- Fixed bug in library: manual and existing libraries use "size", but library.cpp reads "sizeof" as podtype attribute
- Fixed a couple of bugs in handling unknown size in checkbufferoverrun.cpp, get size from library if available.
- Removed a few unit tests of old uninitialized variable checking expecting other results than same test for new checking
- A few tweaks to new uninitvar checking to improve results when run on tests for old check
- switched some (TODO) tests to new check if the TODO is fixed
- Implement shortcut for C code - all types need initialization there
- Break out of loop faster if we encounter a type that needs initialization (it is sufficient if one member needs initialization)
Introduce a new bool setting jointSuppressionReport
that will be set by the analyseWholeProgram() code path.
When the flag is enabled, unmatched suppressions are
collected after running the final whole program analysis
to prevent false positives for the unusedFunction check.
The check functions in the unit test
for single / multi file suppressions were unified.
When something parsing the template syntax
went wrong, "tok2" was NULL and resulted
in deleting all remaining tokens. Whoops.
Triggered by gcc test suite:
gcc/testsuite/g++.dg/cpp0x/variadic87.C
Tracked down the source of the bug with
valgrind's "--track-origins=yes" switch.
Doing "&*some_ptr_var" is redundant and might be the remainder
of a refactoring. Warnings for expanded macros are excluded though:
They are often used with and without pointers and
do something like this: "func(&(*macroarg))".
The new check is fully AST based and was given
strong false positive testing on a large code base.
Frank Zingsheim reported: With the parameter 'flag_'
and the member variable '_flags' having the type 'int'
produces a lot of warings in clang:
"warning: implicit conversion changes signedness: 'int' to 'unsigned int'".
PKEuS
PKEuS
Daniel Marjamäki
Dmitry-Me
Samir Aguiar
Thomas Jarosch
orbitcowboy
Robert Reif