from_ms should do something useful with replicates

[grahamgower]

Currently, from_ms() just converts the first replicate. It might be nice to instead return an iterator over the tree sequences, like when one calls msprime.simulate() with num_replicates>1.

$ mspms 3 2 -T -r 5 6 -seeds 1 2 3
/home/grg/.local/bin/mspms 3 2 -T -r 5 6 -seeds 1 2 3
1 2 3

//
[1](3:0.620,(1:0.150,2:0.150):0.470);
[1](3:2.573,(1:0.150,2:0.150):2.423);
[1](3:0.910,(1:0.150,2:0.150):0.760);
[3](3:0.702,(1:0.150,2:0.150):0.552);

//
[1](1:1.195,(2:0.220,3:0.220):0.975);
[1](1:1.791,(2:0.220,3:0.220):1.572);
[1](1:1.205,(2:0.220,3:0.220):0.986);
[1](2:1.205,(1:0.555,3:0.555):0.651);
[1](2:0.621,(1:0.555,3:0.555):0.066);
[1](1:0.555,(2:0.518,3:0.518):0.036);

$ mspms 3 2 -T -r 5 6 -seeds 1 2 3 | python -c "import tsconvert, sys; ts =
tsconvert.from_ms(sys.stdin.read()); print(ts.num_trees)"
4

[jeromekelleher]

Actually looking through the current implementation of from_ms, it's not really fit for purpose at the moment and needs a full rewrite to use the newick library rather than dendropy, and to harden it against various things (like this). Probably best to avoid using it for now and to maybe reuse some code from msprime's scrm handling code in verification.py for the comparisons in Demes.

[grahamgower]

I ended up splitting the ms output at // and passing the separate strings into from_ms. A bit hacky, but no big deal. Having the tree sequence is quite an advantage for, e.g. getting the SFS, compared with parsing out stats manually like the scrm verification code.

[benjeffery]

I agree that the from_ms code needs a rewrite to use the common, modular newick parser. Adding this to the milestone.