Build r-tiledbsoma with C++ 20

[johnkerl]

Context: single-cell-data/TileDB-SOMA#3154

[jdblischak]

For some reason the osx-64 build isn't using C++20. The errors are the same as in the nightly build (#230 (comment))

[jdblischak]

@johnkerl could you please rebase your PR onto the latest from main to include #228?

[jdblischak]

For some reason the osx-64 build isn't using C++20

Well I guess this makes sense. The osx-64 build is the most complicated since it is the only one that uses the wrapper functions:

tiledbsoma-feedstock/recipe/build-r-tiledbsoma.sh

Lines 9 to 19 in fd059f0

	# https://github.com/conda-forge/r-tiledb-feedstock/commit/29cb6816636e7b5b58545e1407a8f0c29ff9dc39
	if [[ $target_platform == osx-64 ]]; then
	export NN_CXX_ORIG=$CXX
	export NN_CC_ORIG=$CC
	export CXX=$RECIPE_DIR/cxx_wrap.sh
	export CC=$RECIPE_DIR/cc_wrap.sh
	mkdir -p ~/.R
	echo CC=$RECIPE_DIR/cc_wrap.sh > ~/.R/Makevars
	echo CXX=$RECIPE_DIR/cxx_wrap.sh >> ~/.R/Makevars
	echo CXX17=$RECIPE_DIR/cxx_wrap.sh >> ~/.R/Makevars
	fi

These wrapper scripts were initially added to handle C++17 requirements (conda-forge/r-tiledb-feedstock@29cb681, conda-forge/r-tiledb-feedstock#27). Maybe C++20 has different requirements?

[jdblischak]

Ah, looking earlier in the build log, I see we have a nice diagnostic signal. R explicitly tells us it is using C++17 on osx-64.

* installing *source* package ‘tiledbsoma’ ...
** using staged installation
** updated src/Makevars for system library via pkg-config
** libs
using C++ compiler: ‘clang version 18.1.8’
using C++17
using SDK: ‘NA’‘NA’‘NA’‘’

Bizarrely though, the osx-arm64 build also reports using C++17, so unclear why it is able to compile successfully.

[johnkerl]

Commit 3:

mamba upgrade -c conda-forge conda-smithy -n base
conda smithy rerender --commit auto

[jdblischak]

I think it is going to be tough to build r-tiledbsoma with C++20 without the upstream changes in single-cell-data/TileDB-SOMA#3331 that updated configure and src/Makevars.in.

If we want to fix the nightly build, I think we'll need to 1) fix the build on the branch nightly-build, and then 2) modify nightly/update-recipe.sh to incorporate these changes.

[johnkerl]

@jdblischak I was confused

single-cell-data/TileDB-SOMA#3331 is merged to main of TileDB-SOMA & I was confusing this PR with the centralized-nightlies which should be testing the day's main.

Let me take a look at nightly-build and nightly/update-recipe.sh which, embarrassingly, I'm not at all familiar with ...

[jdblischak]

I was confusing this PR with the centralized-nightlies which should be testing the day's main.

Good news! The TileDB-SOMA-R build in centralized-tiledb-nightlies is using C++20 and building without error. Thus if we can get the nightly feedstock build to also use C++20, it should just work.

[johnkerl]

Awesome, thanks @jdblischak ! I'm still working on what I need to type to get us there.