R package fails to install from checkout on Windows due to jemalloc

[jeroen]

What happens?

duckdb fails to install on Windows in r-universe due to jemalloc errors. The problem appears because we create the R source package on Linux, and then install/check this source package on Linux, Mac, and Windows (just like CRAN).

The bug is that when the source package is made on linux, rconfgure.py unconditionally enables the jemalloc extension :

duckdb/tools/rpkg/rconfigure.py

Lines 9 to 10 in b8cf6a9

	if platform.system() == 'Linux' and platform.architecture()[0] == '64bit':
	extensions.append('jemalloc')

This results in a duckdb R package that cannot be installed on Windows. But in addition it also behaves different from the CRAN package on other platforms: the CRAN duckdb configuration does not use the jemalloc extension (regardless of platform) because the R source package was created on MacOS.

Solution

The easiest solution is to remove the two lines above to make this extension opt-in like all other extensions. This way the R source package matches the configuration of the CRAN package, even when you create it on Linux. See #6864

If you would want to auto-enable jemalloc on linux, you would need to check the platform at install time, either in your src/Makevars file or a configure script that you ship with the R package. Your current configure script only runs at source build time but is omitted from the source package that you publish on e.g. CRAN:

duckdb/tools/rpkg/.Rbuildignore

Lines 1 to 6 in b8cf6a9

	.*\.sh
	.*\.tar.gz
	.*\.tgz
	.*\.zip
	git
	configure

This is a bit confusing, you probably should name it something else than configure if it is not an install-time script, but that is another issue :)

To Reproduce

Build the duckdb R source package on Linux, and try to install it on Windows. (the converse also doesn't work btw).

Or: look at the build log on https://duckdb.r-universe.dev/duckdb

OS:

Windows

DuckDB Version:

git master

DuckDB Client:

R

Full Name:

Jeroen

Affiliation:

ropensci

Have you tried this on the latest master branch?

• I agree

Have you tried the steps to reproduce? Do they include all relevant data and configuration? Does the issue you report still appear there?

• I agree