Summary:
**Existing heuristic**: If we have a call `foo(n)` that has no model and summary for `foo`, we underestimate its cost as constant[1].
However, if we have a model for `foo` (e.g.with modeled cost O(n)) but applying the model to arguments causes the cost to be Top (e.g if `n` has Top size), then we could have Top-poisoning where all the callers up the call chain will have Top costs [2].
To prevent these unintended Top-poisioning when adding models, this diff applies *the same heuristic* to modeled calls with Top cost and gives them constant cost. This way, when adding models, we wouldn't be introducing more Tops than if we were to have no models in the first place.
[1] This is problematic in itself and causes many FPs at diff time, but otherwise we would be getting Tops everywhere and would not be able to give any meaningful cost. E.g. for fblite, if we were to give unknown calls Top cost, #procedures with Top cost increases form 5% to 38% and #procedures with linear cost reduces by 99.75%.
[2] This was observed for `containsValue` for Instagram where %Tops increased by 88% :(
Reviewed By: skcho
Differential Revision: D26174644
fbshipit-source-id: 232354923
Summary:
This is several inter-connected changes together to keep the tests
happy.
The ConditionalOperator `b?t:e` is translated by first creating a
placeholder variable to temporarily store the result of the evaluation
in each branch, then the real thing we want to assign to reads that
variable. But, there are situations where that changes the semantics of
the expression, namely when the value created is a struct on the stack
(eg, a C++ temporary). This is because in SIL we cannot assign the
*address* of a program variable, only its contents, so by the time we're
out of the conditional operator we cannot set the struct value
correctly anymore: we can only set its content, which we did, but that
results in a "shifted" struct value that is one dereference away from
where it should be.
So a batch of changes concern `conditionalOperator_trans`:
- instead of systematically creating a temporary for the conditional,
use the `trans_state.var_exp_typ` provided from above if available
when translating `ConditionalOperator`
- don't even set anything if that variable was already initialized by
merely translating the branch expression, eg when it's a constructor
- fix long-standing TODO to propagate these initialization facts
accurately for ConditionalOperator (used by `init_expr_trans` to also
figure out if it should insert a store to the variable being
initialised or not)
The rest of the changes adapt some relevant other constructs to deal
with conditionalOperator properly now that it can set the current
variable itself, instead of storing stuff inside a temp variable. This
change was a problem because some constructs, eg a variable declaration,
will insert nodes that set up the variable before calling its
initialization, and now the initialization happens *before* that setup,
in the translation of the inner conditional operator, which naturally
creates nodes above the current one.
- add a generic helper to force a sequential order between two
translation results, forcing node creation if necessary
- use that in `init_expr_trans` and `cxxNewExpr_trans`
- adjust many places where `var_exp_typ` was incorrectly not reset when translating sub-expressions
The sequentiality business creates more nodes when used, and the
conditionalOperator business uses fewer temporary variables, so the
frontend results change quite a bit.
Note that biabduction tests were invaluable in debugging this. There
could be other constructs to adjust similarly to cxxNewExpr that were
not covered by the tests though.
Added tests in pulse that exercises the previous bug.
Reviewed By: da319
Differential Revision: D24796282
fbshipit-source-id: 0790c8d17
Summary:
Namely `enumerateObjectsUsingBlock` which takes a collection and a block as an arg and the iterates over the collection and applies the block to each element of the collection.
This is a common way to iterate over Objc collections, so let's add a model for it.
Reviewed By: skcho
Differential Revision: D24604590
fbshipit-source-id: 1ceeb4b40
Summary:
This diff revises the translation of message expression's arguments in ObjC frontend. In the
frontend, it massages the arguments when calling a static method, so the class or object value is
not given to the static method as the first parameter.
The problem is that it used a raise-exception-and-catch way to detect where we remove the first
parameter. This way of using an exception is not only hard to understand, but also incorrectly
removed the first parameter, with breaking abstract semantics sometimes. (See the added test.) This diff
avoids using the exception.
