This will now warn for cases like this:
```cpp
auto& f() {
std::vector<int> x;
return x[0];
}
```
It also improves the handling of address of operator, so it can now warn across some function calls, like this:
```cpp
int& f(int& a) {
return a;
}
int* hello() {
int x = 0;
return &f(x);
}
```
Even if `ptr` is a local variable, the object `ptr->item` might be not.
So taking address of `ptr->item` is definitely not unsafe in general.
This commit fixes false positives triggered by commit
1.85-249-gf42648fe2 on the following code of sssd:
https://github.com/SSSD/sssd/blob/d409df33/src/sbus/request/sbus_request.c#L359
This reworks constStatement to find more issues. It catches issue [8827](https://trac.cppcheck.net/ticket/8827):
```cpp
extern void foo(int,const char*,int);
void f(int value)
{
foo(42,"test",42),(value&42);
}
```
It also catches from issue [8451](https://trac.cppcheck.net/ticket/8451):
```cpp
void f1(int x) {
1;
(1);
(char)1;
((char)1);
!x;
(!x);
~x;
}
```
And also:
```cpp
void f(int x) {
x;
}
```
The other examples are not caught due to incomplete AST.
Add a call to simplifyPlatformTypes() in
SymbolDatabase::setValueTypeInTokenList() to simplify return types of
library configured functions. This fixes the FN in #8141. Regression
tests are added, both for the original issue and another FN in the comments.
In order to do that, move simplifyPlatformTypes() to TokenList from Tokenizer.
This is a pure refactoring and does not change any behaviour. The code was
literally copy-pasted from one file to another and in two places
'list.front()' was changed to 'front()'.
When adding the call to simplifyPlatformTypes(), the original type of
v.size() where v is a container is changed from 'size_t' to 'std::size_t'.
Tests are updated accordingly. It can be noted that if v is declared as
'class fred : public std::vector<int> {} v', the original type of 'v.size()'
is still 'size_t' and not 'std::size_t'.
* Fixed#8962 ("(debug) Unknown type 'T'" with template typename parameter)
Only simple one parameter template functions with one function parameter
are supported.
* Added TODO test case for FIXME.
* Fixed#8971 ("(debug) Unknown type 'x'." using alias in class members)
* template simplifier: partial fix for #8972
Add support for multi-token default template parameters.
* template simplifier: fix for #8971
Remove typename outside of templates.
Qt defines `Q_NULLPTR` with `nullptr` if it is available, otherwise with `NULL`.
Since there seems to be no (sane) way to configure it the same way in the library configuration it is just defined with `NULL`.
* Fixed#8960 ("(debug) Unknown type 'x'." with alias in template class alias)
This commit adds non-template type alias support to the template
simplifier. Only relatively simple type aliases are supported at this
time. More complex types will be added later.
--debug-warnings will show unsupported type aliases.
Type alias support will be removed from the symbol database in the
future. Type alias tests have been removed from the symbol database
tests.
* Add the changes.
* Fix codacy warning.
* Fix travis warnings.
* template simplifier: fix crash on windows
Use right token when searching for template type alias to delete.
* template simplifier: fix a cppcheck warning
* Remove newlines after check(
* Remove unneeded statements after if-statements
As an example, the previous test case
check(
"bool foo(int x) {\n"
" if (x < 0)"
" return true;\n"
" return false;\n"
"}");
is changed to
check("void foo(int x) {\n"
" if (x < 0) {}\n"
"}");
This has basic handling of GUI projects. But further work will be needed to handle addons etc, the plan is that we will be able to run addons from the command line soon.
The unsigned less than zero checker looked for patterns like "<= 0".
Switching to use valueflow improves the checker in a few aspects.
First, it removes false positives where instead of 0, the code is using
0L, 0U, etc. Instead of having to hard code the different variants of 0,
valueflow handles this automatically. This fixes FPs on the form
uint32_t value = 0xFUL;
void f() {
if (value < 0u)
{
value = 0u;
}
}
where 0u was previously not recognized by the checker. This fixes#8836.
Morover, it makes it possible to handle templates properly. In commit
fa076598ad, all warnings inside templates
were made inconclusive, since the checker had no idea if "0" came from
a template parameter or not.
This makes it possible to not warn for the following case which was
reported as a FP in #3233
template<int n> void foo(unsigned int x) {
if (x <= n);
}
foo<0>();
but give a warning for the following case
template<int n> void foo(unsigned int x) {
if (x <= 0);
}
Previously, both these cases gave inconclusive warnings.
