- Always fold on the location bases.
- For a given base, reduce if the number of integer offsets is smaller than
  the plevel or the ilevel.

[tests] removes -journal-disable & -check options that are already defined in @PTEST_DEFAULT_OPTIONS@

- use of cannot is apparently preferred over can not
- hov box without formatting hints is useless. Besides, opening
  an inner box very late on the line is usually not a good idea.

Only the functions whose body is analyzed are counted as covered by the
analysis. The functions reached by the analysis but for which a builtin or the
specification is used are not considered analyzed anymore.

Summary: computes statistics only for analyzed function. Returns [None] for
other functions.

Valentin Perrelle authored 3 years ago and David Bühler committed 3 years ago

- Whole program statistics are aggregated from function by function
  statistics
- Also ordered the alarms output by count by category, except for
  the Other category

And makes these options visible.

The fix is correct, but it significantly changes the behavior of the history
partitioning on some case studies.
Keep it for later, in order to fully understand the consequences.

[Datatype.Option] defines a different order than [Stdlib.Option.compare].

Type ['state partition] is a map from partition keys to states.
Such maps are converted into lists to be propagated through statements.
Changes in [Key.compare] impact the list order.
The list order impacts the heuristics used to remove redundant states before
propagating them (see Partitioning_index): redundant states may be propagated
or not depending of the list order.
This impacts the number of states propagated at some statements, and thus the
consumed slevel, and the precision of the analysis afterwards.

This commit reverts the change in the comparison of keys only to minimize
changes in the behavior of the analysis partitioning.