[varLib.interpolatable] minimum_spanning_tree is doing something bad

[simoncozens]

When varLib.interpolatable is run on the attached font, it finds a kink in glyph two, except when scipy.sparse.csgraph is installed.

Log prints for find_parents_and_order without scipy:

Parents: [None, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34]
Order: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35]

With scipy:

Parents: [None, 0, 0, 0, 0, 0, 0, np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(2), np.int32(2), np.int32(2), np.int32(2), np.int32(3), np.int32(3), np.int32(4), np.int32(4), np.int32(7), np.int32(7), np.int32(7), np.int32(7), np.int32(8), np.int32(8), np.int32(9), np.int32(9), np.int32(12), np.int32(12), np.int32(13), np.int32(13), np.int32(20), np.int32(20), np.int32(21), np.int32(21)]
Order: [0, np.int32(1), np.int32(2), np.int32(3), np.int32(4), np.int32(5), np.int32(6), np.int32(7), np.int32(8), np.int32(9), np.int32(10), np.int32(11), np.int32(12), np.int32(13), np.int32(14), np.int32(15), np.int32(16), np.int32(17), np.int32(18), np.int32(19), np.int32(20), np.int32(21), np.int32(22), np.int32(23), np.int32(24), np.int32(25), np.int32(26), np.int32(27), np.int32(28), np.int32(29), np.int32(30), np.int32(31), np.int32(32), np.int32(33), np.int32(34), np.int32(35)]

Additionally these values are not JSON-serializable so --json doesn't work when scipy is installed.

[simoncozens]

(The kink is correct, so we want to find it; but with scipy installed there are a bunch of underweight false positives.)

[behdad]

Parents: [None, 0, 0, 0, 0, 0, 0, np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(2), np.int32(2), np.int32(2), np.int32(2), np.int32(3), np.int32(3), np.int32(4), np.int32(4), np.int32(7), np.int32(7), np.int32(7), np.int32(7), np.int32(8), np.int32(8), np.int32(9), np.int32(9), np.int32(12), np.int32(12), np.int32(13), np.int32(13), np.int32(20), np.int32(20), np.int32(21), np.int32(21)]

Thanks. I can reproduce the 0s, but not the np.int32 part.

[behdad]

Parents: [None, 0, 0, 0, 0, 0, 0, np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(1), np.int32(2), np.int32(2), np.int32(2), np.int32(2), np.int32(3), np.int32(3), np.int32(4), np.int32(4), np.int32(7), np.int32(7), np.int32(7), np.int32(7), np.int32(8), np.int32(8), np.int32(9), np.int32(9), np.int32(12), np.int32(12), np.int32(13), np.int32(13), np.int32(20), np.int32(20), np.int32(21), np.int32(21)]

Thanks. I can reproduce the 0s, but not the np.int32 part.

Wasn't that part fixed by https://github.com/fonttools/fonttools/commit/a531041f3e1c6ae195485bbace3750630646633a

[behdad]

minimum_spanning_tree is not even called when scipy is installed for me.

[behdad]

Sorry, my bad. Ignore me for now.

[behdad]

Can you do some debugging? The different parent order is expected. But that the kink disappears is unfortunate. But for now, the numpy issue is baffling me.