[request] Remove Boost::math_c99 dependency

[philippremy]

Currently the project requires Boost::math_c99 to be available. I would propose to change this dependency to just Boost::math.

I don't think the project actually uses the c99 functions anywhere (it only uses Boost::math), as I don't see that the relevant header is included anywhere (which would be #include <boost/math/tr1.hpp>).

I stumbled across this when trying to switch the AliceVision embedded dependency building process from b2 to CMake for Boost. As it turns out, the CMake build system neither can build the math_c99 module nor informs the user that it is missing from the installation 😅.

I am currently waiting for an answer regarding this (boostorg/math#1329). Nevertheless, it currently makes CCTag unusable for people who built their Boost with CMake.

[simogasp]

It comes from #168, something to do with vcpkg.
If we can safely remove it I'm all for it

[philippremy]

At least on Linux and macOS this worked fine (no dependency on mth_c99). I do not have a Windows machine available right now, so I can't test that. Will try to set-up a VM and see what can be done for vcpkg, if it is an issue after all.

EDIT:

I quickly scanned through the header related to sign.hpp, and I honestly don't see where it calls out to any non-header only function. So theoretically that should not cause a linker issue... but as a long-term Unix user, I don't know what MSVC might do in terms of voodoo-magic here...

[philippremy]

I finally got to spin up the Windows VM. I tested it on with the latest Visual Studio SDK and removed the legacy module with an overlay port (and in turn all links to Boost::math_c99 in the CMakeLists.txts).

I had no issues whatsoever - everything compiled fine and I got no linker issues.

And the maintainer of Boost::math actually proposed to remove this patch from vcpkg completely - so whatever the issue was back then, I assume it got fixed along the way... the maintainer does not seem to have interest in adding the legacy libs into the CMake build system.

So from my side we can go forward with this :).

[simogasp]

@philippremy you can try #232
I suspect that it has something to do with the way vcpkg packages each component, which is different than the official boost one or the old findBoost.
I don't have a Windows machine either, so I cannot test it to see if it works now.

[philippremy]

Will try :)

[philippremy]

Sorry that it took so long, but I was (and technically still am) busy with exams.

But: Good news. I tested the referenced branch on a clean Windows 11 VM (with Visual Studio 2026 Community) and I think you are pretty close to get this working. I installed the dependencies with vcpkg as instructed on the homepage.I did not test CUDA though, but I don't see how that would impact the issue at hand. And just for reference, I used the ARM64 version of Windows 11 (because emulating x86_64 takes ages on my arm64 machine).

I ran into two issues:

• The vcpkg install was missing the Boost.Foreach module (which are only two headers, I just added them manually and then it worked fine). We just need to tell vcpkg and cmake to also search for the boost-foreach module and update the docs.
• The project currently cannot handle Windows ARM64, because in src/cctag/pcg_uint128.hpp there is an endian detection mechanism, which (at least for me) failed to detect the correct setting with MSVC on ARM64. That would probably just be another #elif case to add (and then probably default to little-endian).

What I did not reproduce (and which confirms our assumption) is that we need the Boost.Math legacy modules. I didn't have any include or linker issues; so if this problem persisted at some point, it probably resolved itself :D. So I think it is safe to remove the math_c99 component.

[simogasp]

Yes we still have the dependency of FOREACH here

CCTag/src/cctag/Identification.cpp

Line 170 in 0801af5

BOOST_REVERSE_FOREACH( const MapT::const_iterator::value_type & v, sortedId )

Probably the best solution is to switch and add cmake presets so we have the complete list of needed dependencies.
We actually already have it in the vcpkg port
https://github.com/microsoft/vcpkg/blob/master/ports/cctag/vcpkg.json

it's just a matter of adding the boilerplate thing to build with presets.

[philippremy]

Cool, sounds good!

Regarding FOREACH though, I think we could even remove this one occurrence, remove another dependency and replace it with a regular C++ std::const_reverse_iterator and a for loop in this particular case, since we only iterate a std::map<float, MarkerID (aka int)>, if I see this correctly.

If that truly is the only occurrence, we could just use the C++ STL, no?

And would it make sense to use std::vector::emplace_back() in the loop? That could save us a copy of the newly constructed std::pair, but this could already be optimised away by the compiler anyways.