Summary: Impurity traces are quite big due to recording values histories. Let's simplify the traces by removing pulse's value histories.
Reviewed By: skcho
Differential Revision: D20362149
fbshipit-source-id: 8a2a6115e
Summary: Type is not enough to say a function call of `Provider.get` is expensive or not.
Reviewed By: jvillard
Differential Revision: D20366206
fbshipit-source-id: 83d3e8741
Summary:
This diff uses a type parameter of `Provider.get` to decide whether assigning expensive cost to the
function call or not. For example, if the type is small one like `Provider<Integer>`, it be
evaluated to have a unit cost, otherwise a linear cost.
To get the return type of `Provider.get`, I added a simple analyzer that collects "casted" types
backwards. In Sil, while the function call statement loses the return type, e.g,
```
n$5=_fun_Object Provider.get()(n$3:javax.inject.Provider*);
```
the `n$5`'s value is usually casted to a specific type at some point later.
```
*&$irvar0:java.lang.Object*=n$5
n$8=*&$irvar0:java.lang.Object*
n$9=_fun___cast(n$8:java.lang.Object*,sizeof(t=java.lang.Integer;sub_t=( sub )(cast)):void)
```
So, the analyzer starts from the cast statements backward, collecting the types to cast for each
variables.
Reviewed By: ezgicicek
Differential Revision: D20345268
fbshipit-source-id: 704b42ec1
Summary: This diff adds a model for Java's `Object.clone()` method (similar to existing shallow_copy).
Reviewed By: jvillard
Differential Revision: D20341073
fbshipit-source-id: 30ae40fe7
Summary:
Some (all?) of this is already tested in other tests, but this feature
is important enough (and the implementation is scattered accross the
whole code), so I found it useful to have a small test that ensures the
very basic things are working as expected.
See `NestedFieldAccess.java` that tests far more advances things, but
here we focus only very basic things: conditions, local variable
assignments, and explicit assignments.
Reviewed By: jvillard
Differential Revision: D20339056
fbshipit-source-id: a6cfd0043
Summary: We forgot to take skipped calls into account for state comparison. This diff fixes that.
Reviewed By: skcho
Differential Revision: D20282739
fbshipit-source-id: 7b4d84bb0
Summary:
These were not used (and were actually activated byt the same config
param). They both are in experimental stage that never reached maturity.
Since the team does not have immediate plans to work on ObjC nullability
checker; and since "eradicate" (now known as nullsafe) is the main
solution for Java, removing it is sensible.
Reviewed By: jvillard
Differential Revision: D20279866
fbshipit-source-id: 79e64992b
Summary:
This is the kind of property for which the previous syntax forced one to
use spurious registers.
Reviewed By: ngorogiannis
Differential Revision: D20118863
fbshipit-source-id: b49740d33
Summary:
This diff renames `ZERO_XXX` issues to more appropriately named and descriptive
`XXX_UNREACHABLE_AT_EXIT` and replaces bottom with
unreachable in cost kinds and issues.
Reviewed By: skcho
Differential Revision: D20140301
fbshipit-source-id: eb6076b30
Summary:
1. It is convenient to stick with the policy "ERROR if and only if it is
enforced". Among other, it makes CI integration much easier to implement
(enforcemend, UI and messaging is decided based on severity).
2. Since Nullsafe annotation is an idiomatic way to indicate classes
with enforced nullability checking, we want it to be the only way to
enforce issues.
3. This means we decrease the priority of GraphQL violation issues.
(In practice they were not enforced so we have plenty of violations in
codebase to reflect reality). The proper way dealing with GraphQL will
be detecting such issues as a special issue type and prioritizing fixing
and Nullsafe-ifying corresponding classes.
4. Among other, we downgrade severity of field overannotated to advice
to keep it consistent with condition redundant.
Reviewed By: artempyanykh
Differential Revision: D20141420
fbshipit-source-id: e2f12835a
Summary:
The issue type `ZERO_EXECUTION_TIME` actually corresponds to bottom state but has been mistakenly used to mean
- unreachable nodes (program never reaching exit state)
- having zero cost (e.g. for allocations).
Note that, for execution costs, the latter doesn't make sense since we always incur a unit cost for the start node. Hence, a function with empty body will have unit cost. For allocations or IO however, we only incur costs for specific primitives, so a function with no allocations/IO could have a zero cost. However, there is no point reporting functions with zero cost as a specific issue type. Instead, what we want to track is the former, i.e. functions whose cost becomes 0 due to program never reaching exit state.
