Summary:
We need to make sure that destructors of virtual base classes are called only once. Similarly to what clang does, we have two destructors for a class: a destructor wrapper and an inner destructor.
Destructor wrapper is called from outside, i.e., when variables go out of scope or when destructors of fields are being called.
Destructor wrappers have calls to inner destructors of all virtual base classes in the inheritance as their bodies.
Inner destructors have destructor bodies and calls to destructor wrappers of fields and inner destructors of non-virtual base classes.
Reviewed By: dulmarod
Differential Revision: D5834555
fbshipit-source-id: 51db238
Summary:
The "placement new" operator `new (e) T` constructs a `T` in the pre-allocated memory address `e`.
We weren't translating the `e` part, which was leading to false positives in the dead store analysis.
Reviewed By: dulmarod
Differential Revision: D5814191
fbshipit-source-id: 05c6fa9
Summary:
Simple instance of the problem: analyzing the following program times out.
```
#include <tuple>
void foo() {
std::tuple<std::tuple<int>> x;
}
```
Replacing `std::tuple<std::tuple<int>>` by `std::tuple<int>` makes the analysis
terminate.
In the AST, both tuple<tuple<int>> and tuple<int> have the same template
specialization type: "Pack" (which means we're supposed to go look into the
arguments of the template to get their values). This is not information enough
and that's the plugin fault.
On the backend side, this means that two types have the same Typ.Name.t, namely
"std::tuple<_>", so they collide in the tenv. The definition of
tuple<tuple<int>> is the one making it into the tenv. One of the fields of the
corresponding CxxRecord is of type "tuple<int>", which we see as the same
"tuple<_>", which causes the loop.
update-submodule: facebook-clang-plugins
Reviewed By: mbouaziz
Differential Revision: D5775840
fbshipit-source-id: 0528604
Summary: Destroying local variables that are out of scope after `continue`.
Reviewed By: jberdine
Differential Revision: D5804120
fbshipit-source-id: 638cff5
Summary: Destroying local variables that are out of scope after `break`.
Reviewed By: jberdine
Differential Revision: D5764647
fbshipit-source-id: a7e06ae
Summary: The successor node of `continue` was not correct inside the `do while`.
Reviewed By: sblackshear
Differential Revision: D5769578
fbshipit-source-id: d7b0843
Summary: We inject destructor calls of base classes inside destructor bodies after the destructor calls of fields.
Reviewed By: dulmarod
Differential Revision: D5745499
fbshipit-source-id: 90745ec
Summary: We used to crash whenever we hit these. The simple translation implemented here is not particularly inspiring, but it is better than crashing.
Reviewed By: jvillard
Differential Revision: D5702095
fbshipit-source-id: 3795d43
Summary:
Previous version was hard to understand because it was doing many things within same code. New version has different code for Arrays, Structs and others.
There is some copy-paste, but it's easier to follow code (open to suggestions though)
Reviewed By: dulmarod
Differential Revision: D5547999
fbshipit-source-id: 77ecb24
Summary:
Bumps facebook-clang-plugins to a version that outputs sizeof() info in bytes and not bits.
update-submodule: facebook-clang-plugins
Reviewed By: akotulski
Differential Revision: D5526747
fbshipit-source-id: 6019542
Summary: The `--failures-allowed` was doing for the Clang frontend what `--keep-doing` was doing for the backend. This revision merges the two options to simplify the Infer CLI and our tests.
Reviewed By: jvillard
Differential Revision: D5474347
fbshipit-source-id: 09bcea4
Summary:
This diff tries to achieve the followings: if we have the following C++ codes:
```
bool foo(int x, int y) {
return &x == &y;
}
```
We want the C++ frontend to emit Sil as if the input is written as
```
bool foo(int x, int y) {
if (&x == &y) return 1; else return 0;
}
```
This matches the behavior of our Java frontend.
The reason why we prefer an explicit branch is that it will force the backend to eagerly produce two different specs for `foo`. Without the explicit branch, for the above example the backend would produce one spec with `return = (&x == &y)` as the post condition, which is not ideal because (1) we don't want local variables to escape to the function summary, and (2) with the knowledge that no two local variables may alias each other, the backend could actually determines that `&x == &y` is always false, emitting a more precise postcondition `return = 0`. This is not possible if we do not eagerly resolve the comparison expression.
Reviewed By: akotulski
Differential Revision: D5260745
fbshipit-source-id: 6bbbf99
Summary:
Don't store redundant information in C++ template Type.Name.t.
