- callgraph: add idents used for instantiated nodes
- cgen : added Idents.enter_node
- cmain : removed error when simulated node does not exist (existence
of simulated node was tested for every program, comprising loaded ones)
Bug due to the fact that a variable can be "defined" and "read"
(in scheduling sense) by the same equation, without being a memory:
e.g., a clock defined as the result of a node application, together
with another result on this same clock.
Bug correction: basically removed the "assert false" on killed_vars,
decr_uses; do not count as "use" the self reads.
- We can do a better allocation if we take into
account 'when' in extvalues
(test/good/memalloc_clocks.ept shows the
improvement)
- Fixed a bug with memalloc on records: if we
translate:
o = { a with .f = u }
to
o = a; o.f = u
then we cannot share u and o.f
Added a "Contracts" pass, after inlining, taking care of the
contracts of the nodes called in the body of a node. This pass
"inlines" the code and assume/guarantee parts of these subcontracts.
The "Sigali" pass both generates the sigali ("z3z") code and add the call to
the controller (which is a node generated further by the sigali tool).
Therefore this pass has been included into the mls compiler, and removed
from the targets (a "z3z" dummy target has been kept for backward compatibility
reasons).
On some systems, the targets "all" and "install" must be
made with different rights. Then "all" cannot be a dependendy of
"install", especially since ocamlbuild try to read/modify some
files even if there is no compilation to perform.
The new rule accepts that some branches of a linearly typed merge have linear
type Top, provided that at least one has type "lin". E.g.:
node f(x : int at r) returns (o : int at r)
var ck : bool;
let
ck = true;
o = merge ck (x whenot ck) 0;
tel
is now deemed valid.
To know the variables read by an equation, we should
only look at the clock of variables. Otherwise,
there could be a problem for node calls that
define new clocks.
