Build system wants a newer version of Gambit than the README

[rrthomas]

The README says version 4.7.5+ of Gambit is acceptable, but when I make rsc.exe on my Ubuntu 24.04 system with Gambit 4.9.3 installed from the Ubuntu package, the build system complains:

WARNING: You are using an old version of Gambit, and some ribbit features might not work properly, especially regarding R4RS symbols.
It is suggested to use Gambit v4.9.4+. Type 'gsc -v' to see your version.

It looks like the README needs updating!

[leo-ard]

Thanks for the feedback! It might not be clear from the message (and Readme), but you can use Gambit v4.9.3 to compile/interpret Ribbit. However, some edge-cases around symbols will not work properly. For the best experience, use v4.9.4+. If you just want to try out Ribbit, v4.9.3 is fine.

Do you think this message would be clearer? Feel free to propose suggestions.

WARNING: You are using an old version of Gambit, and some Ribbit features might not work properly, especially regarding uppercase/lowercase R4RS symbols.
You can still use this version to try out the compiler, but it is suggested to use Gambit v4.9.4+ for the best experience. Type 'gsc -v' to see your version.

and in the README:

• Gambit:
  • v4.9.4+ (Recommended)
  • v4.7.5 to v.4.9.3 (Fine for trying out the compiler, but not for compiling large R4RS programs)

Sadly, the Gambit Ubuntu package is quite old (v4.9.3 came out in 2019, almost 7 years ago!)

[rrthomas]

As a project mainly of interest to developers, and given that the minimum required versions are current in current distros, I would not sweat supporting older versions. If you like, make the requirement a soft requirement and emit only a warning rather than an error (as at present), but I would update the README and give readers a strong impression that they should install a supported version of the relevant Scheme compiler, and expect breakage otherwise.

If it were me I'd make it a hard requirement and refuse to compile with <4.9.4.

I sympathise with making it easy for users to try things, but I think you're mostly doing them and your project a disservice when you allow and even encourage compilation with versions of compilers that won't work in some cases, as opposed to merely not supporting some features.

For example, I'm using a supported-but-old Ubuntu LTS version. Had the build system simply said "nope, you gotta have ≥4.9.4", I would've happily installed it from brew, source, or whatever, just as I do when testing the current version of other projects that have recent dependencies. And similarly, I quite happily impose recent deps on my users: from my point of view, the "next release" of my software has to be amenable to packaging for distros' next versions, not to compilation with the current version. Of course I don't do this gratuitously; if there's no advantage to requiring a new version of a dependency, I don't; and I assume you don't either!

[leo-ard]

Hey! Thanks for the feedback. After reflections, you are right that providing a buggy version is probably doing a disservice to users. I'm more convinced that a hard error (that can be bypassed if you know what you are doing) should be the path to go forward.

After further investigation to see the exact version to block, I realized that v4.9.3 works fine with Ribbit (all tests pass!). I know however that v4.7.5 fails, so the right version to support is somewhere between 4.7.5 and 4.9.3. I'll investigate further to pinpoint the version and update the readme/makefile/CI accordingly

[rrthomas]

Again, as a lazy maintainer myself, I would simply require >= 4.9.3 for now, and if anyone complains, try bisecting. Or indeed ask them to test their preferred version.

[leo-ard]

I might have to learn to be lazier 😅. I know 4.9.0 works, I'll require 4.9.0

[rrthomas]

Remember, laziness is one of the 3 cardinal virtues of a programmer!