richlowe: I've found when that happens I need to git fetch the remote is all.

[illumos-gate] 17091 Use DWARFv4 by default -- Robert Mustacchi <rm⊙fo>

Hey, Do I understand right, that currently FC QLogic 2770 Series Fibre Channel Adapters (32Gb) is not supported in illumos?

i don't think the driver has been updated for newer qlogic fc cards, just the emulex ones..

i do recall looking at the changes on freebsd to their qlogic driver to support the newer models, and there weren't too many, so might not be too bad to do if someone's interested (need HW to test)

jbk: Thanks!

jbk: I hope to add support for this if we have enough HW.

did you two ever connect on the rdma ib question?

sort(1) tell me that `-z` is an option without an effect for illumos' `sort` command.

So, how about using this option for NUL-delimited output, as GNU, FreeBSD and OpenBSD do?

msiism: that's not the sort of backwards-incompatible change we make lightly. some distributions ship GNU sort in /usr/gnu/bin so there's a simple workaround (set PATH in/around scripts that need that form of -z).

Good to know, thanks.

erm, if the option is listed obsolete, then we should not be concerned about backwards compatibility. Otherwise this word has no meaning.

I guess the problem is with this option being used in scripts from a time when it did have an effect.

Now if you cahnge that to having a new effect, things will quite likely break.

Or even from when it didn't, but using it with the parameter like `-z 1024`.

Interestingly, Solaris 11 has removed -z from /usr/bin/sort - but not from /usr/xpg4/bin/sort, which seems a little odd

well, fortunately -z and -z recsz are possible to check. But also, *if* this option is really obsolete, using it should give out an error anyhow.

Maybe there could be a deprecation period after which that option is removed. After waiting for another bit, it could be introduced as a new option.

since there is a difference in usage - with or without argument, the implementation can check this and issue a warning/error or switch the behavior accordingly. In that sense we can have zero byte processing without the wait. and just wait till we drop -z recsz.

at least in that sense it is much more easy way out than it is with file(1), which has conflicting options.

I was trying to find when -z was declared obsolete, but tracks end there: illumos.org/issues/243, therefore it must be older than 2010 :)

→ FEATURE 243: system manual pages should live with the software (Resolved)

The manual page I'm reading is dated 2001-11-19.

Reading it at illumos.org/man/1/sort.

It's exactly the same (obsolete) in Solaris 10, so that's a bit further back

and because of separate consolidation, its not present in s7/s8 sources:P

ptribble: because XPG4 specified the -z option - it was removed in later standards: pubs.opengroup.org/onlinepubs/007908799/xcu/sort.html

Thanks for the pointer! Crikey, required by a standard.

looks like Solaris sort made -z obsolete for the 2.0 release when it was changed to use mapmalloc - predating any sort of ARC review

aww

tsoome: I see sort under ON in the S7 & S8 sources - there was a major refactoring for S8, in S7 it was simply usr/src/cmd/sort/sort.c

I meant manuals:)

oh right

which is also why I don't have a copy of those man pages in our source archives

I was working in my alma mater back then and we got the sources when sun started that project to make them available for universities:)

yeah, I was at Berkeley then and remember we could buy the source for $50k if we ordered it by itself, or $50 if we included with another order, so we ordered a cdrom drive to get the source cds

I think they simplified that later

yea, we were buying something every now and then anyhow:)

rmustacc: the fix is good to know, the symptom is _terrifying_

rmustacc: because as sommerfeld said, the "missing object" is an important tree

FWIW, an option being obsolete but harmlessly ignored _is_ a kind of backwards compatibility. Turning obsolete options into hard errors is breaking backwards compatibility.

Also, making option arguments optional is not great -- it makes the parsing ambiguous in some cases. POSIX expressly says don't do that, in guideline 7 of the utility conventions

CLIP too

"optional option arguments" has long been a no-no

(I missed _all_ the context of this)

sort -z

ignoring it and its argument seems fine and compatible, using it for something else seems incompatible

"incompatible with _what_" is perhaps worth asking

right

It's possible that nobody has used that argument in anything for 25 years, but actually being sure about that difficult (if possible) and will require work, not just conjecture, etc

*is difficult

I realize compatibility is important, don't take this question as more than it is: How in the world is GNU `sort -z` not `sort -0`?

am I the only one who thinks of "the xargs way" to do that?

is `-z` somehow prevalent and I never noticed?

OpenBSD and FreeBSD have it too. Also, Busybox.

It's old in GNU sort (30 years), but the comment that came in with it mentioned -print0 so no clues about why it isn't -0 yet

meanwhile it's grep -Z and xargs -0 ;)

ah, grep -z actually

limitless backwards compatibility is just as insane as no consistency in options;) so no need to point the fingers there.

dangergrrl: I don't have ib HW. Did you mean that?

vetal_, you had been asking what ib drivers had working rdma iirc

dangergrrl: I was asking about RDMA Ethernet.

nm

historically, there hasn't been a huge amount of interest in RDMA on illumos... there's the bits that came from Solaris, but that's about it

doesn't mean that can't change, just that's been the history

historically the vendors are not among the friendliest

mostly there's been a small group of (I think) HPC people who ask about it every year or so

mostly in the context of hoping someone else did it already

and I know taking driver support for things from linux is a mess

looking at a couple of those rn :)

mess is one thing, but you do not want to draw yourself into the license wars.

I have some of those already just for working on non gpl projects at all.

there will be some rdma ethernet work using illumos soon (tm)

ib ... is not even the main focus of nvidia in future

wrt license wars, the cuse webcam driver I'm looking at finishing porting to dragonfly is a useful avoidance of gpl code in their kernel

by soon i mean maybe in future :D

would be something nice really.

So yeah, sitting logged in as root, 37MB avail out of 64MB is a nice start.

D-oh, wrong channel. :D Pay no attention to my 486 shenanigans.

I did some research on `sort -z`, github search finds nobody using the solaris way (that I can find, more specific patterns are crashing it), and a great many wanting the GNU way.

but that perhaps says more about the limited ability to tell what'd break something by searching like this

Anyone pxe booting chelsio cards? Evidently Supermicro is having issues "Validating" that PXE is working in our latest SKU for chelsio.com/nic/unified-wire-adapters/t62100-lp-cr