[R][CI]: Compile error on opensuse155

[jonkeane]

Describe the bug, including details regarding any error messages, version, and platform.

Since #45562 (and the error also suggests this PR), we have seen the test-r-rstudio-r-base-4.1-opensuse155 crossbow job fail with:

[ 59%] Building CXX object src/arrow/CMakeFiles/arrow_compute.dir/compute/kernels/aggregate_pivot.cc.o
/arrow/cpp/src/arrow/compute/kernels/aggregate_pivot.cc: In function ‘arrow::Result<std::unique_ptr<arrow::compute::KernelState> > arrow::compute::internal::{anonymous}::PivotInit(arrow::compute::KernelContext*, const arrow::compute::KernelInitArgs&)’:
/arrow/cpp/src/arrow/compute/kernels/aggregate_pivot.cc:152:10: error: could not convert ‘state’ from ‘std::unique_ptr<arrow::compute::internal::{anonymous}::PivotImpl, std::default_delete<arrow::compute::internal::{anonymous}::PivotImpl> >’ to ‘arrow::Result<std::unique_ptr<arrow::compute::KernelState> >’
   return state;
          ^~~~~
make[2]: *** [src/arrow/CMakeFiles/arrow_compute.dir/build.make:496: src/arrow/CMakeFiles/arrow_compute.dir/compute/kernels/aggregate_pivot.cc.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:2045: src/arrow/CMakeFiles/arrow_compute.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

I suspect this is the old(er) GCC compiler in opensuse155 (7.5) which is EOL so we should probably bump the 4.1 test to a different OS.

Component(s)

Continuous Integration, R

[jonkeane]

This might turn into a larger C++ cleanup too, since there are a bunch of things the seemed to be added just for this https://github.com/jonkeane/arrow/blob/0fbf9823542233c5f32c26534c34cc97ce3f0be2/cpp/src/arrow/buffer.cc#L44

[jonkeane]

And judging by #41487 (comment) it looks like we can remove the sharedpointer + move for all of those (or at least start there and try that)

[jonkeane]

Though interestingly, https://arrow.apache.org/docs/developers/cpp/building.html#system-setup says that gcc 7.1 or up should be sufficient. Curious if you have thoughts about if we should update that recommendation for minimum version of gcc, @pitrou? This opensuse r might be the only place we test lower versions of gcc.

[amoeba]

I think we have enough reason to upgrade our minimum GCC version from 7.1 to 8 considering there's work in flight to increase the minimum C++ standard in Arrow C++ to C++20 (#45445) and it looks like you need gcc 8+ for C++20 features. Some of the cleanup mentioned might be possible to do in #45445.

However, what do you think about just fixing the specific failure in this issue for now while the C++20 work is in progress @jonkeane?

[jonkeane]

Sure, I'm fine to be targeted here and let the C++20 upgrade do cleanup

[pitrou]

For the record, the C++20 PR currently removes the OpenSuse build. Is there a reason to keep it?

[jonkeane]

No reason to keep it other than that it was the one place we were testing old versions of C++ we claimed to support. IF we're bumping the minimum C++ version, and we have coverage for the R version elsewhere in our matrix (possibly with a newer opensuse), there's no reason to keep this particular one around.

[amoeba]

PR up to fix just this specific issue at #45874.