Add test cases for assign operators with default implementation

[dan-42]

👋 @danmar

Sorry to bother you again 🙃

So I have found another issue, and after cutting down my code this fails

struct foobar
{
  auto operator=(const foobar&) -> foobar&;
};
auto foobar::operator=(const foobar&) -> foobar& = default;

with:

$ cppcheck --enable=all "foobar.cpp"
Checking foobar.cpp ...
foobar.cpp:0:0: error: Internal Error: Invalid syntax [cppcheckError]
^
foobar.cpp:0:0: note: Internal Error: Invalid syntax
^
foobar.cpp:0:0: note: Internal Error: Invalid syntax
^

• adding { } instead of default works
• adding default inside the struct definition works
• changing it to the classical form foobar& foobar::operator=(const foobar&) = default; works

So I added two test cases to reproduce the issue.
cppcheck failes here:

void CheckClass::initializeVarList(const Function &func, std::list<const Function *> &callstack, const Scope *scope, std::vector<Usage> &usage)
{
    if (!func.functionScope)
        throw InternalError(nullptr, "Internal Error: Invalid syntax"); // #5702
    bool initList = func.isConstructor();

call comes from this function CheckClass::constructors()
and on line 147 it is checked:

   for (const Function &func : scope->functionList) {
            if (!func.hasBody() || !(func.isConstructor() || func.type == Function::eOperatorEqual))
                continue;

So I wonder why the check for cpp func.type == Function::eOperatorEqual

So the issue boils down that func.functionScope is nullptr

Could you point me in a direction, is this more of a tokenizing issue?

[dan-42]

deleted, because this was the wrong direction for this investigation

[dan-42]

@danmar I added my new findings see above

[dan-42]

Ok maybe not I just saw the xml attribute originalName="->" 😞

[danmar]

Yes the token is ".".. well that is a very old decision that made more sense before we had the AST. I think we can consider using "->" someday again but if somebody wants to change it then it must be done carefully.

[dan-42]

@danmar So I found the issue and indeed it was in the symboldatabase
and the "default" was not detected when using trailing return types.

[dan-42]

@danmar I'm pretty sure this can be done in a better way right?
and it also does not cover all cases yet.
So I'm hoping I will get some help here, please.

[danmar]

well a start could be:

Token::Match(closeParen, ") . const| %type% &| = default ;"))

but not sure if that is the ultimate solution, maybe we have some code to handle trailing return type and stuff..

[danmar]

there is some copy/paste problems in the SymbolDatabase. Maybe we should have some utility function to parse function attributes . There is a loop in Function::Function below a comment // parse function attributes that looks promising. I am spontanously wondering if that code can be broken out and reused somehow.

But.. the Token::Match pattern I provided could be a quick fix I could merge that as a start.. if you add a proper test in TestSymbolDatabase also. And then anybody can look at the refactoring later in a separate PR.

[danmar]

another way is to reuse Tokenizer::isFunctionHead.. something like:

const Token *endtok = mTokenizer->isFunctionHead(closeParent->link(), ";");
if (endtok && Token::simpleMatch(endtok->tokAt(-2), "= default ;"))
.....

.. well we have some options. I believe a small fix to start with, with a proper test and a Token::Match is a good first step.

As far as I see, isFunctionHead does not handle trailing return types but it's very important that it can handle those.

[dan-42]

Alright, thank you I will clean it up and add a proper test for this.
maybe I'll find the time tomorrow evening

[dan-42]

@danmar So here you go.
I cleaned it up and added a testcase for the symboldatabase.
Hope this is what you had in mind?

[danmar]

Have you been added to our AUTHORS file? If not, you deserve to have your name there. What name would you like to add there?

[dan-42]

@danmar Wow thank you for the merge and also to give me the honors to call my self a "cppcheck author" :-)
I'm OK with my full name "Daniel Friedrich"