Fix #11553 pop_back on empty container is UB

[chrchr-github]

No description provided.

[pfultz2]

We should be able to warn when the value is not known as well.

[chrchr-github]

The problem is code like

void f(std::vector<int>& v) {
	v.pop_back();
	if (v.empty()) {}
}

We don't seem to take into account that the size changes with pop_back().

[pfultz2]

That looks like a bug in Valueflow. Maybe we don't account for the POP action.

[chrchr-github]

It's not necessarily a ValueFlow issue.
Propagating possible values to locations preceding a size-changing action seems wrong.

We don't warn for this either, although v.resize() does get a possible empty value:

void g(std::vector<int>& v) {
	int n = v.size();
	v.resize(n - 1);
	if (v.empty()) {}
}

containerYieldsElement() is obviously false, and containerAppendsElement() does not check whether resize() adds or removes elements.

[chrchr-github]

We also don't warn for

void g(std::vector<int>& v) {
	v.push_back(1);
	if (v.empty()) {}
}

[pfultz2]

So I fixed the valueflow issues in #4800 and #4803 so you should be able to warn for possible values as well.

For the v.resize(n - 1) case, the checker would need to be updated, but I dont think it needs done in this PR though.

[chrchr-github]

Nice. So we should be able to warn for

void f(std::vector<int>& v) {
	if (v.empty()) {}
	v.pop_back();
}

after those PRs have been merged.
The resize(n - 1) example is equivalent to the pop_back() one in the 2nd comment above, so we shouldn't warn for those.

[pfultz2]

I think #4800 is only needed to support possible values as #4803 needs some more work to fix FPs.