Support setjmp and longjmp?

[rinon]

Although we are unlikely to support it in the near future, this issue will track any ideas and progress on supporting setjmp/longjmp.

[XVilka]

You might employ "No more GOTOs" paper: https://net.cs.uni-bonn.de/fileadmin/ag/martini/Staff/yakdan/dream_ndss2015.pdf
It is even implemented (sort of) in radeco decompiler: https://github.com/radareorg/radeco/commits?author=HMPerson1

While it is not directly applicable, still quite relevant, and might give you ideas as well.

[rinon]

Thanks, I'm actually familiar with the paper. It only applies to transforming local (intra-function) control-flow. We use a variant on the Emscripten relooper algorithm (http://mozakai.blogspot.com/2012/05/reloop-all-blocks.html) to do a similar thing.

However, setjmp/longjmp can be used to transfer control across functions not just in the same function, so these techniques do not apply, at least not directly.

[bjorn3]

According to https://smallcultfollowing.com/babysteps/blog/2016/10/02/observational-equivalence-and-unsafe-code/ setjmp/longjmp is incompatible with the rayon crate, which provides safe scoped threads.