Alessandro Gatti

@dpgeorge the new changes look good to me as well! Just one minor thing: given that picolibc for Arm is also an option on the CI machine, would it make...

Interestingly, Arm targets with picolibc are ever so slightly larger than their newlib counterparts: ``` Picolibc text data bss dec hex filename 181640 16 103124 284780 4586c build-MPS2_AN385/firmware.elf 264044 16...

@dpgeorge I was also wondering, given that there's no RV32 platform in the code size report then maybe adding the size of the qemu repl build there could be nice...

@dpgeorge you may want to increase the heap size for RV32, I just bumped into https://github.com/micropython/micropython/pull/12853#issuecomment-1798784848 again.

Sure, take a look at the test results of https://github.com/micropython/micropython/pull/15714: https://github.com/micropython/micropython/actions/runs/10700895755/job/29665631825?pr=15714 - I rebased the code to be compatible with the merged qemu port. ``` 740 tests performed (22638 individual...

I can also reproduce the issue locally on an Ubuntu 22.04 VM when running tests through `source tools/ci.sh && ci_qemu_build_rv32`. Increasing `HEAP_SIZE` from 100k to 120k makes the test pass...

Added test to keep codecov happy...

Thanks for the explanation! From some random testing I've made the only other place where I was able to reproduce the issue on the REPL is when handling unary operations...

Ok, I've let `afl-fuzz` go wild for a couple of hours using `annotate_var`, `del_deref`, `del_local`, `scope_implicit`, and `unboundlocal` with all functions marked as native and `native_misc` with the extra test...

> I'm surprised fuzzing didn't find more cases. For example the following easily crashes the unix port: It started finding a few more when I upgraded AFL++ to the latest...