This diff aims to split these cases into two by only reporting on the latter and adds traces to bottom/unreachable cost by creating a special category in polynomials.
Next diff will rename `ZERO_XXX` to `XXX_UNREACHABLE_AT_EXIT`.
Reviewed By: skcho
Differential Revision: D20005774
fbshipit-source-id: 46b9abd5a
Summary:
For Mode.Local this is kind of obvious decision.
But this diff does the same for strict mode as well.
See comment in [ExplicitNonnullThirdParty] for the detailed explanation.
Reviewed By: artempyanykh
Differential Revision: D20140056
fbshipit-source-id: 13c66df81
Summary:
In the previos diff we restructured error rendering utils for
TypeOrigin.MethodCall.
In this diff we do the same with TypeOrigin field: lets make the code
consistent.
We also clearly distinct third party from all other possible cases in
this branch.
This changes messaging and reported errors for strict modes (see test cases), and I believe this is a net improvement.
Reviewed By: artempyanykh
Differential Revision: D20139741
fbshipit-source-id: 84f502553
Summary:
> We don't report when the cost is Top as it corresponds to subsequent 'don't know's. Instead, we
> report Top cost only at the top level per function
The previous code just ignored top costed nodes, so it was able to report a non-top cost that was
from another node. For example,
```
void foo() {
linear-cost();
top-cost();
}
```
It reported inconsistent reports: `EXPENSIVE_EXECUTION_TIME` with a linear cost and
`INFINITE_EXECUTION_TIME` at the same time.
This diff fixes it not to report `EXPENSIVE_EXECUTION_TIME` when there is a node with the top cost.
Reviewed By: ezgicicek
Differential Revision: D20139408
fbshipit-source-id: 9fedd4aec
Summary:
In the previous report, it reported the first cost of node that exceeds a threshold. However, this
may hide a bigger cost of node that appears later. This diff changes this to report the biggest
cost of node among the costs exceeding the threshold.
Reviewed By: ezgicicek
Differential Revision: D20116162
fbshipit-source-id: 06199fb46
Summary:
This syntax
- is less confusing (according to several people who are not me);
objectively, there's less magic under the hood
- gives fine control over register number (because condition/action are separated)
- lets one compare values of different arguments of the same call
(e.g., one could have a transition that is taken only if two
arguments of a method call are equal)
Reviewed By: ngorogiannis
Differential Revision: D20005403
fbshipit-source-id: fad8f3b3d
Summary:
The test shows what that TOPL can express, in addition to bugs,
efficiency properties. However, there seems to be an underlying problem
in biabdaction that prevents this particular problem from being caught.
Reviewed By: ngorogiannis
Differential Revision: D20005404
fbshipit-source-id: 466f79050
Summary: The semantics of the `values` function of Java enum class was missing, when it is called outside the class initializer. This diff gets the size of the enum elements from the summary of class initializer function, `<clinit>`.
Reviewed By: ezgicicek
Differential Revision: D20094880
fbshipit-source-id: 7362bba1c
Summary:
Now when typechecking a class `A` marked with `Nullsafe(LOCAL)`,
classes from trusted list are properly recognized and nullability of
method params and return value are refined to `LocallyCheckedNonnull`
in a context of class `A`.
NOTE: refininng nullability when **accessing fields** on trusted classes
is **not implemented yet**, because the whole business of handling fields
in nullsafe is somewhat convoluted. This should not be a huge issue
though, since in Java fields are commonly accessed via getters any
way.
Reviewed By: mityal
Differential Revision: D20056158
fbshipit-source-id: 496433d90
Summary:
This will help making error reporting more actionable.
Often methods that are nullable in general (like View.findViewById) are used as not-nullable due to app-invariants. In such cases suggesting a non-nullable alternative that does an assertion under the hood makes the error report more actionable and provides necessary guidance with respect to coding best practices
Follow up will include adding more methods to models.
If this goes well, we might support it in user-defined area (nullability
repository)
Reviewed By: artempyanykh
Differential Revision: D20001416
fbshipit-source-id: 46f03467c
Summary:
Introduction of `ThirdPartyNonnull` nullability broke nullability
refinement heuristic for enums. This diff fixes it and also adds tests
so that we hopefully avoid such issues in future.
