boostorg/uuid@0f84313 - broke boost::compute

[nega0]

$ brew info boost
==> boost: stable 1.86.0 (bottled), HEAD
Collection of portable C++ source libraries
https://www.boost.org/
Installed
/opt/homebrew/Cellar/boost/1.86.0 (16,215 files, 518.0MB) *
  Poured from bottle using the formulae.brew.sh API on 2024-08-21 at 15:08:42
From: https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/b/boost.rb
License: BSL-1.0
==> Dependencies
Required: icu4c ✔, xz ✔, zstd ✔
==> Options
--HEAD
	Install HEAD version

$ command cat main.cc
#include <boost/compute.hpp>

int main(){}

$ gcc -o main -I/opt/homebrew/include -std=c++11 -Wno-deprecated main.cc
In file included from main.cc:1:
In file included from /opt/homebrew/include/boost/compute.hpp:14:
In file included from /opt/homebrew/include/boost/compute/algorithm.hpp:18:
In file included from /opt/homebrew/include/boost/compute/algorithm/accumulate.hpp:20:
In file included from /opt/homebrew/include/boost/compute/algorithm/reduce.hpp:20:
In file included from /opt/homebrew/include/boost/compute/detail/meta_kernel.hpp:39:
/opt/homebrew/include/boost/compute/detail/sha1.hpp:41:26: error: non-const lvalue reference to type 'digest_type' (aka 'unsigned char[20]') cannot bind to a value of unrelated type 'unsigned int[5]'
            h.get_digest(digest);
                         ^~~~~~
/opt/homebrew/include/boost/uuid/detail/sha1.hpp:179:43: note: passing argument to parameter 'digest' here
inline void sha1::get_digest(digest_type& digest)
                                          ^
1 error generated.

$

There are two issues here.

1. 0f84313 broke boostorg/compute
2. This breaking change made it through the boost release process.

@pdimov given that 0f84313 is your change, and you were also the last to commit to boost::compute @ 36350b7 maybe you can get us a 1.86.1 sooner rather than later.

[nega0]

See boostorg/compute#889
See boostorg/compute#888
See boostorg/compute#887

[pdimov]

My mistake; I didn't notice that Boost.Compute was using an implementation detail of UUID.

[nega0]

Thank you for following up on the Boost ML, it seems my email there is "stuck". Probably for the best, anyway. I don't think anyone would expect you or any boost maintainer to notice an integration issue like this on every commit or merge. As a Boost consumer I do expect an issue like this to be caught in the lead up to release. Hopefully this issue can lead to a broader discussion of the "release manager's" role and responsibilities.

[jeking3]

Boost.Compute was using an implementation detail of UUID

This underscores the need for more automation to find and remove invalid use of implementation details across packages.

[pdimov]

Compute's tests require OpenCL and are automatically disabled when it's not available, which is why all the automation in the world missed it.

[jeking3]

Agree that not having CI on Boost.Compute is also a problem. However, -if- we had a code scanner to detect this inappropriate use of implementation details, and folks did not violate the rules, it never would have happened.