Andrey Semashev wrote:
On that note, what's the use of the tag argument? I have only seen hash_append_tag so far, and I don't think adding new tags in the future would be any different from using separate functions (i.e. to support a new tag would require about the same amount of user's code modifications as adding a new friend function to the user's classes).
This is the tag_invoke pattern: https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2019/p1895r0.pdf Instead of every library making up its own identifier for ADL purposes (e.g. in Hash2 I used do_hash_append in the past), we use tag_invoke with a library-specific tag. The advantage is that the tag is namespaced. If two libraries pick the same name for their ADL customization point, they will conflict; but if two libraries pick the same tag name, they will not conflict because the tags are in their respective namespaces. We use tag_invoke in Boost.JSON as well.