Reviewed By: mityal
Differential Revision: D19975810
fbshipit-source-id: f9245f305
Summary:
We need to be able to differentiate `UncheckedNonnull`s in internal vs
third-party code. Previously, those were under one `UncheckedNonnull`
nullability which led to hacks for optmistic third-party parameter
checks in `eradicateChecks.ml` and lack of third-party enforcement in
`Nullsafe(LOCAL, trust=all)` mode (i.e. we want to trust internal
unchecked code, but don't want to trust unvetted third-party).
Now such values are properly modelled and can be accounted for
regularly within rules.
Also, various whitelists are refactored using
`Nullability.is_considered_nonnull ~nullsafe_mode nullability`.
`ErrorRenderingUtils` became a tad more convoluted, but oh well, one
step at a time.
Reviewed By: mityal
Differential Revision: D19977086
fbshipit-source-id: 8337a47b9
Summary:
Add support for nullsafe mode with `trust=all` and `trust=none` a case
with a specific trust list is not supported yet and needs to be
implemented separately.
Tests introduce one unexpected
`ERADICATE_INCONSISTENT_SUBCLASS_PARAMETER_ANNOTATION` issue which
complains about `this` having incorrect nullability; it is a bug and
needs to be fixed separately.
Reviewed By: mityal
Differential Revision: D19662708
fbshipit-source-id: 3bc1e3952
Summary: In all other cases we have period at the end, which is inconsistent.
Reviewed By: artempyanykh
Differential Revision: D20001065
fbshipit-source-id: 85ec6d751
Summary:
Use a record of package, class name to store (qualified) Java class names. This saves the round trip of concatenating then splitting again, etc, as well as saves some memory in the type environment as now the package paths can be shared across classes of the same package (about 10% in tests).
Also remove some unfortunate APIs.
Reviewed By: jvillard
Differential Revision: D19969325
fbshipit-source-id: f7b7f5a55
Summary: The way `Mangled.t` is used in `JavaClassName` means that it's always a plain string (we never have a "mangled" part). Remove the indirection and extra allocation. Also, simplify the API by throwing away one function that was used just once and wastefully.
Reviewed By: artempyanykh
Differential Revision: D19950672
fbshipit-source-id: b61fcba6e
Summary:
This adds a violation of baos.topl found in github/seata/seata. However,
it is not a bug (see comment in commit).
Reviewed By: ngorogiannis
Differential Revision: D19518641
fbshipit-source-id: e219245ee
Summary:
Since Javalib 3.2, a new feature allows to rewrite
methods that contain (some specific form of) closures. Infer
now uses it. When loading each class we rewrite them and
new classes generated by Javalib to implements closures
(i.e. Java interfaces)<
Reviewed By: ngorogiannis
Differential Revision: D19389227
fbshipit-source-id: 245dd4404
Summary:
We already warn about lack of nullable annotations in `equals()`, and even have a specialized error message for that.
But lack of an annotation is not as severe as direct dereference: the
latter is a plain bug which is also a time bomb: it will lead to an NPE not immediately.
This is widespread enough to be reported separately.
Reviewed By: dulmarod
Differential Revision: D19719598
fbshipit-source-id: a535d43ea
Summary:
Since we fixed a bug in implementation of FalseOnNull (see stack below),
we can finally ship this change.
Side note: this change is essential for the follow up diff (which adds extra check
for user-defined implementations of equal()), without it the follow
up change would introduce a lot of false positives.
Reviewed By: ngorogiannis
Differential Revision: D19771057
fbshipit-source-id: 7d7cf1ef7
Summary:
If we managed to whitelist a function as TrueOnNull, we should teach
nullsafe the nullability of its arguments, otherwise it will ask not to
pass null here.
This fixes a silly FP warning, see the test.
Reviewed By: dulmarod
Differential Revision: D19770341
fbshipit-source-id: 0f861fae1
Summary:
Yay, the previous refactoring finally makes it possible to do some actual
changes to the code in `TypeCheck.ml`!
Changes in this diff:
1. Fixes the bug: TrueOnNull and FalseOnNull were working only for
static methods. Surpsingly nobody noticed that. It is because the first
argument for non-static method was `this`.
2. Behavior change: TrueOnNull/FalseOnNull were not working correctly
where there are several argumens. See the task attached for the example
of the legit usecase. Now the behavior is the following: if there are
several Nullable arguments infer nullability for all of them.
