http://www.eluaproject.net/doc/v0.9/en_arch_ltr.html
As for Rust, why do we think that should use significantly less memory than MicroPython or Lua? I've seen some threads with people doing measurements on simple programs and finding Rust up to 10x heavier on RAM than C.
http://stackoverflow.com/questions/32762102/why-do-rust-prog...
There's nothing about Rust that intrinsically makes it larger than C.
As a side note, there are quite a few options to do safe micro controller system programming for those that prefer AOT compiled languages without C like unsafety.
http://www.mikroe.com/mikropascal/
I think Forth is much more suitable here. You can implement a lot of features in 512K. You can extend the code (and language!) at runtime, you can debug remotely, and do other nice things.
One would wonder how we managed to cope before Rust. Also, you should ask these guys [1] [2] how they know they're using memory safely. Has Rust ever been used to fly aircraft and fricking space shuttles? I didn't think so.
[1] https://web.archive.org/web/20110204160744/http://forth.gsfc...
However the ESP8266 is built around Tensilica's Xtensa, not ARM's Cortex. There currently is no support whatsoever for Xtensa in LLVM (let alone in Rust).
Current microcontroller work is on MSP430[3] with work being planned for Amtel AVR[4] since the AVR branch was merged into LLVM in November 2016. Xtensa is quite literally nowhere[5].
[0] guaranteed to build & binaries provided but tests are not automatically run
[1] supported in the codebase but no guarantees that it even builds
[2] only Core, no standard library
[3] https://github.com/rust-embedded/rfcs/issues/20
DOH!
I have to say, though, that having the AVR code generator merged is very cool. At one time I was a big user of avr-gcc and remember when the AVR LLVM effort was just getting started. I haven't been following it, though.
One of the annoyances with gcc is that the optimizer kept wanting to use code motion to relocate the interrupt disable instruction to the beginning of the function it was found in. The only way to keep the optimizer from doing that would have been to make the instruction depend on all of RAM, which would effectively disable optimization for any function that disabled interrupts. So coding critical sections required a bit of jiggery-pokery to keep the optimizer from breaking your code. I hope the LLVM back-end found a way around that.