Summary:
This is mostly useful to avoid duplicating error states, which are
propagated unchanged through both branches of, say, conditionals, and
can end up duplicated if the join is not careful:
```
{[Abort(Error 1), Abort(Error 2), Continue σ']}
if (..) { .. } else { .. }
{JOIN([Abort(Error 1), Abort(Error 2), Continue σ_then],
[Abort(Error 1), Abort(Error 2), Continue σ_else])}
{[Abort(Error 1), Abort(Error 2), Continue σ_then, Continue σ_else]}
```
Whereas before this diff we got
```
{[Abort(Error 1), Abort(Error 2), Continue σ_then, Abort(Error 1), Abort(Error 2), Continue σ_else]}
```
Detect states that do not change simply using `phys_equal` as they
should literally not change. Refactor the code to be able to re-use the
same logic in the stronger join used in widening, that compares states
using the domain's `leq` relation to establish implication.
Reviewed By: ezgicicek
Differential Revision: D27908529
fbshipit-source-id: b461165da
Summary:
When a block value is passed via more than one-depth of function calls, it is not analyzed correctly
because current inlining mechanism (specializing objc block parameters) of the frontend works for
only one-depth of block passing. This diff gives up analyzing initialized-ness of captured
variables in ObjC to avoid FPs.
Reviewed By: da319
Differential Revision: D27885395
fbshipit-source-id: fc6b4663c
Summary: We have some FPs due to unknown init methods that are added dynamically.
Reviewed By: ezgicicek
Differential Revision: D27856371
fbshipit-source-id: b6fb46df3
Summary:
This diff filters known expensive callees when cost is constant.
previous:
```
foo() {
known_expensive_call();
}
```
current:
```
foo() {
known_expensive_call();
goo();
}
// cost is constant
goo() {
known_expensive_call();
unknown_call();
}
```
When callee's cost is constant and its summary includes known expensive callees, the checker addressed it as a non-constant-cost callee, i.e., it copies all ungated callees from the callee's summary. However, sometimes this full-copying introduces unexpected issues. For example, suppose a callee `goo` is added and `goo`'s cost is constant as above. Since it includes `known_expensive_call`, all ungated callees of its summary is copied to the caller `foo`'s summary:
* `foo`'s ungated callees (before): {`known_expensive_call`}
* `foo`'s ungated callees (after): {`known_expensive_call`, `unknown_call`}
As a result, it would report about `unknown_call` is added. However, this is not what we intended: In the example, `unknown_call` is reported because it is called in the same function with `known_expensive_call`, not because it is expensive.
To fix that issue, this diff filters known expensive callees from `goo`'s summary in that case.
Reviewed By: ezgicicek
Differential Revision: D27852552
fbshipit-source-id: d207eef1c
Summary:
It introduced a FP due to reporting addition of `__cast`.
* This diff added known cheap model for `__cast`.
* In addition, moved `match_builtin` to `BuiltinDecl`.
Reviewed By: ezgicicek
Differential Revision: D27791495
fbshipit-source-id: 55aec1728
Summary:
Previously we were only taking constexpr into account on constructors.
Add this info to ProcAttributes.t instead by exporting it from the
plugin for all functions.
This allows SIOF to take constexpr into account in more cases as it's not
always good at capturing which functions *can* be constexpr-evaluated,
which caused false positives.
Delete now-useless is_constexpr in constructor types. This generated the
changes in frontend tests.
Some minor renamings of variants of is_const_expr -> is_constexpr.
Reviewed By: da319
Differential Revision: D27503433
fbshipit-source-id: 3d1972900
Summary:
When a field is assigned by a value,
```
_.field = exp;
```
it should collect the field when the abstract config value of `exp` is non-bottom, rather than
non-top.
Reviewed By: ezgicicek
Differential Revision: D27766188
fbshipit-source-id: a0b1f2c28
Summary: Added a new issue type for sending a message to nil when its return type is non-POD. To distinguish these issues from other nullptr dereference issues, we extend the `MustBeValid` attribute to contain the reason of why an address must be valid. For now a reason can only have `SelfOfNonPODReturnMethod` as it's value, but in the future we will use it for other nullability issue types, such as nil insertion into collections.
Reviewed By: jvillard
Differential Revision: D27762333
fbshipit-source-id: 689e5a431
Summary:
In order to use Inferbo's analysis result, a checker should know current instruction index.
