* Tokenize: Set varId for variables in enum
Set varIds in enum values. It was previously disabled in 5119ae84b8
to avoid issues with enums named the same as global variables. Take care
to only set varids to variables used to set the value of an enumerator,
not the enumerator itself. This is somewhat complicated by the fact that
at the time this happens, astOperand1(), astOperand2(), astParent() etc
are not set. The current implementation is not perfect, for example in
the code below, y will not have a varid set, but x and z will. This is
deemed sufficient for now.
int x, y, z;
enum E { a = f(x, y, z); };
* Fix#9647: Value of enums with variables as init values
C++ allows enum values to be set using constexprs, which cppcheck did
not handle before. To solve this, add a new pass to valueflow to update
enum values after global consts have been processed. In order to do so,
I moved all settings of enum values to valueflow. After setting the enum
values, we need another call to valueFlowNumber() to actually set users
of the enums.
There is still room for improvements, since each pass of
valueFlowGlobalConstVar() and valueFlowEnumValue() only sets variables
that are possible to set directly, and not if setting the value of a
variable allows us to set the value of another. For example
constexpr int a = 5;
constexpr int b = a + 5;
enum E { X = a };
constexpr E e = X;
Here both b and e will not have their values set, even though cppcheck
should be possible to figure out their values. That's for another PR
though.
This was tested by running test-my-pr.py with 500 packages. The only
difference was one error message in fairy-stockfish_11.1, where cppcheck
now printed the correct size of an array instead of 2147483648 which I
assume is some kind of default value. In that package, using a constexpr
when setting enum values is common, but as mentioned, there was no
change in the number of warnings.
On some platforms, the 'd_name' field of struct dirent is not a static
fixed-sized array but a "flexarray" (i.e. a single character); in this
situation, 'd_name' points to a buffer allocated somewhere, usually
at the end of the buffer used for dirent (which is then allocated in a
bigger memory). Because of this, creating a struct dirent on stack as
buffer for readdir_r is not enough to store all the memory needed for
a dirent on those platforms.
As result, create an helper union with all the needed space, calculated
statically at build time. NAME_MAX+1 is still not a perfect option, but
it will do the job in the vast majority of cases.
Previously, if an enum value was set to a value unknown to cppcheck, the
next enum value would erroneously be set to the last set value plus one
(or zero, if no enum value had been set before). This partially fixes
Trac ticket #9647, in the sense that it no longer sets wrong values for
these enum values. Further improvements to this would be to set the
correct values instead. It also fixes the false positive mentioned in the
comments in the ticket.
- gui.pro: Update the path of the output file, so that it is also built
inside the help sub-directory (add $$PWD/help/ to the path):
-o $$PWD/help/online-help.qhc
- Use qcollectiongenerator in case of Qt version < 5.12, otherwise
qhelpgenerator
- appveyor.yml: Add %QTDIR%\bin to PATH. This allows qhelpgenerator
or qcollectiongenerator to be found
Fix return value types of library functions returning unsigned.
Previously, the valueType of auto x = f() would be signed even if f()
was specified to return an unsigned type.
This fixes#9941, which is a regression in cppcheck 2.2 compared to 2.1.
The regression was introduced in 32df807b22.
Daniel Marjamäki
Wolfgang Stöggl
Even Rouault
miltolstoy
amai2012
Rikard Falkeborn
Paul Fultz II
Ken-Patrick Lehrmann
Pino Toscano
lakshit24
Armin Müller
orbitcowboy
Georgy Komarov
Steve Browne