mcc
Wasm question
I was given https://hachyderm.io/@unlambda/114582671790895240 as some example code for printing to stdout from inside wasmtime/WASI
It calls a function named "fd_write" taken from "wasi_snapshot_preview1"
"wasi_snapshot_preview1" does not Google well and one link https://github.com/WebAssembly/WASI/blob/main/legacy/preview1/docs.md seems to suggest this is deprecated. How were the functions in wasi_snapshot_preview1 chosen? Is there a wasi_snapshot_preview2 or whatever?
✧✦Catherine✦✧
@mcc there is preview2 (and preview3) but the latter two are component model based. component model is an additional layer of complexity on top of wasm, comparable with wasm itself in terms of the amount required to learn
✧✦Catherine✦✧
@mcc the functions in preview1 were chosen based on something called "cloudlibc" / "cloudabi"
mcc
@whitequark all right. if i'm trying to implement "putc" by pulling fd_write out of preview1 and telling it to print a single char from a single iov , am I to your knowledge missing a simpler method?
✧✦Catherine✦✧
@mcc this is the simplest thing you can do to accomplish this task in plain wasm, yes
mcc
By the way if anyone has an easy way to point me to any of the following it would be much appreciated
- An example of the (block) syntax in textual wasm
- An "alignment" is described in "memarg" in the spec, but this shows up nowhere in the textual wasm examples I find. Is this implicit in the use of `i32.` instructions?
- Is there a way to query the current wasm stack depth?
✧✦Catherine✦✧
@mcc block example: https://developer.mozilla.org/en-US/docs/WebAssembly/Reference/Control_flow/block
querying stack depth: no, you're not allowed to do anything that would make stack depth dynamic, therefore there is no need to have an operation to query it
mcc
OK so
I'm struggling with this week due to many issues that feel to me artificial. I don't like how this spec is written. I don't like a lot of the decisions
Buuut *sigh*
I guess I got the Good Idea out of the wasm spec I was hoping to find when I picked it. I was feeling like this S-expression form didn't "feel like" a LISP. And then I found this bit in the PDF spec. Using a syntax reordering trick to represent a Forth in LISP style. That's *saying this as if miserable* really smart actually
✧✦Catherine✦✧
@mcc alignment: i think it works like this:
i32.load align=2
mcc
@whitequark well that's good to know because the next thing i was about to do was use a counter in a local to make the stack depth dynamic lol
✧✦Catherine✦✧
@mcc I do emphasize the similarity between the Wasm stack and LLVM variables here! you can't create new LLVM %n variables at runtime in the same way
the Wasm stack is conceptual, it is not an actual physical stack you can use for normal stack type things. it's basically a compression mechanism (oversimplifying)
mcc
@whitequark by the way, the block example, Katelyn G was saying there's some sort of unusual limitation on which basic blocks may follow each other, is there anything that stops me from br'ing between blocks?
✧✦Catherine✦✧
@mcc Wasm does not have basic blocks. (block) is structured control flow: think {...} in C, and there's no `goto`
mcc
@whitequark fuck's sake this is worse than LLVM IR
✧✦Catherine✦✧
@mcc it allows for linear time verification, which is pretty important for the web, where anything you can encounter can be malicious. not having the equivalent of a zipbomb in my wasm engine is pretty important
✧✦Catherine✦✧
@mcc that being said, it's a pain, and even lowering LLVM IR to Wasm is... nontrivial. when I build Yosys (which has dozens or hundreds of `goto`s) to Wasm, LLVM spends about 10 minutes in codegen resolving those `goto`s to reducible control flow
✧✦Catherine✦✧
@swetland @mcc yeah, I also found the spec difficult to navigate (to make it worse it has some obnoxiously resource-hungry javascript that renders formulas)
my understanding is that the spec is written with one main consumer in mind: people doing formal verification of the spec. the secondary consumer is people with a lot of PL experience writing runtimes. everyone else isn't really considered, in practice if not in theory
Brian Swetland
@mcc @whitequark Between that and how just incredibly convoluted the spec is (any needed piece of information on an opcode or how it works is inevitably spread between 3-5 different sections of the spec), WASM really makes my head hurt. There are some good ideas in there and people are doing good things with it, but from a "sit down and understand how it all works" standpoint it's a pain in the butt.
mcc
@whitequark @swetland really startling to find a situation where I'm like "this HTML page is so hard to read and navigate, I'll switch to the PDF ahh that's better"
✧✦Catherine✦✧
Brian Swetland
@whitequark @mcc That definitely aligns with my experience trying to read it from a "how do I build a functional WASM runtime" and "how do I generate correct WASM binaries" standpoint.
Some notes from my initial encounters with it illustrate specific grumbles with the presentation:
https://chaos.social/@swetland/111271568378452789
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.