However, for the checkers using `ProcCfg.Normal` CFG, it was impossible to get the instruction
index. To solve the issue, this diff changes the AbsInt framework to give the index together to
`exec_instr`.
Reviewed By: ezgicicek
Differential Revision: D27680894
fbshipit-source-id: 1dc8ff0fb
Summary:
The problem is that `Models.is_field_nonnullable` didn't differentiate
between
- having a nullable model (in which case we want the field to be NULLABLE),
- not having a model at all (in which case we want the field to be THIRDPARTY_NONNULL).
The problem was noticed only now because previously we didn't have any
NULLABLE field models.
Reviewed By: ngorogiannis
Differential Revision: D27709508
fbshipit-source-id: b98c8f86f
Summary:
This diff evaluates a cpp vector given as a parameter symbolically. Especially, it addresses it as an array, so the cost checker can use its symbolic length correctly.
**About handling `cpp.vector_elem` field:**
The field is a virtual field of vector object that points to the array of vector elements. It was introduced in Inferbo to model semantics of vector operations.
Since many semantics of Inferbo depends on type information, it had collected type information of vector elements, whenever `cpp.vector_elem` field was introduced, as a *side-effect*. A problem is that it has *side-effect*, which means it may introduce non-deterministic analysis results depending on the type information of the virtual field.
This diff changes it not to collect the type information on `cpp.vector_elem` as a side-effect. Instead, it tries to write the information to the abstract states (abstract location) when possible.
Reviewed By: ezgicicek
Differential Revision: D27674935
fbshipit-source-id: f3d52cae7
Summary:
In D27430485 (a6ab4d38cf), we used the static cost of the callee to determine whether it was cheap/expensive. This diff improves on that by taking the whole instantiated cost of the function call (not just the callee's cost).
Also, if the callee is an unmodeled call, we consider it to be expensive as before.
Note: cost instantiation was used by hoisting. I refactored bunch of code there to reuse as much as code possible.
Reviewed By: skcho
Differential Revision: D27649302
fbshipit-source-id: 07d11f3dd
Summary: When instantiating the callee's cost, we have picked up the InferBo memory at the node corresponding to the last instruction. Instead, we should pick up right at the call instruction. Picking it up later might cause arguments to go out of scope.
Reviewed By: skcho
Differential Revision: D27652474
fbshipit-source-id: 5ab35cabb
Summary:
The output differs on Java 11 compared to Java 8: one prints an
interface, the other resolves to a class name.
Reviewed By: ezgicicek
Differential Revision: D27678552
fbshipit-source-id: c5a5d0c39
Summary: We have been referring to the arguments of a function call as "params". This has been bothering me. Let's fix it!
Reviewed By: ngorogiannis
Differential Revision: D27649158
fbshipit-source-id: 10e0b28cb
Summary:
To avoid too big abstract states due to instantiated templates in C++,
this diff loosens the compare functions of field names and ungated
callees.
Reviewed By: ezgicicek
Differential Revision: D27625775
fbshipit-source-id: e33e9d34c
Summary:
Nullsafe/biabduction tests were sensitive to Java version: they were recorded for Java 8 but if the machine that is used to run the tests had Java 11, tests would fail. This diff aims to resolve this issue by
- making our tests produce java8-compatible bytecode so that tests don't fail on Java 11 machines
- removing nullsafe tests that exercise obscure Java 8 behavior that cannot be alleviated with backward compatible bytecode on Java 11
- changing lambda argument printing to be Java 11 compatible
Reviewed By: martintrojer
Differential Revision: D27500731
fbshipit-source-id: 77fe302ea
Summary:
Reporting all ungated (un configed?) function calls causes many FPs. Instead, we rely on complexity analysis to determine whether a function is cheap/expensive: if the callee's complexity is not symbolic (e.g. constant), we consider it as cheap and don't keep track of it.
Note that we don't take the instantiated/modeled cost into account yet. So, if we have `foo(int n)` with complexity `O(n)`, and call it as `foo(3)`, we would still keep track of it. Similarly, if `foo` is a modeled function with constant time complexity, we would have no summary for it hence would keep track of it.
These will be improved later.
Reviewed By: skcho
Differential Revision: D27430485
fbshipit-source-id: d5f66320d
Summary:
This diff removes additional inferbo options `--bufferoverrun` from cost tests, since printing
inferbo issues is not that useful to understand cost results.