Finally, it makes it possible to give warnings for the following code:
void f(unsigned x) {
int y = 0;
if (x <= y) {}
}
Also, previously, the checker for unsigned variables larger than 0, the
checker used the string of the astoperand. This meant that for code like
the following:
void f(unsigned x, unsigned y) {
if (x -y >= 0) {}
}
cppcheck would output
[unsigned-expression-positive.c] (style) Unsigned variable '-' can't be negative so it is unnecessary to test it.
using expressionString() instead gives a better error message
[unsigned-expression-positive.c] (style) Unsigned expression 'x-z' can't be negative so it is unnecessary to test it.
Use `--check-library` for all tests as it was done before.
Re-enable all tests in runtests.sh again.
The regressions where runtests.sh would fail are disabled via "FIXME"
comment in the inline suppression comment.
* Add regression test for #6906
Ticket #6906 was fixed in f65cf220ba.
Add a test to make sure there are no regressions.
* Add regression test for #7284
Ticket #7284 was fixed in 5d1fdf7958.
Add tests to avoid regressions.
This will use the lifetime checker for dangling references. It will find these cases for indirectly assigned reference:
```cpp
int &foo()
{
int s = 0;
int& x = s;
return x;
}
```
This will also fix issue 510 as well:
```cpp
int &f( int k )
{
static int &r = k;
return r;
}
```
As discussed in https://trac.cppcheck.net/ticket/8931 a regression test is added
to the test/cfg/runtests.sh script to make sure that unmatchedSuppression messages result in an Cppcheck exit code that signals a failure.
On linux systems (like travis) Qt often seems to be built with the option "reduce_relocations" which requires an application using it to specify the option "-fPIC".
* fix for CMake compile_commands.json input - director does not include trailing / which makes include directories wrong - so add it if it doesnt exist
* fix the bugfix for trailing / in the directory name of CMAKE JSON file, add also new test case to see if it works in both cases (with and without trailing /)
* revert adding accidental new line
To be able to use real Qt-Code in "test/cfg/qt.cpp" and still do a
syntax check the Qt settings are read out via pkg-config now if it is
available. This way the test now can contain Qt macros and functions and
the syntax check can still be used.
Additionally the same options as for the other tests are used now for
the Qt config tests.
Installing the package "qtbase5-dev" should be enough to enable the
syntax checks (already installed for travis tests).
This fixes valueflow to have a value for `||` operator here:
```cpp
bool f()
{
bool a = (4 == 3); // <-- 0
bool b = (3 == 3); // <-- 1
return a || b; // <-- 1
}
```
When comparing if the shift is large enough to make the result zero, use
an unsigned long long to make sure the result fits. Also, a check that
avoids setting the value if the shift is equal to or larger than the
number of bits in the operand (this is undefined behaviour). Finally,
add a check to make sure the calculated value is not too large to store.
Add test cases to cover this.
This was detected by an MSVC warning.
valueflow.cpp(1350): warning C4334: '<<' : result of 32-bit shift implicitly
converted to 64 bits (was 64-bit shift intended?)
This fixes issue in:
```cpp
void f()
{
char stack[512];
RGNDATA *data;
if (data_size > sizeof (stack))
data = malloc (data_size);
else
data = (RGNDATA *)stack;
if ((char *)data != stack)
free (data); // <- data is not stack
}
```
It seems the `ProgramMemory` can't handle two known values(such as int and tok) together. So instead `ValueFlowAfterAssign` runs `ValueFlowForward` with tok values and then runs it with the other values.
I sometimes find myself wondering which test cases I have broken when I fiddle with some check (since I then can try running these specific tests in the debugger, or make a small change and see if the tests pass). This PR adds the testclass and the test case names to the file and line number. I took special care to cover the case where an assert would be placed directly in the ```run()```-function, i.e., not inside a test case (from what I could see, no such case exists). If there is no need to handle this case, the code can be simplified (there wouldn't be need for the ```teardownTest()```-function for example).
The exact format for how to print the test name is very much up for discussion.
This makes arrays non-null in valueflow, so it can catch comparisons against null that is always true:
```cpp
void f(void) {
int buf[42];
if( buf != 0) {;} // << always true
}
```
Add a check for function arguments that can be constant:
```cpp
extern void bar(int);
void f(int x) {
bar((x & 0x01) >> 7); // function 'bar' is always called with a '0'-argument
}
```
* Implement hexadecimal float conversion in MathLib::toDoubleNumber independent of C99/C++17
* Refactor MathLib::isFloatHex and cure some false results
Paul Fultz II
orbitcowboy
versat
Kamil Dudka
rikardfalkeborn
IOBYTE
Daniel Marjamäki
Simon Martin
Lauri Nurmi
amai2012
juremenart
practicalswift
amai2012
Igor