New signature:
```
| CppClass (qual_name, template_args)
```
For example, for `std::shared_ptr<int>`, will look like this:
```
| CppClass (["std", "shared_ptr"], Template [int])
```
While it used to be:
```
| CppClass (["std", "shared_ptr<int>"], Template (["std", "shared_ptr"], [int]))
```
Reviewed By: jberdine, mbouaziz
Differential Revision: D4834512
fbshipit-source-id: cb1c570
Summary:
Title.
The way types are printed is completely valid, but little weird for some C++ programmers:
`int const` - same as `const int`
`int * const` - pointer is `const`, value under it is not
`int const *` - pointer is not `const`, but the value is
`int const * const` - both pointer and value are const
Reviewed By: jberdine
Differential Revision: D4962180
fbshipit-source-id: dcb02e3
Summary:
We were including hex of empty string if mangled name was not empty (so for all C++ functions).
Instead, include hex of a source file only if it's not empty
Reviewed By: mbouaziz
Differential Revision: D4705388
fbshipit-source-id: 55b6587
Summary:
Procnames files are now reversed qualifier lists with `#` as separator (instead of `::` which needs to be escaped in bash).
Because of the mechanism that is used to obtain qualifiers, it also affects naming for ObjC classes.
Examples:
```
std::unique_ptr<int>::get -> get#unique_ptr<int>#std#__MANGLED,...__ // C++ method
folly::split -> split#folly#__MANGLED,..._ // function within namespace
NSNumber numberWithBool: -> numberWithBool:#NSNumber#class // ObjC method
```
Reviewed By: jvillard
Differential Revision: D4689701
fbshipit-source-id: c3acfc6
Summary: These direct tests were still mostly relying on PHONY targets.
Reviewed By: jberdine
Differential Revision: D4326469
fbshipit-source-id: 37b2d0a
Summary: Globals that are constexpr-initializable do not participate in SIOF.
Reviewed By: sblackshear
Differential Revision: D4277216
fbshipit-source-id: fd601c8
Summary:
Currently cfg nodes are written into dot files in whatever order they
appear in a hash table. This seems unnecessarily sensitive, so this
diff sorts the nodes.
Reviewed By: dulmarod
Differential Revision: D4232377
fbshipit-source-id: a907cc6
Summary:
Dealing with symbolic links in project root is tricky. To avoid it, always normalize all paths to sources with `realpath`.
Changes to tests are expected - infer started to resolve symbolic links which screws up with our testing mechanism.
Reviewed By: jberdine
Differential Revision: D4237587
fbshipit-source-id: fe1cb01
Summary:
Run all clang tests with project-root at `infer/tests`. I need it because we'll start resolving symbolic links
soon and some tests would lead outside of project root which means we'd start seeing absolute paths in recorded tests.
Diff that does same thing for java tests: D4233236
Reviewed By: jberdine
Differential Revision: D4233194
fbshipit-source-id: c261a2b
Summary: These are dangerous if you are trying to compare a type to a string, and they're also unsightly.
Reviewed By: jvillard
Differential Revision: D4189956
fbshipit-source-id: 14ce127
Summary:
SIOF is only for interactions between objects of non-POD types. Previously the
checker was also reporting for POD types.
Reviewed By: akotulski
Differential Revision: D4197620
fbshipit-source-id: 7c56571
Summary:
- set SHELL to bash explicitly in Makefiles (Debian uses dash)
- avoid using system headers when using our own clang's headers in tests
- do not rely on the name of the object file to write the frontend debugging scripts. It turns out that `-o` is *not* always present in the arguments of `-cc1` functions so the `Option.get` could crash. Since we don't actually need to get the object file name, just a nice enough name, don't try to be smarter at guessing what object will be created and pick a different name built from the source name instead.
Reviewed By: akotulski
Differential Revision: D4159516
fbshipit-source-id: c7bc2b9
Summary:
In several places the tests were using whatever 'infer' executable was
found in PATH, instead of the one build from the source to be tested.
Reviewed By: jeremydubreil
Differential Revision: D4065019
fbshipit-source-id: 9b65099
Summary:
Create dummy functions representing the initializers of global variables. This
is so we can implement checks in the backend that can look at the initializer
expressions of global variables. We try not to create these dummy functions
when the initializer is not present, although for some reason we sometimes end
up with empty initializers.
Also add source file info to global variables in the backend (Pvar.re).