Reviewed By: ngorogiannis
Differential Revision: D27592496
fbshipit-source-id: 6ab3e6528
Summary:
Whenever an equality "t = v" (t an arbitrary term, v a variable) is
added (or "v = t"), remember the "t -> v" mapping after canonicalising t
and v. Use this to detect when two variables are equal to the same term:
`t = v` and `t = v'` now yields `v = v'` to be added to the equality
relation of variables. This increases the precision of the arithmetic
engine.
Interestingly, the impact on most code I've tried is:
1. mostly same perfs as before, if a bit slower (could be within noise)
2. slightly more (latent) bugs reported in absolute numbers
I would have expected it to be more expensive and yield fewer bugs (as
fewer false positives), but there could be second-order effects at play
here where we get more coverage. We definitely get more latent issues
due to dereferencing pointers after testing nullness, as can be seen in
the unit tests as well, which may alone explain (2).
There's some complexity when adding term equalities where the term
is linear, as we also need to add it to `linear_eqs` but `term_eqs` and
`linear_eqs` are interested in slightly different normal forms.
Reviewed By: skcho
Differential Revision: D27331336
fbshipit-source-id: 7314e127a
Summary:
It's better (=possibly more efficient) to take the opportunity to
normalize linear terms when we can instead of possibly having to apply
the same normalization over and over on individual terms until the next
round of proper normalization.
Reviewed By: skcho
Differential Revision: D27464885
fbshipit-source-id: 0dc01a089
Summary:
When we don't know the value being shifted it may help to translate
bit-shifting into multiplication by a constant as it might surface
linear terms, eg `x<<1` is `2*x`.
Reviewed By: skcho
Differential Revision: D27464847
fbshipit-source-id: 9b3b5f0d0
Summary:
The simplifications done by `simplify_shallow` are all taken care of by
`eval_const_shallow` as well, they just also happen to help when not
*all* of the term is a constant. However, they might be less
precise/efficient than in the constant case, in particular in the next
diff that translates `x << c` into `x * 2^c` when `c` is constant.
Reviewed By: skcho
Differential Revision: D27464805
fbshipit-source-id: 452bc6ab1
Summary:
On some pathological examples of crypto primitives like libsodium, later
diffs make pulse grind to a halt due to an explosion in the size of
literals. This is at least partly due to the fact the arithmetic doesn't
operate modulo 2^64.
Due to the fact the arithmetic is confused in any case when we reach
such large numbers, cap them, currently at 2^128. This removes pathological
cases for now, even now on libsodium Pulse is ~5 times faster than before!
Take this opportunity to put the modified Q/Z modules in the own files.
Reviewed By: jberdine
Differential Revision: D27463933
fbshipit-source-id: 342d941e2
Summary: Just some scaffolding to save a bit of churn from the next diff.
Reviewed By: skcho
Differential Revision: D27328348
fbshipit-source-id: 4f5bfcc65
Summary:
This was added in C++14. Was investigating how SIOF dealt with this but
it turns out it already does the right thing as the translation unit of
global variable templates shows up as the place they are instantiated
(not the one where they are declared), which works well for SIOF
checking.
Reviewed By: da319
Differential Revision: D27500998
fbshipit-source-id: b8b9b9c48
Summary:
This is better suited than the generic "cGeneral_utils", and saves
exporting one of them too.
Reviewed By: da319
Differential Revision: D27500933
fbshipit-source-id: f4224f63b
Summary: One source of non-deterministic diff result is when there are multiple overloaded methods the cardinals of unchecked callees of which are the same. This diff tries to select one of them in a more deterministic manner.
Reviewed By: ezgicicek, ngorogiannis
Differential Revision: D27430757
fbshipit-source-id: 38ba5d8dc
Summary: Error message was accidentally changed to a specific nullptr error message (D26887140 (cba144b779)) for any invalidation (use after delete, etc). This diff reverts back the error message for a general case and keeps the special case for nullptr dereference. Also fixed spacing for nullptr dereference error message.
Reviewed By: jvillard
Differential Revision: D27407628
fbshipit-source-id: 2649f3032
Summary:
The title
Also notice that there is a duplication of an error.
Reviewed By: skcho
Differential Revision: D27426933
fbshipit-source-id: dbd2f861a
Jules Villard
Sungkeun Cho
Ezgi Çiçek
Artem Pianykh
Loc Le
Daiva Naudziuniene
Radu Grigore
Nikos Gorogiannis