[libcxx] [test] Need to include <memory> for std::unique_ptr.

[STL_MSFT]

Summary

[libcxx] [test] Need to include <memory> for std::unique_ptr.

Fixes MSVC "error C2039: 'unique_ptr': is not a member of 'std'".

[Eugene.Zelenko]

It'll be good idea to run Include What You Use to make sure that all dependencies are explicit.

[EricWF]

Woops. My bad.

[EricWF]

r267958.

[rsmith]

In D19698#416255, @Eugene.Zelenko wrote:

It'll be good idea to run Include What You Use to make sure that all dependencies are explicit.

You can also remove all the export *s from the module map and try running the testsuite with modules enabled for a nasty shock about how common this problem is...

[EricWF]

In D19698#416306, @rsmith wrote:

In D19698#416255, @Eugene.Zelenko wrote:

It'll be good idea to run Include What You Use to make sure that all dependencies are explicit.

You can also remove all the export *s from the module map and try running the testsuite with modules enabled for a nasty shock about how common this problem is...

Most tests pretty much rely on <type_traits> getting dragged in indirectly, same thing with <memory> and the bits of <utility> we actually define in <type_traits>.

I was hoping that maybe "Include What You Use" had a fix-it mode. I'll look into this further tonight.

[Eugene.Zelenko]

I was hoping that maybe "Include What You Use" had a fix-it mode. I'll look into this further tonight.

It has fix_includes.py script, but I encountered quite a lot of false positives, so I prefer to fix problems manually.

[rsmith]

In D19698#416309, @EricWF wrote:

Most tests pretty much rely on <type_traits> getting dragged in indirectly, same thing with <memory> and the bits of <utility> we actually define in <type_traits>.

There are also a few symbols that libc++ defines in the wrong header and then #includes into the right one. Any automated fix is going to mess up on those.

[EricWF]

In D19698#416315, @rsmith wrote:

In D19698#416309, @EricWF wrote:

Most tests pretty much rely on <type_traits> getting dragged in indirectly, same thing with <memory> and the bits of <utility> we actually define in <type_traits>.

There are also a few symbols that libc++ defines in the wrong header and then #includes into the right one. Any automated fix is going to mess up on those.

Ah, I stupidly assumed that the automatic fix would handle "std::" symbols based off of the spec and not the implementation.

test/std/utilities/meta/meta.rel/is_callable.pass.cpp

 
 // is_callable
 
 // Most testing of is_callable is done within the [meta.trans.other] result_of
 // tests.
 
 #include <type_traits>
 #include <functional>
+#include <memory>
 
 #include "test_macros.h"
 
 struct Tag {};
 struct DerFromTag : Tag {};
 
 struct Implicit {
   Implicit(int) {}