hrm.. i don't know if anyone knows offhand if there's a way in node.js to distinguish between solaris and illumos without having to run and parse the output of uname -o ?

jbk: os.version() is likely to be distinct

(even if it's just an easier way to do the same as you've suggested)

Successfully using the latest LLMs for code generation seems to me to hinge upon breaking up tasks into pieces that are clearly defined. Using something for project management (beads) can help. Tell LLM to create a detailed plan for conversion that breaks it up into manageable pieces. Then assign or auto assign various agents to each of the tasks. In my experience, the problems come when the planning and ex

ecution are mixed. Without specific instructions for a manageable duration task, the goals seem to get lost in the growing context. The latest LLMs are shockingly consistent at completing small tasks (although typos do creep in). Codex rebooted my smartos box due to a quoting error while searching a log for "reboot."

Dixie_F: :D

b.exp

oops

gwr: circumstantial evidence from aarch64 automated tests: the libc symbol tests are half of my runtime for the (non-filesystem) test suites

and there's one fewer ISA on aarch64

so I imagine your header-test thing will make a bunch of people happy :)

What do the make variables ARCH and ARCH64 expand to on ARM64?

both expand to aarch64 for unfortunate reasons

but uses are guarded based on whether the platform is multilib or not

I'd like to figure out at build time if the target for which we're creating things does 64-bit, 32-bit or both.

gwr: I just put up a huge fucking ugly PR to the arm64-gate for that

I designed it not to be clean, but to merge easily for me

if I were doing it properly, I'd probably do the `$(MACH)/Makefile`, `../Makefile.com` dance, but I converted one suite to that and did some experimental merging and it was not going to fly

the gist of it is ISA-d binaries land in 32/ and 64/ directories, so we can guard one against i86pc in the runfiles

and still uses test groups

(that is, the directories only exist to pacify test-runner)

I think the other thing to know is that for obvious reasons arm64-gate has a `$(BUILD32)` (cf. $(BUILD64))