HeuristicResolver cannot resolve chained dependent field access

[HighCommander4]

In the following code:

int next;

template <typename T>
struct Link {
  Link<T> *next;
};

template <typename T>
void foo(Link<T> *link) {
  link->next->next;
}

HeuristicResolver is able to resolve the first next in link->next->next, but not the second.

This can be seen in clangd: go-to-definition on the second next will also offer the unrelated global variable as a result, indicating that it's falling back to the textual heuristic because semantic resolution failed.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Nathan Ridge (HighCommander4)

In the following code:

int next;

template <typename T>
struct Link {
  Link<T> *next;
};

template <typename T>
void foo(Link<T> *link) {
  link->next->next;
}

HeuristicResolver is able to resolve the first next in link->next->next, but not the second.

This can be seen in clangd: go-to-definition on the second next will also offer the unrelated global variable as a result, indicating that it's falling back to the textual heuristic because semantic resolution failed.

[HighCommander4]

Slightly simplified testcase that does not involve pointer access (which introduces additional complications):

struct A {
  int waldo;
};

template <typename T>
struct B {
  A &a;
};

template <typename T>
void foo(B<T> &b) {
  b.a.waldo;  // waldo is not resolved by HeuristicResolver
}

[HighCommander4]

Proposed fix for the simplified testcase: #124451

The original testcase remains to be fixed.

[HighCommander4]

Proposed fix for the simplified testcase: #124451

The original testcase remains to be fixed.

The revised version of the patch at #124451 fixes both the original and simplified testcases.