Reviewed By: jvillard
Differential Revision: D24565513
fbshipit-source-id: 0a84ca394
Summary: This diff fixes on-demand symbolic value generation of a class that inherits NSEnumerator.
Reviewed By: ngorogiannis
Differential Revision: D24504955
fbshipit-source-id: bcb20e8aa
Summary:
This diff replaces overridden method calls in ObjC when possible, ie the first parameter of the
method has a sub-class type of the method's class. For example,
when `MyEnumerator` is a sub-class of `NSEnumerator` and there is overridden `nextObject`,
```
[my_enumerator nextObject]
```
in Sil, it was translated to like
```
NSEnumerator.nextObject(my_enumerator : MyEnumerator*)
```
and the analyzer missed the overridden method. This diff replaces the function call to
```
MyEnumerator.nextObject(my_enumerator : MyEnumerator*)
```
Reviewed By: ezgicicek
Differential Revision: D24477290
fbshipit-source-id: 6842a76f8
Summary:
This diff revises `nextObject` model to handle multiple symbolic enumerators. Instead joining the
symbolic offsets of them, which sometimes introduces top, it sums the offsets. This is a sound &
conservative semantics since they are all non-negative integers.
Reviewed By: ezgicicek
Differential Revision: D24474513
fbshipit-source-id: 6707aa907
Summary:
This diff revises memory model of enumerator in ObjC to enable passing it as a parameter.
The cost checker was not able to analyze a function precisely when it gets an enumerator as a
parameter because the offset of an enumerator was available only when the analyzer knew the correct
relation between the enumerator and an array.
This diff simplifies the enumerator to have a similar value with `array->elements`, so its offset can
be taken without the relation between enumerator and array to get them.
Reviewed By: ezgicicek
Differential Revision: D24446574
fbshipit-source-id: 27cdc051e
Summary:
This diff adds an option hiding function pointers in costs to users: `cost-suppress-func-ptr` is
true by default.
Reviewed By: ezgicicek
Differential Revision: D24448212
fbshipit-source-id: 88f6b5ea1
Summary:
This diff adds a model for NSFileManager.contentsOfDirectoryAtURL as returning a constant-length
collection.
The analyzer cannot know files in a directory. We have some options to handle such unknown data.
1. Use `Unknown` value, ie `top`
2. Use a symbolic value
3. Use a constant value
We had been used the first option. An upside of this is that the analyzer can remain as sound.
However, a downside of this is the top value can be propagated to other procedures, making their
costs top, thus we may miss some cost changes of them.
The second option is to introduce a symbolic value, ie. that for the number of files. A problem is
that the symbolic value will never be concretized. As a result, the symbol can be propagated to
other procedures, increasing the coefficient of the complexity or making top costs. Note that handling multiple
symbols is somewhat limited in Inferbo's interval domain.
The last option is to introduce a constant value. I think this is the best approach we can take among above.
Even though we may have FNs when there are a lot of files in a directory, we cannot reason or expect about
that at the analysis time anyway.
Reviewed By: ezgicicek
Differential Revision: D24418099
fbshipit-source-id: bf8cf3538
Summary: This diff adds closure symbols to operation/allocation costs, when function pointer is called.
Reviewed By: ezgicicek
Differential Revision: D24308550
fbshipit-source-id: 6c5889d41
Summary:
In ObjC, `NSObject.copy` returns the object returned by `copyWithZone:` on the given class. This method must be implemented if the class complies with `NSCopying` protocol. Since we don't have access to `NSObject`'s code, to follow calls into `copyWithZone:`, we replace such `copy` calls with calls to `copyWithZone:` when a) such a method exists in the class and b) the class conforms to `NSCopying` protocol.
This is done in the preanalysis because
- we need to know if there is a `copyWithZone:` method in the class.
- so that other analyses also benefit (as opposed to doing this in cost and inferbo models).
