mirror of
https://github.com/Z3Prover/z3
synced 2025-06-27 08:28:44 +00:00
#7667 - add API documentation
The scenario works as expected when disabling model compression: ``` from z3 import * pre_processor = Tactic("bit-blast") solver = Solver() set_param("model.compact", False) x = BitVec('x', 8) pre_processed = pre_processor(x == 5) print(pre_processed[0]) solver.add(pre_processed[0]) # add the sole assertion if solver.check() == sat: print(solver.model()) model = pre_processed[0].convert_model(solver.model()) print(model) print(model[x].as_long()) ```
This commit is contained in:
parent
2714dc2623
commit
b4c2b455bd
1 changed files with 5 additions and 0 deletions
|
@ -6284,6 +6284,11 @@ extern "C" {
|
|||
The model may be null, in which case the returned model is valid if the goal was
|
||||
established satisfiable.
|
||||
|
||||
When using this feature it is advisable to set the parameter \c model.compact to \c false.
|
||||
It is by default true, which erases variables created by the solver from models.
|
||||
Without access to model values for intermediary variables, values of other variables
|
||||
may end up having the wrong values.
|
||||
|
||||
def_API('Z3_goal_convert_model', MODEL, (_in(CONTEXT), _in(GOAL), _in(MODEL)))
|
||||
*/
|
||||
Z3_model Z3_API Z3_goal_convert_model(Z3_context c, Z3_goal g, Z3_model m);
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue