GitLab

Otherwise we can't install the binary wheel and need 'gcc' and other tools to
install mypy.

Update hypothesis to 5.5.2+.

See merge request !4

Commit cf2b0e44 remove the double-max referred
to by the comment.  So the list is not longer needed.

Don't check the xotl namespace's __init__ with mypy.

See merge request !3

Those are really non-interesting pieces of code.  The only thing worth
mentioning is the change in Registry model: now we require the timestamp.

Still we don't reach 100%.  But hypothesis may choose to generate data that
would touch part of the code.  For instance, in two different runs I get:

Name                             Stmts   Miss  Cover   Missing
--------------------------------------------------------------------------------------------
xotl/crdt/register.py               60     11    82%   46, 66-67, 72-73, 78-79, 84-87, 108
xotl/crdt/testing/registers.py      57      7    88%   87-93
--------------------------------------------------------------------------------------------
TOTAL                              579     18    97%

Name                             Stmts   Miss  Cover   Missing
--------------------------------------------------------------------------------------------
xotl/crdt/register.py               60     11    82%   46, 66-67, 72-73, 78-79, 84-87, 108
xotl/crdt/testing/registers.py      57      7    88%   87-93
xotl/crdt/testing/sets.py          104      6    94%   232-237
--------------------------------------------------------------------------------------------
TOTAL                              579     24    96%

This happens because the assumption in xotl/crdt/testing/sets.py:231 was never
met in the second run.

Therefore, I'm going to settle with 95% coverage.

Dockerize CI pipeline.

See merge request !2

We split the items into *newly* created items and *previously added*.  The
assumption in add_item is left for a safe-guard (but current statistics show
no invalid case).

This change doesn't diminish our testing capabilities in remove_item:

- Hypothesis still generates removes in a replica that does have the item.

- Hypothesis generates reset_all_replicas after adding items.

For the comment in the code:

   ... we must ensure every process present in `self` is also present in
   `other` with the same counter ...

if follows this is just the same as comparing the lists stripping all 0
counters.

It was feeding the items *unwrapped* to the internal USet.  The test
'reset_all_replicas_with_item' now checks that this won't happen again (for
any set).

This was not caught by our tests; there are two possibilities:

- The tests are not good enough

- The condition is not needed.