Note that `NSObject` doesn't itself conform to `NSCopying` but all its subclasses must confrom to the protocol and support the same behavior as above.
https://developer.apple.com/documentation/objectivec/nsobject/1418807-copy
Similarly for `mutableCopy` -> `mutableCopyWithZone:` for classes implementing `NSMutableCopying` protocol.
Reviewed By: skcho
Differential Revision: D24218102
fbshipit-source-id: 42900760e
Summary: Model it similar to `NSArray.initWithArray` as copying from the given dictionary elements. Removes a FP as expected.
Reviewed By: ngorogiannis
Differential Revision: D24136868
fbshipit-source-id: ed31c3c8f
Summary: Dispatch & function call mechanism was so jumbled up together. Let's refactor it to be cleaner.
Reviewed By: skcho
Differential Revision: D24049889
fbshipit-source-id: 42a218016
Summary:
The problem: current enumerator semantics does not work on symbolic enumerator that is given as a
parameter.
Reviewed By: ezgicicek
Differential Revision: D24017059
fbshipit-source-id: 378e75bb0
Summary:
This diff substitutes closure parameter when it is given via variable. For example,
```
x = ^{ ... }
foo(x)
```
this diff substitutes the closure variable `x`,
```
x = ^{ ... }
foo(^{ ... })
```
so that the specialization of `foo` can be done by `CCallSpecializaedWithClosures.process`.
Reviewed By: jvillard
Differential Revision: D23814595
fbshipit-source-id: a89f1530f
Summary:
This diff reports paths under the xcode isysroot as relative in tests.
This was a problem when another machine that has a different isysroot
directory is running the test.
Reviewed By: ezgicicek
Differential Revision: D23729222
fbshipit-source-id: 4e9681f65
Summary:
As title.
Facebook
Found this case by examining db.
>
D23394545
Time complexity of `_registerTabPreloadables` has **decreased** from `Top` to `O(1)`. Please make sure this is an expected change. You can inspect the trace to understand the complexity increase:
Reviewed By: ezgicicek
Differential Revision: D23448099
fbshipit-source-id: 108fd1ef2
Summary:
This diff adds translation of `dictionaryWithObjects:forKeys:count:`. In the previous implementation it was
translated as if it was `dictionaryWtihObjectsAndKeys:`, but their function parameters are different.
In this diff, it translates an array literal `NSDictionary* a = @ [ @"firstName": @"Foo", @"lastName":@"Bar" ];` to
```
n$1=NSString.stringWithUTF8:(@"firstName")
n$2=NSNumber.stringWithUTF8:(@"Foo")
n$3=NSNumber.stringWithUTF8:(@"lastName")
n$4=NSNumber.stringWithUTF8:(@"Bar")
temp1[0]:objc_object*=n$1
temp1[1]:objc_object*=n$3
temp2[0]:objc_object*=n$2
temp2[1]:objc_object*=n$4
n$3=NSDictionary.dictionaryWithObjects:forKeys:count:(temp2:objc_object* const [2*8], temp1:objc_object*const [2*8], 2:int)
```
where `temp` is an additional local variable declared as array.
See,
https://developer.apple.com/documentation/foundation/nsdictionary/1574184-dictionarywithobjects?language=objchttps://developer.apple.com/documentation/foundation/nsdictionary/1574181-dictionarywithobjectsandkeys
{F316854542}
Reviewed By: skcho
Differential Revision: D23447042
fbshipit-source-id: 14b7c3f2b
Summary:
This diff finishes the migration from the specialization of methods that take blocks as arguments. Here we delete all the old code and change the way we model dispatch functions so that the tests pass.
- Remove the code for specializing the methods in biabduction.
- Remove the call flags `cf_with_block_parameters` that was only used in this algorithm.
- Removes models for dispatch functions.
- Adds models for dispatch functions as program transformation only in biabduction. To be added in other checkers in the future.
