hello all, I have question on this link: pastebin.com/NjQT1Wy4 , any advice? best wishes

hm, how do I tell gcc that cmn_err(CE_PANIC, ...) does not return? it whines that "control reaches end of non-void function" otherwise

en.cppreference.com/w/cpp/language/attributes/noreturn ?

or just return something

well, my point was not having useless return :)

it is a bit annoying since it appears smatch understands but gcc does not

i've had to add a superfluous return surrounded by #ifndef __CHECKER__ to make everything happy

yuripv: You _could_ just call panic()

yeah, there's no valid way to say "returns _sometimes_" in a way the compiler fully understands

you can return once, never, or more than once.

this is basically the thing sommerfeld ran into with ASSERT()

but rather than `aok` it's the flag argument

Hey! Is it possible to increase iscsi target tcp rcv buffer size? Setting it via 'ipadm set-prop -p recv_buf=1048576 tcp' and 'svcadm restart iscsi/target' do not change that for the tcp socket.

sommerfeld: I looked at your rationale for nightly bringover of git, and you're basically avoiding all the points that caused me to not do in the first place. (I'm not sure if `git worktree` even existed then).

seems really handy, with those limitations placed on it.

Probably iscsi target sets it some another way, however tcp_create_common() is executed when do "svcadm restart iscsi/target"

sommerfeld: a suggestion, though I haven't looked at the code yet... if we could specify bringover/target ws on the nightly command line, you could share an env file and just say `nightly --parent /builds/richlowe/arm64-gate/illumos/gate --builddir /tmp/foo env/shared`

(I'm aware I could do that setting CODEMGR_WS and BRINGOVER_WS on the command line, it just seemes neater not to)

richlowe: yeah, I haven't shared my WIP just yet.

Good idea to put the dirs on the command line - I'll consider doing that as a separate change.

jclulow: thanks, that's what I did eventually, just thought that it's somewhat "unclean" as cmn_err() seems to be used much more than panic()

I'm not using `git worktree` as a git branch can IIRC only be checked out in one worktree of a repo at a time. using the objects/info/alternates mechanism is nearly as lightweight as worktrees.

richlowe: unlike aok, the cmn_err(CE_PANIC, ..) thing is at least theoretically possible for an sufficiently advanced compiler to get right because it's (almost always) a compile-time constant at each call site rather than a global variable that can be patched in /etc/system or kmdb..

richlowe: anyhow, back to "bringover_git": I'm curious about what gotchas you were thinking of (I have my own but I'd like to hear more on why you punted)

I punted almost entirely because I couldn't think of a way for merging to work out right, or even ok

the former joyent people might remember more details, they were big on git.

I keep forgetting about "aok"

Probably to protect myself