DeriNet._roots is not updated

[vidraj]

During database load, the field DeriNet._roots is initialized to contain a list of internal IDs of all cluster roots. But as far as I can see, it is not updated when adding derivations, so the information goes out of date quickly as clusters are merged or split. The information is used when saving the database, possibly resulting in a broken file.

I didn't actually test any of this, I just noticed it reading the source. If I'm wrong, correct me, please.

[vidraj]

Commit 2974338 possibly fixes a part of this issue, that the roots weren't correctly created on DB load and when adding lexemes. An internal ID is now added to ._roots when the new node doesn't have a parent, i.e. is a root.

I wrote tests for both this and the not-updated-when-adding-derivations issue.

[vojtsek]

I followed with commit e3e89bb. It changes _roots representation to set to achieve efficient removal. Changes in the roots that are caused by derivations should be hnadled now. I leave this issue open for now till it will be confirmed.

[vidraj]

The roots should probably be updated in sort() as well.

[vojtsek]

You are right @vidraj, thanks. It should be easy to add, I will also create a test for that.

[vidraj]

Thank you for solving this. :-)