Find a solution to binary size

[Edouard-chin]

On linux, the resulting binaries are big. For instance the precompiled json gem is around 7MB big. This includes all the binaries that compose the fat gem. The reason its big is because of debugging symbols.
By stripping them LDDFLAGS="-s" we reduce the size of the gem to around 800KB.

That's what rake-compiler-dock do by default. However, stripping debugging symbols make it painful for gem maintainers to investigate a problem when a crash happens on a user machine and they a bug report.

Is it possible to reduce the size of the binaries without stripping debugging symbols ?

[gjtorikian]

👋 I'm interested in this problem, too. I was actually pointed to your PR into json from a similar issue I posted in psych.

I looked into this size issue because I thought it would be a good first step in understanding how this CI system works. I ran the same build on JSON and got an artifact of ~5MB on x84 for Linux.

...Or did I? I downloaded the artifact and investigated:

❯ pwd
/Users/gjtorikian/Downloads/stage-Linux-X64

❯ ls -al
drwxr-xr-x@   4 gjtorikian  staff   128 Dec  4 11:35 pkg
drwxr-xr-x@   3 gjtorikian  staff    96 Dec  4 11:35 tmp

Right off the bat, tmp can be dropped, which is a good thing because that's where 4MB of the size comes from:

❯ du -hs tmp
4.0M	tmp

Within pkg, the non-architecture specific gem is also in the artifact:

❯ ls -alh
-rw-r--r--@  1 gjtorikian  staff   1.8M Dec  4  2025 json-2.16.0-x86_64-linux.gem
-rw-r--r--@  1 gjtorikian  staff   102K Dec  4  2025 json-2.16.0.gem

It's just 100K, but I don't think it's doing anything?

The fact that tmp was kept in the artifact bothered me—maybe I was misremembering something, it has been a while since I mucked around with native Ruby gems—so I unpacked the x86_64-linux gem itself:

❯ gem unpack json-2.16.0-x86_64-linux.gem

❯ ls -alHR json-2.16.0-x86_64-linux/lib/json/ext/
drwxr-xr-x@ 4 gjtorikian  staff  128 Dec  4 11:51 3.1
drwxr-xr-x@ 4 gjtorikian  staff  128 Dec  4 11:51 3.2
drwxr-xr-x@ 4 gjtorikian  staff  128 Dec  4 11:51 3.3
drwxr-xr-x@ 4 gjtorikian  staff  128 Dec  4 11:51 3.4
drwxr-xr-x@ 3 gjtorikian  staff   96 Dec  4 11:51 generator

json-2.16.0-x86_64-linux/lib/json/ext/3.1:
-rwxr-xr-x@ 1 gjtorikian  staff  292368 Dec  4 11:51 generator.so
-rwxr-xr-x@ 1 gjtorikian  staff  228448 Dec  4 11:51 parser.so

json-2.16.0-x86_64-linux/lib/json/ext/3.2:
-rwxr-xr-x@ 1 gjtorikian  staff  287944 Dec  4 11:51 generator.so
-rwxr-xr-x@ 1 gjtorikian  staff  220600 Dec  4 11:51 parser.so

json-2.16.0-x86_64-linux/lib/json/ext/3.3:
-rwxr-xr-x@ 1 gjtorikian  staff  288816 Dec  4 11:51 generator.so
-rwxr-xr-x@ 1 gjtorikian  staff  222528 Dec  4 11:51 parser.so

json-2.16.0-x86_64-linux/lib/json/ext/3.4:
-rwxr-xr-x@ 1 gjtorikian  staff  290976 Dec  4 11:51 generator.so
-rwxr-xr-x@ 1 gjtorikian  staff  223024 Dec  4 11:51 parser.so

Cool, that's what I wanted to check: the Ruby 3.x versions for each native compilation in its own folder. So now I'm more certain that tmp can be dropped.

[Edouard-chin]

Hey! Thank you for the investigation.

Ah great point I forgot about temporary folders that gets uploaded during the compilation phase. I do this in order to test the binary in other CI jobs in order to not have to recompile (and ultimately save time).

1.8MB is the size of the gem with precompiled binaries as you pointed out. I don't know whether this is acceptable for a small gem like json. By comparison, the gem with precompiled binary on MacOS is only 300KB.

[gjtorikian]

Probably the "best" solution is to change bundler to figure out a way to package and fetch gems by Ruby version. Since fat gems need to bundle together every compiled portion for all Major.minor, it would be nice if one could push and pull specifically from x86_64-3.0, x86_64-3.1, etc. I'm not sure there's any way around it.

[Edouard-chin]

Yes totally agree 👍 . There was an RFC drafted in RubyGems recently about this https://github.com/segiddins/rfcs/blob/segiddins/wheels/text/0000-improved-binary-gem-support.md

[gjtorikian]

oh nice!