C core functions.yaml) and see if either of the two suggestions I made work?