Peter Dimov

I agree that it should win, if we can assume that everything is written correctly, but until I/we figure out a way to make these mistakes less likely to occur,...

"Otherwise" implies that they are tried in order, and the first match wins, which isn't currently the case (for the same reason I haven't made `tag_invoke` "win" over the rest.)...

This looks OK at a first glance, but you should change ``` if (BOOST_SUPERPROJECT_VERSION) ``` to ``` if(NOT CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR) ``` which checks whether this is the root project...

Also, if you're going to make changes to CMakeLists.txt, you should first add CI jobs that verify that current uses aren't broken. Or I can add them for you if...

CI jobs for testing CMake subdir/install use added. I arbitrarily picked the `adj_list_edge_list_set.cpp` test as the `main.cpp` file in both cases.

Here: https://github.com/boostorg/graph/commit/f1e51a08578a648c27ef61c0581adb9e8440cac4

Interesting, thanks. I knew about rurban/smhasher (and used it to test https://github.com/pdimov/mulxp_hash), but wasn't aware of your fork.

As explained in the Boost.JSON documentation: https://www.boost.org/doc/libs/1_88_0/libs/json/doc/html/json/ref/boost__json__is_described_class.html structs with base classes don't get an implementation by default. To enable it, you have to specialize `boost::json::is_described_class` like this: ``` template struct...

Yes, I think that we should just make it work.

If a member shadows another member, and you're using mod_inherited, the shadowed member is not returned by default (because it's mod_hidden.) If a member is ambiguous and not hidden, this...