Add FindBoost_EP.cmake

[davisp]

Pull in Boost as an external project when building without vcpkg.

---

TYPE: NO_HISTORY
DESC: Add FindBoost_EP.cmake

[teo-tsirpanis]

Suggested change

set(Boost_NO_SYSTEM_PATHS ON)

See #4723 (comment)

[teo-tsirpanis]

How about we use boost-cmake instead of b2? 🤔

[teo-tsirpanis]

The boost repository has a CMakeLists.txt but the source artifacts don't. Why? 😭

[teo-tsirpanis]

Recursively cloning boost and all its submodules takes an extraordinary amount of time…

[teo-tsirpanis]

Given that

• Adding a cross-platform ExternalProject for Boost is going to be a nontrivial amount of work
  • The ExternalProjects are going to be removed soon either way
• Conda and MariaDB are the only remaining first-party users of TILEDB_VCPKG=OFF
• MariaDB can just enable vcpkg with Enable Azure and vcpkg. TileDB-MariaDB#319
• Conda can not yet enable vcpkg but can just acquire vcpkg from conda-forge with Add dependeny to Boost. conda-forge/tiledb-feedstock#234

How about we

• Don't try to write Boost_EP
• Require the user to have Boost installed in the system if building without vcpkg
• Manually install Boost in nightly builds on Unix and use vcpkg on Windows

What do you think @KiterLuc, @ihnorton?

[eddelbuettel]

Why not follow KISS and vendor the Boost parts we need as proposed by @eric-hughes-tiledb and (implicitly) endorsed by @lums658 ?

[davisp]

Why not follow KISS and vendor the Boost parts we need as proposed by @eric-hughes-tiledb and (implicitly) endorsed by @lums658 ?

Because the bits we need aren't header only so just vendoring it means that we'll need to reverse engineer whatever sort of build system which is where things start to get a bit weird.

[eddelbuettel]

Because the bits we need aren't header only

Dang. Consider my proposal withdrawn.

[davisp]

@teo-tsirpanis I just used what the docs and tutorials were using. I didn't see boost-cmake. In terms of cross platform, it appears to work just fine on both macOS and Linux. Is there something non-trivial preventing the branch from working on Windows? I've got no idea what's breaking in that build in CI so I can't tell if its as simple as invoking the build command incorrectly because I don't know how Windows CLIs work, or if there's some deeper issue there. But if you tell me that it's untenable to fix then I'm fine going with another approach.

I'm not against the idea of relying on a system installed version of boost, but I'd need to @ihnorton or someone else a lot more familiar with packaging and distribution to know if that's a viable approach.

[davisp]

Recursively cloning boost and all its submodules takes an extraordinary amount of time…

Why are you cloning things? This PR just downloads the release tarball.

[teo-tsirpanis]

The tarball doesn't have CMakeLists.txt for some reason.