get_path_count() returns 0 when run on a GBZ, breaking vg paths

[faithokamoto]

@adamnovak here is the bug I mentioned in vg meeting today. It is in v1.71.0.

On December 27 I merged in #4780 which added a check to vg paths. If the input graph has no paths, the program now errors and refuses to continue. My logic here was that, as the command fundamentally operates on paths, a complete lack of paths will make it useless.

vg/src/subcommand/paths_main.cpp

Lines 534 to 536 in 010e9ed

	if (graph->get_path_count() == 0) {
	logger.error() << "graph does not contain any paths" << std::endl;
	}

I tested this on a few kinds of graphs and it looked fine. Notably, it passed CI tests. Today, I attempted to run it with a GBZ input. The graph definitely has paths. To my surprise, the command errored. To my bigger surprise, commenting out the error made it work fine. Which means that there are paths, just they aren't being detected by the get_path_count() function.

vg/src/vg.cpp

Lines 441 to 443 in 010e9ed

	size_t VG::get_path_count() const {
	return paths._paths.size();
	}

GBZ creation:

GRAPH=/private/groups/patenlab/fokamoto/centrolign/graph/unsampled/chr12

# PG format needs nodes with length <= 1024bp for distance indexing
vg convert --gfa-in $GRAPH.gfa | vg mod --chop 1024 - > $GRAPH.pg
# Convert to GBZ format by smooshing GBWT and PG together
vg gbwt --index-paths -x $GRAPH.pg -o $GRAPH.gbwt
vg gbwt --gbz-format -x $GRAPH.pg $GRAPH.gbwt -g $GRAPH.gbz

vg paths --list -x $GRAPH.gfa | wc -l # 373
vg paths --list -x $GRAPH.pg | wc -l # 373
vg paths --list -g $GRAPH.gbwt | wc -l # 373
vg paths --list -x $GRAPH.gbz | wc -l # error

Again, though, if I comment out the check on line 535 then vg paths works fine and the 373 paths are found.

[jltsiren]

GBZ and GBWTGraph only expose reference paths and generic paths through the PathHandleGraph interface. Haplotype paths are excluded, because a lot of older code assumes that it's perfectly fine to spend time and/or memory proportional to the total length of the paths. That assumption often fails with human graphs with a large number of haplotypes.

Path iteration code in vg paths uses the more modern PathMetadata interface, which also exposes haplotype paths in GBZ / GBWTGraph.