Reviewed By: skcho
Differential Revision: D19770219
fbshipit-source-id: 7dffe42cd
Summary:
This refactoring unblocks the changes in follow up diffs (plus fixes a
bug).
So what was happening?
Each comparison with null leads to CFG being splitted into two branches, one branch
is PRUNE(a == null) and another is PRUNE(a != null).
PRUNE(a != null) is where most of logic happens, it is the place where
we infer non-null nullability for a, and this is a natural place to
leave a check for redundancy.
Before this diff we effectively checked the same thing twice, and used
`true_branch` (only one of 2 instruction will have it set to true) as a symmetry breaker.
This diff removes the `true_branch` checks, but leaves only one call out
of two, hence breaking symmetry in a different way.
## Bug fix
The code around the removed check was (crazily) doing two things at
once: it processed results of (returning booleans!)
TrueOnNull-annotated functions AND
results of (returning Objects!) other functions, using the fact that all
of them are encoded as zero literals (sic!).
Not surprisingly that lead to a bug where we accidentally call the check
for non intended places (arguments of trueOnNull functions), which lead
to really weird FP.
This diff fixes it.
Reviewed By: dulmarod
Differential Revision: D19744604
fbshipit-source-id: fe4e65a8f
Summary: We don't use allocation costs in prod at the moment. There is no plan to do so in the near future. Let's not report them anymore and also save some space in `costs-report.json`.
Reviewed By: skcho
Differential Revision: D19766828
fbshipit-source-id: 06dffa61d
Summary:
This test tests PropagatesNullable and TrueOnNull/FalseOnNull
annotations.
Both tests suites grew big so it is hard to observe them at glance and
make changes.
I could not figure out better name for TrueFalseOnNull.java, it is sort
of silly but I optimized for searchability, "FalseOnNull" will be
directly searched and "TrueOnNull" will be searched in IDEs that are
smart enough.
Reviewed By: skcho
Differential Revision: D19724512
fbshipit-source-id: 703961342
Summary: This diff returns non-symbolic value (top) for unknown external function calls because the symbolic values sometimes make it hard to understand costs.
Reviewed By: ezgicicek
Differential Revision: D18685715
fbshipit-source-id: 1b39c718b
Summary:
Pulse has an extra invalidation mechanism (introduced in D18726203) to prevent something invalid (e.g. `null`) to be passed by reference to an initialisation function. Therefore, it havocs formals passed by reference to skipped functions. However, I don't think this makes sense in Java. So, let's turn it off.
A nice consequence of this is that in impurity analysis, we do not consider functions that call skipped library calls with object arguments as writing to their formals.
Reviewed By: skcho
Differential Revision: D19697110
fbshipit-source-id: 6e3a71f2a
Summary:
- Thread the two types into one instead of having a record where the `path` field doesn't always make sense (`Class` case).
- Improved pretty printing of class objects (java only).
- Move starvation-specific stuff out of `AbstractAddress` (eg `make_java_synchronized`).
- Slight optimisation of `apply_subst` for when a parameter is used without additional accesses inside a method (then, the substitution need not modify the term substituted for the parameter in any way).
Reviewed By: ezgicicek
Differential Revision: D19639922
fbshipit-source-id: 1cebecf5d
Summary:
`String` and `StringBuilder` both implement `CharSequence`. Let's generalize the model for `String` to `CharSequence` wherever possible and add missing models for
- `StringBuilder.append`
- `StringBuilder.toString`
Reviewed By: skcho
Differential Revision: D19558009
fbshipit-source-id: 0dfdb21af
Summary:
Java's String models were broken for
- initializing a String object with a locally defined constant string (which is an `Object*` in SIL).
- initializing a String object with a `char`/`byte` array
This diff fixes them and also adds models for `new String ()`.
Reviewed By: skcho
Differential Revision: D19662180
fbshipit-source-id: 23968d0aa
Summary:
Prevent returning a negative cost bound when calling `substring(begin_index, end_index)` when either is possible
- `begin_index < 0`
- `begin_index > end_index`
Instead, return unit cost since such cases either throw `IndexOutOfBoundsException ` at runtime or correspond to having two symbolic bounds that cannot be semantically compared.
Reviewed By: dulmarod
Differential Revision: D19619410
fbshipit-source-id: cf5e8cb7b
Ezgi Çiçek
Sungkeun Cho
Mitya Lyubarskiy
Radu Grigore
Artem Pianykh
Nikos Gorogiannis
David Pichardie