alice-vision: init at 3.0.0, add 3.1.0

[zeuner]

Description of changes

Opening this since #224262 appears to be stale. @hesiod left off at a very promising state where a build was already configured to work without pulling in submodules. I moved forward up to the point where the package has working (usable for doing actual 3D meshing) binaries and added a package for the more experimental release upstream made in the meantime.

I would consider the packages as a valuable addition since they would move us a significant step forward in achieving #216403

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 23.11 Release Notes (or backporting 23.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

[zeuner]

@AndersonTorres I had to touch your nanoflann package since alice-vision requires an older version to compile. You might want to review whether you are fine with the changes.

[AndersonTorres]

Is the :BOOL needed?

[zeuner]

With the :BOOL omitted, some of the CMake cache entries are created as strings instead of booleans. Not sure whether it can make a practical difference, though. The build configuration doesn't change here.

[AndersonTorres]

This is because I am writing a helper function for CMake:

#256418

[zeuner]

For a generic helper function, I would always mandate for setting more exact typing data if it's possible. This way, maintainers could possibly get higher quality metadata for CMake without extra effort, while getting to that point might beconsidered as too tedious at a single package scope.

[AndersonTorres]

I have found only one place where those "data types" are defined:

https://cmake.org/cmake/help/latest/command/set.html

It looks too gui-driven, and also too dependent on the upstream - what if this variables are truly strings that happen to act like booleans?

Nonetheless, it should be easy to adjust.

[AndersonTorres]

Too off-topic. Solving this.

[zeuner]

Any blockers left?

[AndersonTorres]

Not on my part.

[SomeoneSerge]

Ez, lib.cmakeBool name value

EDIT: Oh, I see, they rolled their own custom stuff that doesn't allow TRUE/FALSE...

[zeuner]

Investigating the failing tests it turned out they happened due to a binary incompatibility with cctag. They are now fixed by disabling libEigen alignments when cctag support is enabled. In order to always use alignment, #258950 would have to be fixed.

At this point, from what I perceive this package is on-par with the monolithic build (8411621).

[infinisil]

The pkgs/by-name check failure can be safely ignored, it's a problem with the check that's to be fixed with #285089

[infinisil]

The above PR is now merged and used in CI, so it won't fail anymore the next time you push :)

[infinisil]

Although I don't think it's great how the expression is effectively duplicated between the two versions. See #285922 (comment) for my recommendation for multiple package versions.

[zeuner]

The above PR is now merged and used in CI, so it won't fail anymore the next time you push :)

Thanks for your work! I need to retrigger CI anyway due to #284145

[zeuner]

Although I don't think it's great how the expression is effectively duplicated between the two versions.

Fully agree. When I started where #224262 left off, I wanted to keep the changes to 3.0.0 minimal considering that this was the code @hesiod initially agreed to maintain. A few months having passed without a reaction, I'm not even sure whether this is still the case. So I'll refactor as I think it will fit.

I still find it useful to have multiple versions since there seem to be 3D reconstruction pipelines that perform worse on newer versions (see alicevision/AliceVision#1501).

[pbsds]

Suggested change

	src = old.src // {
	src = old.src.override {

[pbsds]

consider marking broken on darwin

[pbsds]

It also seems the nanoflann override needs more work, as it fails to build when merged with master

[bouk]

FYI there's a fix for the darwin failure in staging: #287255

So I think it's fine to merge with the broken Darwin build

[pbsds]

#256115 (comment) is still blocking

nix-repl> alice-vision_3_0_0.src    
«derivation /nix/store/icpw6kvspi5ws2rksv2dswkwwwgvngbz-source.drv»

nix-repl> alice-vision.src       
«derivation /nix/store/icpw6kvspi5ws2rksv2dswkwwwgvngbz-source.drv»

[SomeoneSerge]

In addition to the nanoflann failure, I'm seeing this:

alice-vision> /nix/store/bgcaxhhxswzvmxjbbgvvaximm5hwghz1-binutils-2.41/bin/ld: warning: libboost_thread.so.1.81.0, needed by /nix/store/2viiyhdypfw0v0ygxk0vvgyzyq0fad76-openimageio-2.5.5.0/lib/libOpenImageIO.so.2
.5.5, may conflict with libboost_thread.so.1.79.0

[luzpaz]

soft bump