dirstate: drop all logic around the "non-normal" sets
The dirstate has a lot of code to compute a set of all "non-normal" and
"from_other_parent" entries.
This is all used in one, unique, location, when `setparent` is called and moved
from a merge to a non merge. At that time, any "merge related" information has
to be dropped. This is mostly useful for command like `graft` or `shelve` that
move to a single-parent state -before- the commit. Otherwise the commit will
already have removed all traces of the merge information in the dirstate (e.g.
for a regular merges).
The bookkeeping for these sets is quite invasive. And it seems simpler to just
drop it and do the full computation in the single location where we actually
use it (since we have to do the computation at least once anyway).
This simplify the code a lot, and clarify why this kind of computation is
needed.
The possible drawback compared to the previous code are:
- if the operation happens in a loop, we will end up doing it multiple time,
- the C code to detect entry of interest have been dropped, for now. It will be
re-introduced later, with a processing code directly in C for even faster
operation.
Differential Revision: https://phab.mercurial-scm.org/D11507
$ hg init repo
$ cd repo
$ touch foo
$ hg ci -Am 'add foo'
adding foo
$ hg up -C null
0 files updated, 0 files merged, 1 files removed, 0 files unresolved
this should be stored as a delta against rev 0
$ echo foo bar baz > foo
$ hg ci -Am 'add foo again'
adding foo
created new head
$ hg debugindex foo
rev linkrev nodeid p1 p2
0 0 b80de5d13875 000000000000 000000000000
1 1 0376abec49b8 000000000000 000000000000
$ cd ..