Reviewed By: sblackshear
Differential Revision: D3780238
fbshipit-source-id: 2dca87e
Summary:
It's not necessary if compiling tests in infer environemnt. It may be required if compiling some C++ tests
without infer. `infer/tests/codetoanalyze/cpp/shared/attributes/depracated_hack.cpp` is one of them
Reviewed By: cristianoc
Differential Revision: D4008850
fbshipit-source-id: 5d94bdf
Summary:
1. models no longer need access to private fields (shared_ptr needed that)
2. create macro for __attribute__((deprecated("__infer_replace_with_deref_first_arg"))) and use it in models
Reviewed By: jberdine
Differential Revision: D3791113
fbshipit-source-id: 532dd33
Summary:
1. Add capability to clang frontend to replace some function calls with another SIL code based on `__deprecated__` attribute.
2. Given this capability, use those attributes for shared_ptr getters to generate `Sil.Load` instruction instead of method call
3. Add test that mimics shared_ptr model, but it doesn't have that much scary C++ templated code
Reviewed By: jvillard, jberdine
Differential Revision: D3729176
fbshipit-source-id: 2a330d5
Summary:
Make std::shared_ptr<T> translated as T* inside infer. This will make reporting better
since smart pointers are really pointers not structs - this form is much easier for the analyzer to understand.
This requires changes to the model of shared_ptr as well.
Reviewed By: jvillard
Differential Revision: D3587255
fbshipit-source-id: b86fb36
Summary: With this approach, all the global consts will be inlined in the places where they are used.
Reviewed By: dulmarod
Differential Revision: D3703133
fbshipit-source-id: 3c19479
Summary:
Previously, we would translate `throw` with `return`. However, `throw` in
ObjC/C++ is often used to mean "abort". We now translate `throw` the same as
`exit` to prune these paths.
Reviewed By: akotulski
Differential Revision: D3594156
fbshipit-source-id: 81083bb
Summary:
No longer allow pointer types to be passed inside var_exp_typ. We used to accept both forms,
but it won't be possible any longer once shared_ptr becomes pointer type.
Reviewed By: dulmarod
Differential Revision: D3593003
fbshipit-source-id: a830914
Summary:
This call was producing confusing false positives when deleted object was possible to be null.
Changing frontend to add that check is not trivial so I turned it off for now (we don't handle
destructors in other cases anyway)
Reviewed By: dulmarod
Differential Revision: D3509354
fbshipit-source-id: c23dc81
Summary:
When clang instantiates template function with argument pack, it will
give the same name to all parameters coming from the pack. To avoid
name collisions, always add index of argument's position to mangled part
of the variable.
Seemingly unrelated changes are to make existing tests pass (don't use
simple variable name where it matters)
Reviewed By: dulmarod
Differential Revision: D3503608
fbshipit-source-id: 794093a
Summary:
Pass object by reference every time struct object is passed by value
in C++. Do it only for C++/objC++ where we have guarantee that the
object which is passed will be temporary one (created by copy constructor).
Reviewed By: jberdine
Differential Revision: D3346271
fbshipit-source-id: d3e5daa
Summary:Local variable created by conditional operator translation is now declared in scope of whole
procedure. Semantically there is no difference, hopefuly backend will not complain about this
change. Also, nullifying that variable is deferred to preanalysis instead of calling it manually
Reviewed By: jvillard
Differential Revision: D3155733
fb-gh-sync-id: 6cec8fc
fbshipit-source-id: 6cec8fc
Summary:BinaryConditionalOperator should evaluate condition expression once, but we used to evaluate it twice.
Fix translation to account for it.
Reviewed By: dulmarod
Differential Revision: D3179803
fb-gh-sync-id: a801a7e
fbshipit-source-id: a801a7e
Summary:This diff translate cpp lambdas. For the moment it does not take care of
captured variables. Captured variables will come in the next diff.
Reviewed By: dulmarod
Differential Revision: D3114790
fb-gh-sync-id: bf36450
fbshipit-source-id: bf36450
Summary:public
When a conditional is the last instruction, there will be a join node leading directly to the exit node.
Some instructions, such as nullification of dead variables, and abstraction, are added to the control flow graph automatically. But, join nodes cannot contain instructions. So when a procedure ends with a conditional, there might be no place to store these instructions.
This diff adds one extra node between the join and the exit node in that situation.
Reviewed By: jvillard
Differential Revision: D3179056
fb-gh-sync-id: 2b9cd7e
fbshipit-source-id: 2b9cd7e
Summary:public
Instead of translating code from headers blindly, translate only gets transitively referenced from source code.