Reviewed By: ngorogiannis
Differential Revision: D23345342
fbshipit-source-id: b5e8542ce
Summary:
Two issues are fixed
1. For this diff, when the condition `curr_langauge_is Java` is removed in some part of the analysis, some c bufferoverrun tests are broken. This is fixed in this diff by inspecting if we are dealing with a `Size` alias referring to an `objc_internal_collection_array`.
2. Previously, when we are modifying mutable array in a loop, e.g. adding element to the array or removing element from the array, we are unable to give an estimable size of the array after exiting the loop. This is now fixed, and the corresponding FPs are resolved.
Reviewed By: skcho
Differential Revision: D23342350
fbshipit-source-id: 200f5261c
Summary:
In the previous diffs, we implement enumerator in order to estimate the cost of for-each loop in ObjC, but when we have FP case when enumerator is used not in for-each loop. For example, the following code has top cost before the fix.
```
void nsarray_enumerator_linear_FP(NSArray* array) {
id obj;
NSInteger sum = 0;
NSEnumerator* enumerator = [array objectEnumerator];
while (obj = [enumerator nextObject]) {
sum += (NSInteger)obj;
}
}
```
Reviewed By: skcho
Differential Revision: D23294895
fbshipit-source-id: 50c7b359f
Summary:
Fix the FP when iterating through constant collection.
facebook
This fix is a hack for now.
Reviewed By: ezgicicek, skcho
Differential Revision: D23241338
fbshipit-source-id: e2e0c05f8
Summary:
As title.
This diff is co-authored by SungKeun Cho and me.
facebook
This diff is co-authored by skcho and me.
Original comments from skcho
"For the record:
1. Rory and I tried to write models for ObjC iterator.
2. We could not use Java's iterator semantics: In Java's, `hasNext` returns the size of collection, rather than a boolean, and which is used as a control variable. On the other hand, in ObjC, it calls only `nextObject`, not calling `hasNext`, and the return value of which is being checked as `null`.
3. We added an artificial field `objc_iterator_offset` to keep the index of the iterator, and the models added in this diff are handling that integer value.
A problem is that `array.objc_iterator_offset` is not included in the control variables, since the condition of the loop is `nextObject() != null` that does not include the iterator offset. We need to make `array.objc_iterator_offset` as a control variable, by changing the part collecting control variables.
"
Reviewed By: ezgicicek, skcho
Differential Revision: D22944278
fbshipit-source-id: 7e71b79c1
Summary:
When implementing iterator, we find out that because some semantics of inferbo is Java-specific, we cannot simply use Java's `Collection` model for `NSCollection`.
So this diff writing `NSCollection` model separately.
This diff also extracts the common parts of `NSCollection` and `Collection` into `AbstractCollection` to combine the duplicate the parts.
Reviewed By: ezgicicek
Differential Revision: D22975159
fbshipit-source-id: daed3f99f
Summary: Before, `NSCollection` are modelled like array, but this will have issue when we want to say add object to array. This diff changes `NSCollection`'s model to use Java's Collection model.
Reviewed By: ezgicicek
Differential Revision: D22840079
fbshipit-source-id: b944b743b
Summary: Implement `alloc` and implement initialisation method that uses the return value of `alloc`, i.e. `NSString.init`.
Reviewed By: ezgicicek
Differential Revision: D22840080
fbshipit-source-id: 47a7523e3
Summary:
This diff translates for-in block in objc as a simple for-loop. For example,
`for (item_type item in items) { body }` is translated to
```
NSEnumerator *enumerator = [items objectEnumerator];
item_type item;
while (item = [enumerator nextObject]) { body }
```
Reviewed By: ezgicicek
Differential Revision: D22841524
fbshipit-source-id: 296ee84df
Summary:
We check for supertypes in Java. Why not ObjC?
Would be good to get dulmarod's input here.
Reviewed By: roro47
Differential Revision: D22817126
fbshipit-source-id: 52c1c3f3c
Ezgi Çiçek
Jules Villard
Sungkeun Cho
Qianyi Shu
Dulma Churchill