Bug 3254 – [module] Module member visibility depends on declaration order

The following code compiles: --a.d--------- import b; void main() { foo(0); } -------------- --b.d--------- void foo(int x) {} private void foo(float x) {} -------------- > dmd -c -o- a.d > But swap function declarations in b.d, and it breaks: --b.d--------- private void foo(float x) {} void foo(int x) {} -------------- > dmd -c -o- a.d a.d: module a b.foo is private

(DMD 1.046 Windows) If b.d is changed to ---------- private void foo(int x) {} void foo(float x) {} ---------- then it doesn't compile either way round: ---------- bz3254a.d: Error: module bz3254a bz3254b.foo is private bz3254a.d(3): Error: function bz3254b.foo is not accessible from bz3254a ---------- The first of these errors disappears if they're swapped over. But I'm not sure whether this is meant to work or not. Are function calls meant to be matched over all overloads, or only those that are accessible?

From bug 314 > Access protection is defined to happen after lookup and overload resolution. So the correct behavior seems to match foo(int) first, then look for its access attribute. Though I don't understand this scheme too.

(In reply to comment #2) > From bug 314 > > Access protection is defined to happen after lookup and overload resolution. Defined where in the spec? > So the correct behavior seems to match foo(int) first, then look for its access > attribute. Though I don't understand this scheme too. Basically, all overloads of a function name are looked at, regardless of their protection attributes, in order to pick an overload. Then the call is validated on the protection attribute of the one that's been picked. My guess is that the point is to protect against accidental behaviour changes if some code is moved between modules. But maybe it's better to avoid having a private function and a public function with the same name in the same scope. But the current behaviour does seem to be overdoing it. If a module imports two modules each of which defines a symbol, but in one of them it's private, an import conflict is reported. It's been my view for a while that private symbols should not be imported at all. Maybe the best policy is to see whether _all_ overloads are private and, if so, the module will keep the symbol to itself; otherwise, invoke the the current behaviour.

(In reply to comment #3) > My guess is that the point is to protect against accidental behaviour changes > if some code is moved between modules. The code moved, it moved from one module to another. The accidental behavior change still can happen in the module the code moved to. If there was public foo(int), the module used to call it, then private foo(float) appears and the module silently starts to call it.

> > From bug 314 > > > Access protection is defined to happen after lookup and > > > overload resolution. > Defined where in the spec? That was Walter's comment. This exact comment made me experiment with module-level protection attributes and led to this bug report. Well, I think the "overload before protection" rule has its merit for classes. Class methods *hide* methods of the same name in base classes to protect from overload set hijacking. It probably would be confusing if hiding rules worked differently depending on whether your function is in the same module with the class or not. But module members cannot hide anything. Any name collision is an error, and any conflict resolution is explicit. Therefore I think overload resolution should be changed for modules, too. Consider: module a; class A {} void foo(A a) {} module b; class B {} void foo(B b) {} module c; import a, b; private alias a.foo foo; private alias b.foo foo; void bar() { foo(new A); foo(new B); } So far so good. But now: module d; import a, c; void baz() { foo(new A); // error ?!! } a.foo conflicts with c.foo ? But I don't know about c.foo. There is no documentation for c.foo. There is no c.foo. Still, this phantom messes up overloading.

*** Issue 5422 has been marked as a duplicate of this issue. ***

I think that disallowing mixed protections within one overload set could work out very well.

Commit pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/e527108f9153cedab2ca658d053b2ac81610e55d fix Issue 3254 - Module member visibility depends on declaration order

This exact bug has reappeared.

Oh, I misread the original report: it's now in reverse; foo(0.0f) compiles, even though it calls a private function. Should I open a new bug?

*** Issue 21829 has been marked as a duplicate of this issue. ***

*** Issue 23397 has been marked as a duplicate of this issue. ***

@ibuclaw created dlang/dmd pull request #14713 "fix Issue 3254 - [module] Module member visibility depends on declaration order" fixing this issue: - fix Issue 3254 - [module] Module member visibility depends on declaration order https://github.com/dlang/dmd/pull/14713

*** Issue 23991 has been marked as a duplicate of this issue. ***

*** Issue 19080 has been marked as a duplicate of this issue. ***

*** Issue 24388 has been marked as a duplicate of this issue. ***

THIS ISSUE HAS BEEN MOVED TO GITHUB https://github.com/dlang/dmd/issues/18058 DO NOT COMMENT HERE ANYMORE, NOBODY WILL SEE IT, THIS ISSUE HAS BEEN MOVED TO GITHUB