This commit adds field initializers to the CFG for non-static constructors. For
example, in
```
class C
{
int Field1 = 0;
int Field2 = Field1 + 1;
int Field3;
public C()
{
Field3 = 2;
}
public C(int i)
{
Field3 = 3;
}
}
```
the initializer expressions `Field1 = 0` and `Field2 = Field1 + 1` are added
to the two constructors, mimicking
```
public C()
{
Field1 = 0;
Field2 = Field1 + 1;
Field3 = 2;
}
```
and
```
public C()
{
Field1 = 0;
Field2 = Field1 + 1;
Field3 = 3;
}
```
respectively. This means that we no longer have to synthesize calls, callables,
parameters, and arguments in the data flow library, so much of the work from
d1755500e4 can be simplified.
The predicate `maxSplits()` was previously applied dynamically to ensure that
any control flow node would keep track of at most `maxSplits()` number of splits.
However, there was no guarantee that two different copies of the same AST element
wouldn't contain different splits, so in general the number of copies for a given
AST element `e` could be on the order `$\binom{n}{k}c^k$`, where `n` is the total
number of splits that apply to `e`, `k = maxSplits()`, and `c` is a constant.
With this change, the relevant splits for `e` are instead computed statically,
meaning that the order is instead `$c^k$`.
When completions are inherited by elements inside `finally` blocks, we previously
threw away the underlying completion. For example, in
```
try
{
if (b)
throw new Exception();
}
finally
{
if (b)
...
}
```
the completions for `b` inside the `finally` block are `true` and `throw(Exception)`,
where the latter is inherited from the `try` block, with an underlying `false`
completion. Throwing away the `false` completion meant that we were unable to prune
the `false` edge (Boolean CFG splitting).
- Extract names of properties in a propery match, using the `exprorstmt_name` relation.
- Simplify extraction of properties by not distinguishing between top-level patterns
and nested patterns.
- Introduce `PatternExpr` to capture patterns in `is` expressions, `case` statements,
and `switch` expression arms.
- Generalize `IsTypeExpr`, `IsPatternExpr`, `IsRecursivePatternExpr`, and `IsConstantExpr`
to just `IsExpr` with a member predicate `PatternExpr getPattern()`.
- Generalize `TypeCase`, `RecursivePatternCase`, and `ConstCase` to just `CaseStmt` with
a member predicate `PatternExpr getPattern()`.
- Introduce classes `Switch` and `Case` as base classes of switch statements/expressions
and case statements/switch expression arms, respectively.
- Simplify CFG logic using the generalized classes.
- Generalize guards library to cover `switch` expressions tests.
- Generalize data flow library to cover `switch` expression assignments.
- Cache predicates in the same stage using a cached module.
- Introduce `DefUse::defUseVariableUpdate()` and use in `CallableReturns.qll`.
The updated file `csharp/ql/test/library-tests/cil/dataflow/Nullness.expected`
demonstrates why this is needed.
- Utilize CIL analysis in `Guards::nonNullValue()`.
- Analyze SSA definitions in `AlwaysNullExpr`, similar to `NonNullExpr`.
Write accesses in assignments, such as the access to `x` in `x = 0` are not
evaluated, so they should not have entries in the control flow graph. However,
qualifiers (and indexer arguments) should still be evaluated, for example in
```
x.Foo.Bar = 0;
```
the CFG should be `x --> x.Foo --> 0 --> x.Foo.Bar = 0` (as opposed to
`x --> x.Foo --> x.Foo.Bar --> 0 --> x.Foo.Bar = 0`, prior to this change).
A special case is assignments via acessors (properties, indexers, and event
adders), where we do want to include the access in the control flow graph,
as it represents the accessor call:
```
x.Prop = 0;
```
But instead of `x --> x.set_Prop --> 0 --> x.Prop = 0` the CFG should be
`x --> 0 --> x.set_Prop --> x.Prop = 0`, as the setter is called *after* the
assigned value has been evaluated.
An even more special case is tuple assignments via accessors:
```
(x.Prop1, y.Prop2) = (0, 1);
```
Here the CFG should be
`x --> y --> 0 --> 1 --> x.set_Prop1 --> y.set_Prop2 --> (x.Prop1, y.Prop2) = (0, 1)`.
