Watchman fails to build with fmt 8.1.0+

[carlocab]

Hi there, we're trying to update Homebrew's version of Watchman at Homebrew/homebrew-core#92869. We were last able to build Watchman against fmt 8.0.1, and it now no longer builds since Homebrew's fmt was updated to 8.1.0. The failure occurs on multiple versions of macOS and on Linux.

The Watchman build fails with this error:

  In file included from /tmp/watchman-20220110-77236-142y8ko/watchman-2022.01.10.00/watchman/PendingCollection.cpp:8:
  In file included from /tmp/watchman-20220110-77236-142y8ko/watchman-2022.01.10.00/watchman/PendingCollection.h:11:
  In file included from /opt/homebrew/include/folly/futures/Promise.h:23:
  In file included from /opt/homebrew/include/folly/futures/detail/Core.h:27:
  In file included from /opt/homebrew/include/folly/Executor.h:25:
  In file included from /opt/homebrew/include/folly/Range.h:45:
  In file included from /opt/homebrew/include/fmt/format.h:48:
  /opt/homebrew/include/fmt/core.h:1722:3: error: static_assert failed due to requirement 'formattable_pointer' "Formatting of non-void pointers is disallowed."
    static_assert(formattable_pointer,
    ^             ~~~~~~~~~~~~~~~~~~~

Complete build logs available at https://github.com/Homebrew/homebrew-core/actions/runs/1680135847.

I tried to follow the call stack (see the log after the snippet I pasted above), and it doesn't look to me that Watchman is passing any pointers at all to fmt. However, I don't really do C++, and I'm unfamiliar with Watchman's codebase, so I could be mistaken.

Any assistance trying to resolve this would be appreciated. Thank you!

[vitaut]

I'll look into it, thanks for reporting.

[vitaut]

The problem is that w_string is implicitly convertible to a pointer which is caught by a new diagnostic (compile-time error) introduced in {fmt} 8.1:

Added missing diagnostic when trying to format function and member pointers as well as objects convertible to pointers which is explicitly disallowed

Fixed in 8f8a1a0 by only giving the diagnostic if there is no formatter specialization.

[carlocab]

Thanks for the lightning-quick fix. I've confirmed that it works.

[gh-andre]

@vitaut I just ran into the same bug. When this is planned to be released? If it's going to be some time, would you mind pointing out which commits can be used as a patch for 8.1.0 or 8.1.1 in the meantime? Thanks.

[carlocab]

8f8a1a0 (mentioned above) applies cleanly to at least 8.1.1.

[gh-andre]

@carlocab Thanks. There's a lot of commits around I was hoping to hear from the maintainers what would work well for a patch until the next release. It would be nice if an 8.1 branch was maintained for these patch fixes.

[vitaut]

8f8a1a0 is pretty self-contained. It will also be included in the next release but I don't have any specific dates yet.

[gh-andre]

@vitaut Sounds good. Thanks for confirming.