The case of the missing 32-bit Ruby package

I have access to a 32-bit Arch server that hosts this website and I use Ruby on it to run jekyll for static site generation. I don’t touch it very often, but recently I started getting this exception:

$ irb
irb(main):001:0> require "date_core"
<internal:/usr/lib/ruby/3.0.0/rubygems/core_ext/kernel_require.rb>:85:in `require': cannot load such file -- date_core (LoadError)
        from <internal:/usr/lib/ruby/3.0.0/rubygems/core_ext/kernel_require.rb>:85:in `require'
        from (irb):1:in `<main>'
        from /usr/lib/ruby/gems/3.0.0/gems/irb-1.4.2/exe/irb:11:in `<top (required)>'
        from /usr/bin/irb:25:in `load'
        from /usr/bin/irb:25:in `<main>'

I reproduced it with just the require, but I first encountered it running bundle exec jekyll build. The error is happening because Ruby is trying to load date_core (which is a compiled-from-C .so file) and can’t find it on any of the load paths. It’s the same error you get if you try to import nonsense like require "aoeuaoeu".

tl;dr This can be worked around by setting Ruby’s $LOAD_PATH to explicitly include the directory

export RUBYLIB=/usr/lib/ruby/3.0.0/x86-linux/

but it’s weird that it’s happening at all.

I eventually tracked the problem down to being in /usr/lib/ruby/3.0.0/x86-linux/ which was not on Ruby’s path so it couldn’t find it to load it.

$ ls /usr/lib/ruby/3.0.0/x86-linux/
/usr/lib/ruby/3.0.0/x86-linux/   # This is in the directory x86-linux
$ ruby -e 'puts $LOAD_PATH'
/usr/lib/ruby/3.0.0/i686-linux  # The directory it's looking in is i686-linux

The path that Ruby does look in also exists and has other installed Ruby packages in it. It’s very strange that 32-bit .so files for Ruby packages are in both i686 and x86-linux and that Ruby only seems to know about one of them.

$ ls /usr/lib/ruby/3.0.0/x86-linux/  cgi
$ ls /usr/lib/ruby/3.0.0/i686-linux/  json  racc      rbconfig.rb      digest  enc  io    rbconfig

For each directory, I found a package that owned files there so I could look at their build scripts and compare them: ruby-date & ruby-digest.

$ yay -Qo /usr/lib/ruby/3.0.0/x86-linux/ /usr/lib/ruby/3.0.0/i686-linux/
/usr/lib/ruby/3.0.0/x86-linux/ is owned by ruby-date 3.2.2-4.0
/usr/lib/ruby/3.0.0/i686-linux/ is owned by ruby-digest 3.1.0-5.8

Arch Linux has officially deprecated 32-bit support, so this is where the research became harder. I think ArchLinux32 use the same packages as ArchLinux unless something needs to be different. I didn’t find anything suspicious in the 32-bit specific packaging scripts so I looked at the upstream PKGBUILDs.

I think both packages use local _platform="$(gem env platform | cut -d':' -f2)" to generate the platform-specific version, but it’s producing different results. Since previously I didn’t get the exception, I looked in the pacman logs to find the version difference - ruby-date (3.2.2-3.6 -> 3.2.2-4.0) which means it was only a change in how it was packaged - but both before and after use gem env platform so it’s probably not that.

I couldn’t find a definite answer, but my suspicion is that the different packages (and package versions) are built on different computers, and on some of them gem env platform produces i686 and on others x86-linux, but I’m not sure why.

I didn’t get any further than this - the server was reinstalled with a 64-bit Arch which didn’t have the same problems. If you run into the same problem or figure out what was going on, please let me know!