* Remove duplicate namespace aliases so they don't produce syntax errors.
DACA2 results showed new SymbolDatabase syntax errors when duplicate
namespace aliases were simplified improperly. The solution is to remove
them in the tokenizer when found.
* Add tests for deleting namespace aliases at end of token list.
* Use eraseTokens to delete multiple tokens at once.
* Optimised simplifyKnownVariables
Changed the check for references to constant variables so instead of iterating through all tokens looking for references (which is very slow for large files), usages of each known variable are recorded so each usage can be checked for whether or not it is a reference. Just checking known usages is a lot quicker than checking through all tokens.
* Fixed test error caused by incorrect code
* Fixes to constant variable simplification optimisation
Indexing variables by varId is easier than by Token pointer, but means we also have to store a reference to the constant variable Token in another collection.
Switched to using unordered_map over map as it has slightly better find performance.
* Fixed incorrect simplification behaviour
This should remove constant variables correctly and efficiently. Requires additional functions, Token::deleteThisInPlace() and TokenList::front(Token*).
* Added setter for TokenList::first
This allows code that adds and removes Tokens from the list to do so without copying nodes into other nodes, which sometimes create difficulties.
* Added deleteThisInPlace function
This allows a token to delete itself without invalidating pointers to the node after it. If this token is the first or last in a list the calling code will have to remember to change the list's front or back.
* Added declaration for deleteThisInPlace()
* Removed premature MatchCompiler optimisation
* Added and removed some functions
Added declaration for deleteToken(Token*) which deletes a single token from the list
Removed public front(Token*) because it broke encapsulation
* Implemented deleteToken(Token*) function
* Removed 'delete this' from deleteThisInPlace
* Switched to using safer function
TokenList::deleteToken is better than calling straight into Token::deleteThisInPlace because it doesn't call delete this and doesn't break the TokenList's encapsulation.
* Replace constant variables in reverse order
This fixes the problem where you have two constant value assignment statements in a row. Replacing and deleting them in reverse order means we avoid the problem of deleteThis() potentially invalidating the pointer to the start of the next assignment statement
* Removed unneeded and unsafe deleteThisInPlace
* Removed unneeded and unsafe deleteThisInPlace
* Removed unneeded deleteToken
* Removed unneeded deleteToken
* Removed extra whitespace
* Don't remove the volatile keyword so we can properly overload functions.
I fixed all the checks that had tests that use volatile. There will
probably be more changes needed due to lack of test coverage for
volatile in some checks.
* Fix unused private function warning.
* Add support for namespace aliases and C++17 nested namespaces.
These are implemented as tokenizer simplifications so changes are not
needed to the tokenizer and symbol database.
* Fix codacy warning.
In 9cea2d6df, simplifications were removed for a number of functions
which should instead be handled with configurations. The commit did
however not update the description of the function, do this now.
Also sin() and sinh() and their float and long double versions were
missing from the comment so add these as well.
Remove TEXT() macro from windows.cfg and handle it internally where it
can be correctly simplified (Ansi vs. Unicode).
Also add handling of _TEXT() macro which is just a synonym for _T().
Add tests to verify correct function and macro simplification.
* Fix#8382 (Syntax error when scanning code with template and attribute)
This commit only addresses #8382. There are issues concerning which
versions of C++ should be supported and also generic C++ 14 attribute
support which can be revisited later.
* Remove all C++ style attributes.
Remove all C++ style attributes when C++ version is 11 or greater.
Rename simplify function to simplifyCPPAttributes.
Handle more cases of roreturn function attribute.
The following snippet triggerd the error:
template<typename DerivedT>
template<typename T>
auto ComposableParserImpl<DerivedT>::operator|( T const &other ) const -> Parser {
return Parser() | static_cast<DerivedT const &>( *this ) | other;
}
Whenever simplifyFunctionParameters was called on a templated class'
templated member function (and probably any nested template), the
tokenizer would recognise it as a syntax error, assuming that return
type *must* come after a template<> token.
* Fixes issue with case inside switch that is not a compound statement was treated as garbage
This fixes an issue with the check for case keywords outside of switch
detection that would treat a case statement inside a switch that is not
a compound statement as garbage, but this is perfectly valid C++. This
construct is used in several libraries, i.e. Google Test.
* Tweak check and handle missing semicolon
Tweaks the check with feedback from danmar.
Handle the case where there is no semicolon and document it with a unit
test.
* Add support for simple c++ 11 type ailases like: using INT = int;
Only types supported by ValueType are supported. Complex types like
function pointers are not supported. Template type aliases are not
supported.
* Fix crash when type in using type alias is simplified away.
This fixes a crash when size_t is replaced with unsigned long in: using
size_t = unsigned long; by the tokenizer.
This does not fix the problem where Tokenizer::simplifyPlatformTypes()
simplifies away size_t in other cases. This is only a problem when the
new type is different from the platform type.
* fix#8284: False positive: "Label 'class' is not used." for anonymous C++ class
Add support for annonymous derived structures and classes.
* Fix travis build (use findsimplematch).
* Fix bug in simplifyLabelsCaseDefault which was inserting ; in wrong place.
* fix for correct parsing of embedded PL/SQL blocks (Oracle Pro*C)
* enforce SQL block end at the end of nearest outer C block, when
appropriate terminator is not found
* added check for ; at the end of END-EXEC and made SQL block detection
more readable
All issues were found with PVS-Studio:
V560 A part of conditional expression is always true: tok. astutils.cpp 407
V560 A part of conditional expression is always true: size > 0. checkbufferoverrun.cpp 709
V547 Expression 'secondTrue' is always true. checkcondition.cpp 1013
V547 Expression 'firstTrue' is always true. checkcondition.cpp 1020
V560 A part of conditional expression is always true: !scan. checkio.cpp 1036
V560 A part of conditional expression is always true: scope->function. checknullpointer.cpp 395
V560 A part of conditional expression is always true: tok2. checkstl.cpp 268
V560 A part of conditional expression is always true: par. tokenize.cpp 9440
V547 Expression '!erased' is always true. symboldatabase.cpp 3990
IOBYTE
PKEuS
Simon Martin
Daniel Marjamäki
rebnridgway
rikardfalkeborn
Sebastian
PKEuS
jrp2014
Lauri Nurmi
Alexander Mai
Matthias Krüger
Jørgen Kvalsvik
David Hallas
Martin Ettl
Alexey Eryomenko
Dmitry-Me
Simon Martin
Oleksandr Redko
Ayaz Salikhov
Ayaz Salikhov