It won't translate functions from system headers, but in the future we could do that as well
since most of them aren't used and it shouldn't add much overhead.
For now this functionality is hidden behind --cxx-experimental flag
Reviewed By: dulmarod
Differential Revision: D3163519
fb-gh-sync-id: 0c53b10
fbshipit-source-id: 0c53b10
Summary:public
Instead of using location of init_stmt, use location of variable when translating initialization.
Most of the time it change anything with some exceptions:
// example1 - C/C++/objC
int x = // now: assignment happens in this line
3; // past: assignment happens in this line
// example2: valid in C++11 only
struct X {
int x = 0; // now: one assignment here
int y = 2; // now: one assigmnent here
X() = default; // before: 2 assignments in this line
};
Reviewed By: dulmarod
Differential Revision: D3155870
fb-gh-sync-id: f38c78c
fbshipit-source-id: f38c78c
Summary:public
Add extra dereference when accessing fields that have T& type. It is similar
to what is done when accessing variables of T& type.
The only difference is that we need to handle constructor initializer list
separately (this is the only place where the field can be initialized)
Reviewed By: ddino
Differential Revision: D2965887
fb-gh-sync-id: 1b8708b
shipit-source-id: 1b8708b
Summary:public
Do same thing we do to CXXDefaultArgExpr
Reviewed By: dulmarod
Differential Revision: D2954128
fb-gh-sync-id: 2c92c16
shipit-source-id: 2c92c16
Summary:public
Translate headers every time they are included provided that they are located inside project_root directory.
While this is suboptimal (we might end up translating same header many times), doing it exactly once
is hard due to parallel compilation and template instantiations
Reviewed By: dulmarod
Differential Revision: D2916799
fb-gh-sync-id: 93b72c4
shipit-source-id: 93b72c4
Summary:
public
Avoid problems of overwriting good type information with incomplete information
when type declaration happens after its complete definition.
The solution is that we will only time we *update* type information is
when struct declaration has definition as well (which should happen once)
Reviewed By: cristianoc, sblackshear
Differential Revision: D2921811
fb-gh-sync-id: 16baba3
shipit-source-id: 16baba3
Summary:
public
This expression is used to value-initialize non-class types. Per definition of value initialization for non-class types:
1. If it's an array, value-initialize each of its elements
2. Otherwise, zero-initialize it
http://en.cppreference.com/w/cpp/language/value_initialization
I was unable to reproduce (1) in a way that produced CXXScalarValueInitExpr and so this diff
deals with case (2)
Reviewed By: jvillard
Differential Revision: D2901311
fb-gh-sync-id: beeafa2
Summary:
public
Add type of return parameter to the context. It allows for better translation
of returnStmt and will be necessary for easy implementation of constructor init lists
Reviewed By: cristianoc
Differential Revision: D2890838
fb-gh-sync-id: e791c3d
Summary:
public
xvalues is concept introduced in C++11. While they are not same as lvalues, they have one common trait:
They have identity which means that:
> it's possible to determine whether the expression refers to the same entity as another expression, such as by comparing addresses of the objects or the functions they identify (obtained directly or indirectly);
It means that as far as backend is concerned, they should be treated in same way. Right now there is no concept of "move" in the backend and so we don't have
to differentiate between them.
Reference:
http://en.cppreference.com/w/cpp/language/value_category
Reviewed By: cristianoc
Differential Revision: D2895593
fb-gh-sync-id: 5101e28
Summary:
public
C++ allows for parameters with empty names (unused/default copy constructors). Make backend happy by assigning a non-empty
name to these variables
Reviewed By: jvillard
Differential Revision: D2895550
fb-gh-sync-id: b466397
Summary:
public
It turns out that C-like structs in C++ may have methods generated by clang (constructors for example).
If struct has a method, it needs to have Sil.Class type - make all CXXRecordDecls Sil.Class types by default.
Reviewed By: cristianoc
Differential Revision: D2895567
fb-gh-sync-id: 8eb18c3
Summary:
public
Backend needs to know whether parameter has Derived* type - otherwise subtyping in backend doesn't work. Skipping `DerivedToBase` does that
Reviewed By: dulmarod
Differential Revision: D2890673
fb-gh-sync-id: a79abbc
Daiva Naudziuniene
Sam Blackshear
Jules Villard
Andrzej Kotulski
Jia Chen
Jeremy Dubreil
Dulma Churchill
Josh Berdine
Cristiano Calcagno
Martino Luca
Dino Distefano