Crash in recursive function with many intermediate allocations #55192
Labels
Comments
|
Should the GC segfault not get its own issue? |
|
I suspect you are very close to the stack edge when you called the GC and just calling a few sweep functions made you overflow. Instrumentation: diff --git a/src/signals-mach.c b/src/signals-mach.c
index ad5788ea23..5cce9fe715 100644
--- a/src/signals-mach.c
+++ b/src/signals-mach.c
@@ -296,6 +296,13 @@ static void segv_handler(int sig, siginfo_t *info, void *context)
jl_call_in_state(ptls, (host_thread_state_t*)jl_to_bt_context(context), &jl_sig_throw);
}
else {
+ ucontext_t *ucontext = (ucontext_t*)context;
+ // read the fault address
+ uintptr_t addr = (uintptr_t)info->si_addr;
+ // read the stack pointer
+ uintptr_t rsp = ucontext->uc_mcontext->__ss.__sp;
+ jl_safe_printf("Fatal error: %s (%d)\n", sig == SIGSEGV ? "Segmentation fault" : "Bus error", sig);
+ jl_safe_printf("Faulting address: %p, stack pointer: %p\n", (void*)addr, (void*)rsp);
sigdie_handler(sig, info, context);
}
}Results: Judging by how suspiciously close the faulting address extracted from Though I think that the error message here could be better. |
|
Duplicate of #53033 |
|
For what is worth, I wasn't asking to add tail-call optimisation, @vchuravy suggested on slack to open the issue because this code shouldn't blow up the stack (and this wasn't until v1.10) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
While reading https://www.modular.com/blog/mojo-vs-rust-is-mojo-faster-than-rust, I wanted to see how Julia fares with the TCO example. I tried this translation:
The stackoverflow error is new in Julia v1.11, it works up to Julia v1.10:
For the record, LLVM IR of the julia version:
Julia LLVM IR
For comparison, the LLVM IR of the Rust version, which doesn't blow up the stack, from the Mojo blogpost:
Rust LLVM IR
Side note, it'd be amazing if the compiler would detect that all the allocations and stores are unused and it'd just elide them 🙂
Tangentially related, at the beginning I wrote a slightly different implementation of the recursive function which has much better performance, but on Julia nightly (same version as bove) it runs in a segfault in the garbage collector for larger arrays:
The text was updated successfully, but these errors were encountered: