scoped_allocator with unordered_map on macos

[boostorg]

scoped_allocator with unordered_map on macos

[cmazakas]

@kleunen Can I go ahead and close this issue or is there anything else you'd like to add/ask about?

[kleunen]

I think I figured it out myself now, it seems unordered_map does not detect by itself it can use allocator traits and cxx11 construction with libc++. I had to add these defines:

#define BOOST_UNORDERED_USE_ALLOCATOR_TRAITS 1
#define BOOST_UNORDERED_CXX11_CONSTRUCTION 1
#include <boost/unordered_map.hpp>

I think it would be good to add detection of llvm libc++ to unordered_map, now that it is used more often

[cmazakas]

Hi kleuen!

Unfortunately, I tried recreating your issue but everything compiles:
https://godbolt.org/z/hePrb5YGq

It's not a perfect recreation of your exact setup but should be similar enough so as to trigger an error.

Could I possibly get some more information about your setup?

My question is: what compiler are you using? I'm guessing you're using Apple Clang which unfortunately I can't easily reproduce.

I'm also wondering, what C++ version? If I have more information around your compiler version and the C++ standard, I might have better luck recreating the issue.

[kleunen]

The issue occured on the Github CI setup, so the virtual environment of Github CI with macos was used.

I am not sure what version that was, possibly this is already been updated.

[cmazakas]

Would you be able to provide a smaller, minimal example that causes the failure you experienced?

[kleunen]

I have to think about that. Actually, i am not using the scoped_allocator anymore. It was easier to use a stateless allocator, so I made a workaround with a global allocator.

The example you showed was similar to the code I was using.

[cmazakas]

Ha ha, no worries.

You made me realize, Unordered actually doesn't have a test case like what you mentioned. So I'm going to work on adding it to the CI test matrix and then hopefully, we'll be able to trigger a failure.

[kleunen]

If i have time, i will have a look if i can break it. I have some old test code lying around.

[kleunen]

This was a code sample i made:
https://godbolt.org/z/9snoYPM8q

On godbolt it seems to compile fine, so maybe really related to macos ?

[cmazakas]

Yeah, I'm guessing it's probably some weirdness surrounding Apple Clang specifically. Maybe just not a robust enough check on the side of Unordered.

Thank you for the godbolt link!

[kleunen]

My guess was that libc++ on macos was missing certain defines.

[cmazakas]

Hi @kleunen, I have a PR up: #31

Would you mind taking a look and seeing if this mirrors your use-case sufficiently simplified?

The Github Actions test matrix does include OS X and unfortunately, a compiler error doesn't seem to have been triggered.

[kleunen]

I will have a look.

Yes, I think the macos virtual environment is probably upgraded now.
Also ubuntu 16.04 was removed recently, so.

[kleunen]

The PR looks good. You are checking now the stateful scoped allocator is passed to the underlying vector inside the map. So, at least you have a test for this now, to make sure the C++ environment is handling this correctly.

[cmazakas]

Awesome, then I'll work on fixing the msvc compilation failures (so hey, at least we found one failure in the test matrix) and then we'll get this bad boy merged!