[whispersofthedawn]

i think we did this update right, and diffs are in the commit messages.

original description detailing an outstanding problem that mszabo-wikia gave us a solution for below

we're not done figuring this out and mostly opening this to ask for help (sorry).

current status: everything builds fine on x86_64-linux and seems to work. on aarch64-linux specifically, building fbthrift currently fails with

[239/622] Building CXX object thrift/lib/cpp2/CMakeFiles/thriftmetadata.dir/dynamic/TypeSystemTraits.cpp.o
FAILED: [code=1] thrift/lib/cpp2/CMakeFiles/thriftmetadata.dir/dynamic/TypeSystemTraits.cpp.o
/nix/store/hy2xqdbh7mw1w1yhqggfb3s1dmf89h10-gcc-wrapper-15.3.0/bin/g++ -DBOOST_ATOMIC_DYN_LINK -DBOOST_ATOMIC_NO_LIB -DBOOST_CONTEXT_DYN_LINK -DBOOST_CONTEXT_NO_LIB -DBOOST_FILESYSTEM_DYN_LINK -DBOOST_FILESYSTEM_NO_LIB -DBOOST_PROGRAM_OPTIONS_DYN_LINK -DBOOST_PROGRAM_OPTIONS_NO_LIB -DBOOST_REGEX_DYN_LINK -DBOOST_REGEX_NO_LIB -DFMT_SHARED -DGFLAGS_IS_A_DLL=0 -DGLOG_USE_GLOG_EXPORT -DTHRIFT_HAVE_LIBSNAPPY=0 -Dthriftmetadata_EXPORTS -I/build/source/. -I/build/source/build -isystem /nix/store/dg6pgq3k8j2rj4j004h3rqks3kd3bay5-fast-float-8.2.10/include -isystem /nix/store/rh9samgza9bhdsz27g2n6afag5ch2vdg-libevent-2.1.13-dev/include -isystem /nix/store/gc8kg459z9kjzx4yg96q7vkvm4wlm4b8-xz-5.8.3-dev/include -isystem /nix/store/5ky8ppq46nrzqx5fkn6ickajvzllss3y-lz4-1.10.0-dev/include -isystem /nix/store/symmwwcz269glcxpzmlss970immizkz6-libiberty-15.3.0-dev/include/libiberty -isystem /nix/store/m4qhnwkm1n2l7hkjzr9p9i1z5f7nb6a5-libunwind-1.8.3-dev/include -O3 -DNDEBUG -std=c++20 -fPIC -fcoroutines -MD -MT thrift/lib/cpp2/CMakeFiles/thriftmetadata.dir/dynamic/TypeSystemTraits.cpp.o -MF thrift/lib/cpp2/CMakeFiles/thriftmetadata.dir/dynamic/TypeSystemTraits.cpp.o.d -o thrift/lib/cpp2/CMakeFiles/thriftmetadata.dir/dynamic/TypeSystemTraits.cpp.o -c /build/source/thrift/lib/cpp2/dynamic/TypeSystemTraits.cpp
In file included from /build/source/./thrift/lib/cpp2/type/Any.h:26,
                 from /build/source/thrift/lib/cpp2/dynamic/TypeSystemTraits.cpp:18:
/build/source/build/thrift/lib/thrift/gen-cpp2/any_rep_types.h:12:10: fatal error: thrift/lib/thrift/gen-cpp2/type_types.h: No such file or directory
   12 | #include "thrift/lib/thrift/gen-cpp2/type_types.h"
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.

we are struggling to trace what could possibly be platform specific here and causing this to fail, and putting it here in the hopes that someone else has marginally more of a clue. (our best guess right now is that it might be a situation with an undeclared dependency that happens to be generated in the right order on our x86_64-linux system with more parallelism than the small raspberry pi building for aarch64-linux.)

tested with some basic usage of watchman on the command line, subscribing to, querying events for, and triggering commands on changes to a directory. not sure how to meaningfully test the other packages here beyond their own test suites.

also: we offer our sincerest condolences to the folks with the unfortunate responsibility of maintaining these packages. it seems like hell, especially on darwin. (though this comment from emily did give us a good laugh :3)

Things done

• Built on platform:
  • x86_64-linux
  • aarch64-linux
  • aarch64-darwin
• Tested, as applicable:
  • NixOS tests in nixos/tests.
  • Package tests at passthru.tests.
  • Tests in lib/tests or pkgs/test for functions and "core" functionality.
• Ran nixpkgs-review on this PR. See nixpkgs-review usage.
• Tested basic functionality of all binary files, usually in ./result/bin/.
• Nixpkgs Release Notes
  • Package update: when the change is major or breaking.
• NixOS Release Notes
  • Module addition: when adding a new NixOS module.
  • Module update: when the change is significant.
• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.
• Follows the automation/AI policy.

[mszabo-wikia]

@whispersofthedawn if you figured this out in the meantime feel free to ignore this, but otherwise I suspect adding type-cpp2-target as an explicit dependency of thriftmetadata or maybe even any_rep-cpp2-target may help. FWIW I haven't been able to reproduce the original issue on my local so probs this is heavily dependent on # of jobs or possibly the ninja version

[whispersofthedawn]

thank you so much for the pointer! that seems to have fixed it for us :) pushed a patch to fbthrift to include that. it seems like everything is working for us now across the stack.

[mszabo-wikia]

@whispersofthedawn glad to hear it worked! I sent an upstream PR with the fix. I took the liberty of committing it under my own ID since there's a CLA gate. If you'd rather submit your own PR or would like the authorship changed, let me know and I can close/amend as needed.

[whispersofthedawn]

@mszabo-wikia thanks! i've added a link to the pr in the comment by the patch. and i'd rather not deal with the facebook cla myself so i appreciate you submitting it <3

[vcunat]

Cross-ref: #550199 (comment)

[mszabo-wikia]

Looks like meta/folly@ac1bb81 is included in this bump and should fix the folly include issue.

[vcunat]

I confirm that this PR does fix the folly failure (when merged into nixkpgs master).

[whispersofthedawn]

thanks for checking! i think for this we mainly need to verify that darwin does not completely fall apart if we take this update, since it's caused problems in the past (well, and maintainer review). though i suppose if it's already broken it can't get more broken.

[vcunat]

On an aarch64-darwin (with nixpkgs master + this PR) I tried to build folly fizz mvfst wangle fbthrift fb303 edencommon watchman and only watchman failed.

[vcunat]

Unfortunately, x86_64-linux goes even worse and breaks on edencommon already (not getting to watchman). The darwin error is from fmt and this linux error also looks like consequence of something in PR #550199

[vcunat]

Still, some packages do get fixed by this PR, so 🤷🏽

[whispersofthedawn]

i'll take a look at at least the -linux failures this evening, though no -darwin for me. though if it's fmt related it'll hopefully be cross-platform.

[vcunat]

That fmt deprecation is cross-platform, hitting many packages.

[whispersofthedawn]

the whole stack builds on x86_64-linux now. had to sed in a missing <string> header (that was transitively included from fmt/core.h before) for edencommon, and vendor an open pr for fmt/core.h → fmt/format.h for watchman, but everything should be fine now.