gm!

Where would it plug

i tried to do it the normal way at first but after like 10 minutes and some blood on my fingers i was like "Cannot hold the pieces together if it's literal dust". It cannot

what's the emcc stack size nowadays? wasi-sdk's is 64K which is really way too small (but more contentious to change and I won't even try)

re: stack overflows: despite wasting a significant % of my time porting on them, I spent several years not even reporting them on the tracker, so I am not convinced this is a good proxy for how often people encounter them

maybe what could be useful is a -Wl,--output-fliter,somecmdline option to wasm-ld that would let the end user filter the output via wasm-opt, stackcheck, etc?

and it's not that easy to wrap a full clang invocation in a shell or python script to manually run binaryen afterwards (which I'm sure you're aware, I'm just enumerating the possibilities here)

I actually think this difficulty is somewhat of a problem on its own

I would be very _very_ happy to have the binaryen pass enabled in the pipeline, but afaik technically including it in the build pipeline is somewhat tricky (wasm-ld used to invoke wasm-opt if it found it in $PATH and that was was a pretty bad decision, binaryen would be subject to similar issues...)

eyes...

UBSan, ASan and LSan do not work at all on WASI SDK (there is an open issue but it seems difficult to port), binaryen isn't available in a default configuration (and not trivial to integrate in a flow even if it was), there is no documentation for how to do any of this either

I think you are making ane error I see quite a few people make: equate WebAssembly with Emscripten

I ~only use WASI SDK, which does not have the mitigations. are you collecting data for these users?

ye, it's normal nowadays

there is and i can't imagine anybody removing it, either (it would frankly make no sense given thread stacks are heap allocated anyways...)

(I think we basically agree and I don't object to the existence of --no-stack-first, but I am quite tired of the "I want my program to be fast, nevermind it calculates some complete nonsense" style of argument over the years...)

I've heard that argument, yes. I do not really care how big my binary is if it crashes in mysterious ways

the answer is "conditionally yes" depending on the type of SSD, but even if the architecture is suitable you will need a lot of knowledge about the drive internals (mainly the whitening algorithm) to be able to make use of the memory captures

define "proper"?

yep!

huh, I didn't realize the coverage area is that small; that makes it sound like putting a few pirate cubesats into LEO is cheap and useful enough that it's a matter of time before someone does it!