indolering (Mar 15 2026 at 02:57): indolering (Mar 15 2026 at 02:57):I've been tracking the blockchain VM space, and it seems like a lot of projects have converged on RISC-V-based VMs. What I find interesting is that they're lowering to a RISC-V VM as a distribution artifact. Has that influenced how y’all think about WASM and the Cranelift IR stack at all?
Chris Fallin (Mar 15 2026 at 04:17):Not really? Wasm is a reasonable portable ISA all on its own; Cranelift follows a pretty established path in terms of abstraction level for a compiler IR (SSA CFG); I'm not sure what we would gain by trying to use a specific ISA as an IR rather than our existing one
Chris Fallin (Mar 15 2026 at 04:19):In particular note that there is value in reifying Wasm semantics in an IR that is close to Wasm (CLIF takes explicit inspiration from Wasm for e.g. floating-point semantics and other things); and having a separate IR (VCode) close to the machine ISA; and an explicit lowering between them. I wrote about this ~5 years ago at https://cfallin.org/blog/2020/09/18/cranelift-isel-1/ when I introduced the design. Using a low-level thing "in the middle" would scramble all of the nice architectural properties we have there
Chris Fallin (Mar 15 2026 at 04:19):Maybe you could say more specifically what sort of thoughts you have here?
indolering (Mar 15 2026 at 04:21):I just find it really interesting that the unit of deployment for Ethereum code has turned out to basically be a RISC-V shaped virtual ISA. But I guess that has a lot to do with lowering the requirements to just zk proving.
Chris Fallin (Mar 15 2026 at 04:26):Sure. I think they're solving a pretty different problem than Wasm; that's as far as I've thought about it
indolering (Mar 15 2026 at 04:28):Do you have thoughts on Determinate Nix's use of WASM tech for reproducible build systems?
Chris Fallin (Mar 15 2026 at 04:30):"hey that's pretty cool!" (more seriously: as someone who tried to use Nix for a while and then "fell off the wagon" back into impurity, I think it has much bigger papercuts that still need solving, but allowing little pure computations as part of builds is neat, and can't hurt)
indolering (Mar 15 2026 at 04:31):Care to elaborate?
Chris Fallin (Mar 15 2026 at 04:34):About which part?
indolering (Mar 15 2026 at 04:35):What needs fixing in Nix?
Chris Fallin (Mar 15 2026 at 04:35):I mean, that's a big topic and not really something I can offer you a ready-made essay on at the moment, sorry
indolering (Mar 15 2026 at 04:36):It's a saturday evening, thank you for taking time to respond to me.
Chris Fallin (Mar 15 2026 at 04:36):I will say that I hit several "I have no idea what's going on and it seems the build is broken" moments in the span of a few months while defining much of my dev environment under Nix, and the overzealous compiler patching always bit me in unexpected ways
Chris Fallin (Mar 15 2026 at 04:37):I got tired of trying to "work around" the thing and decided that going against the grain wasn't really a good use of my time
Chris Fallin (Mar 15 2026 at 04:38):I went down the route of "maybe I can use nix-ld to define an environment that looks just like my Debian dev machine" before, well, switching back to my Debian dev machine shrug
indolering (Mar 15 2026 at 04:42):Yeah, the friction is real. I'm trying to use an OSTree distro and Bluefin is just polished enough to make it worth the hassle.
Last updated: Mar 23 2026 at 18:16 UTC