heh..

so a coworker managed to get illumos to cross compile on macOS

(not sure how the resulting binaries look, but you can at least run everything through the compiler)

macos intel or even arm?

intel, but it's building the cross-compiler illumos gcc, illumos ld, so hopefully we can test it on some arms as well (still producing x86.. i don't think anyone here aside from me has looked at the arm stuff since it's still under development)

i might see if I can get all the tweaks upstreamed -- i've not looked at them closely yet, but from first glance, they seem like either legitimate things we should fix regardless, or just a 'yeah, that's probably not good practice'

nice

Rich's ld(1) fix should help.

one of the bigger things is it had to skip some symbol check on all of the IP modules I wasn't aware was there

(the '$MODULE.global-objs.{obj,debug}64' files)

though that might be able to be fixed with a little work

whatever 'sis_check' is meant to be (appears to be something from Sun)...

I found out about those checks during gcc10 bring-up, and they were useful in spotting differences in what the compiler was doing. They do generally get in the way of things though, including some of the new build systems that people have experimented with over they years!

yeah, but if you look.. it appears the only things that use them are all the IP related modules

the rest of the kernel does not

Yes. Just saying that I was glad they are there (even if it was an experiment that didn't get expanded to the rest of the tree), but I appreciate they're a pain for things like you're doing.

the name 'global-objs' also is a bit confusing since it's looking at all objects defined in the module, not just global ones

(why it'd be nice to understand the reason for it, and why it's only the ip related modules)

danmcd: so I take it 'sis check' doesn't ring any bells from your Sun days?

I don't think so?

heh

(I wonder if no one knows what it's for if maybe we can just get rid of it)

if it's some private sharing of symbols amongst the modules, we could probably handle that with mapfiles explicitly exporting what's needed by others (and we'd be able to document it as well)

or on the consumer side, noting it's extern (and again documenting)

[illumos-gate] 18155 Typos of "pseudo" -- Dan McDonald <danmcd⊙ei>

the IP global objects checks were introduced with private netstacks, to prevent people adding global state that wasn't zone-aware

ahh

We should do that with NFS too once we finish the job there (/me side-eyes rpcsec ....)