gcv

gcv (graph file converter) is a command-line utility distributed with Scotch, the graph partitioning, static mapping, and sparse-matrix ordering package developed by François Pellegrini and colleagues at LaBRI and INRIA (Bordeaux). Its role is narrow but useful: it translates graph descriptions between Scotch's native adjacency format (Scotch, the .grf files consumed by tools like gpart, gmap, and gord) and several external representations, so that graphs produced elsewhere can be fed into Scotch's partitioning and ordering pipeline, or exported from it. The program was formerly named scv.

The converter reads one input graph and writes one output graph, selecting formats via -i and -o flags. Supported keywords are Chaco/MeTiS adjacency format (Chaco, an unweighted or weighted adjacency listing also used by METIS), Matrix Market (Matrix Market), Harwell-Boeing (Harwell-Boeing), and Scotch's own format, which is the default output. Because Harwell-Boeing and Matrix Market are sparse-matrix formats, gcv treats their nonzero patterns as edges and symmetrizes any unsymmetric structure, reflecting the fact that Scotch's algorithms operate on undirected graphs. When source data carries vertex coordinates, an optional geometry file (.xyz) can be emitted alongside the topology; purely combinatorial formats produce no geometry.

Ergonomically, gcv follows Unix conventions: omitted filenames or a bare dash '-' route data through standard input and output, allowing it to sit in a pipeline. If the supporting libraries are compiled in, it transparently reads and writes compressed streams keyed off the filename suffix, supporting bzip2 (.bz2) and gzip (.gz) for both directions and lzma (.lzma) on input only.

Its strengths are simplicity, tight integration with the Scotch/PT-Scotch ecosystem, and a focus on the sparse-matrix and mesh formats common in scientific computing and finite-element work. Its limitations are equally clear: it is not a general graph-interchange tool. It ignores directionality, has no notion of rich attributes, labels, or property-graph semantics, and does not touch mainstream graph-visualization or exchange formats such as GraphML, GML, or DOT. For those broader conversions a dedicated multi-format converter such as GraphInOut or a scripting library like NetworkX is more appropriate; gcv is best understood as the format bridge into and out of Scotch itself.