drop in xg0 (take two)

[glennhickey]

This is a fresh stab at #2311, made much easier by all the handlefication efforts in the meantime.

As of this commit, it's compiling but not linking due to a couple trailing XGPath usages..

[glennhickey]

Having the libhandlegraph version specified in two places

• https://github.com/vgteam/libhandlegraph/tree/47791b556bb963d7f1922548cf512749e59dab86
• https://github.com/vgteam/xg/blob/64165b94e1b114d6e42341d1b2fd65a05418f816/CMakeLists.txt#L55

is a wee bit dangerous (a little mismatch gave me weird segfaults)

[glennhickey]

From the first failed bash tap test, it looks like creating an xg index is dropping a bunch of paths

vg construct -m 1000 -r small/xy.fa -v small/xy2.vcf.gz -R x -C -a 2> /dev/null | vg view -j - | sed s/_alt/alt/g | vg view -Jv - >w.vg
vg paths -v w.vg -L | wc -l
39
vg index w.vg -x w.xg
vg paths -x w.xg -L | wc -l
30

[ekg]

It might be that the specific code path that is used by vg wasn't tested. I had been building from GFA during testing.

…

On Fri, Aug 16, 2019, 19:36 Glenn Hickey ***@***.***> wrote: From the first failed bash tap test <https://github.com/vgteam/vg/blob/master/test/t/05_vg_find.t#L93-L96>, it looks like creating an xg index is dropping a bunch of paths vg construct -m 1000 -r small/xy.fa -v small/xy2.vcf.gz -R x -C -a 2> /dev/null | vg view -j - | sed s/_alt/alt/g | vg view -Jv - >w.vg vg paths -v w.vg -L | wc -l 39 vg index w.vg -x w.xg vg paths -x w.xg -L | wc -l 30 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2402?email_source=notifications&email_token=AABDQEOXAFXV7DUHNQ4NZY3QE3QQBA5CNFSM4IMKVXS2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4PHSEA#issuecomment-522090768>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AABDQEM3HH2YWRVT4FQTC73QE3QQBANCNFSM4IMKVXSQ> .

[glennhickey]

Yeah, this error seems to be coming from empty paths getting dropped by XG::from_path_handle_graph(). Pretty minor, I think (provided the format supports empty paths).

[glennhickey]

Most of these changes are now in master via #2409. All the test failures in this PR are therefore due to either

• bugs in new xg
• subtle build problems (My main suspect, as everything runs fine when I build locally)
• bugs in VGSet::to_xg() (only vg code change here that's not in master)

A future PR adding the new xg should only need to adjust the Makefile and submodules as well as the aforementioned VGSet function. Hopefully the third time is the charm. Anyone else wanna give it a shot? @adamnovak @ekg ?

[adamnovak]

I'm going to set up a PR that makes your other changes on top of this. We will see how that goes.