<rjo>
whitequark: it is a bit weird that you actually have to read it from bottom to top to "tell the story"... but also llvm does that to me all the time.
<whitequark>
should probably quote "duration" there
<rjo>
"while doing A we hit B which then resulted in C". A, B, C being the items in that message in reverse order.
<whitequark>
yeah, I parroted the structure from LLVM
<whitequark>
usually LLVM does the right thing. though not always. I am open to change here
<rjo>
the problem is generally that the "notes" refer to contextual information that is "outside" and "wider" than the actual error. when reading it, one would want to reverse them. same way people initially get confused by a traceback because -- while the call trace is top-to-bottom -- the error actually propagated from bottom to top.
<whitequark>
well
<whitequark>
I'm not convinced that Python got tracebacks right
<rjo>
anyway. my grit is that "more information is not always better" but here it is certainly good enough.
<GitHub124>
[artiq] whitequark pushed 2 new commits to master: https://git.io/vzyoJ
<rjo>
i would guess that a previous test messes with the asyncio global state and pipe_ipc stumbles over the residue.
<whitequark>
oh right
evilspirit has joined #m-labs
fengling has quit [Ping timeout: 272 seconds]
fengling has joined #m-labs
<cr1901_modern>
Use the msys2 version of LLVM? Does all ARTIQ do is call LLVM binaries via subprocess()?
<whitequark>
ARTIQ does not call LLVM binaries
<cr1901_modern>
I think I meant MiSoC lol. Basically my question is- you think anyone will notice if you swap out a MSVC-compiled LLVM with an MSYS2-compiled LLVM?
<whitequark>
dunno, I am not about to find out because that's more work than installing VC12
rohitksingh has joined #m-labs
sb0___ has joined #m-labs
<sb0___>
whitequark, rjo: note the ImportError: cannot import name 'pipe_ipc' above
<sb0___>
so it's not about asyncio
<whitequark>
bb-m-labs: force build --props=package=llvm-or1k conda-win32
<bb-m-labs>
build #6 forced
<bb-m-labs>
I'll give a shout when the build finishes
<whitequark>
sb0: all our 'old' packages are uploaded ie current ARTIQ should function on Windows.
<whitequark>
I will now proceed working on Qt5.
sb0__ has joined #m-labs
<sb0__>
whitequark: good
<sb0__>
can you try it? this shit internet connection i have right now blocks vnc
<whitequark>
sb0__: it doesn't
<sb0__>
everything but 80 and 443 are blocked
<whitequark>
ssh lab -L 5900:localhost:5900
<sb0__>
ssh is blocked
<whitequark>
oh
<whitequark>
fuckers
<sb0__>
i guess you could configure the openssh server to listen on the https port... but are you in hk so you can fix it in case it doesn't restart?
<whitequark>
sb0__: I put ssh on 443
<whitequark>
it's not necessary
<whitequark>
ssh does not kill existing connections when you restart it
<whitequark>
it forks and all active shells remain active
<whitequark>
but that didn't even matter because openssh has a reload command
<whitequark>
sb0__: later we should set up https-only proper and instead put ssh on 80
<whitequark>
(I'm waiting for letsencrypt to become usable on stable debian)
<whitequark>
oh, the buildserver is testing. so, usable on testing.
<whitequark>
it's still in unstable only I think.
<sb0__>
ssh: connect to host lab port 443: Connection refused
<whitequark>
oh I need also add a forward
<whitequark>
sb0__: done
<sb0__>
works, thanks
<whitequark>
lucky. they don't do DPI.
<whitequark>
please do another separate install of miniconda wherever, to not fuck up the builders
<whitequark>
oh, I will later make a snapshot of this so that you could run wild with it
<whitequark>
or maybe I should do it now (requires reboot)
<sb0__>
of course, the remmina vnc client craps out on my hi-dpi display
<whitequark>
do not expect to be able to use stock vnc client on that vnc
<whitequark>
it will not work
<whitequark>
i've tried at least four
<sb0__>
it displays the windows desktop, but zoomed in and the mouse doesn't work
<whitequark>
yes, it doesn't update the display
<whitequark>
because to make it send you new pixels you need to do wildly noncompliant things
<whitequark>
so that it would do wildly noncompliant things in response and send that shit anyway
<sb0__>
can anyone connect to that 5900 port and get the windows desktop?
<whitequark>
anyone with ssh access to our buildserver, yes
<whitequark>
deliberately not exposed to the internet and will not be exposed to the internet because there are hordes of thirsty bots that scan and own vnc endpoints
<whitequark>
not even on a different port with a password
<whitequark>
you should see the stuff I've seen on shodan
<whitequark>
like coal power plants in Netherlands or waste treatment plants in Germany
<whitequark>
with unauthenticated vnc
<sb0__>
so how do people who did not write a vnc client use the qemu stuff?
<sb0__>
and where do I find this rustc debian package?
<whitequark>
1) absolutely no idea, I've spent several hours googling different variations on this
<whitequark>
there are zero results
<whitequark>
maybe it's something they broke relatively recently, there was one another bug in the xen debian package
<whitequark>
2) it's in unstable. it shouldn't drag any more crap than mere its own dependencies
<sb0__>
whitequark: and cargo?
<whitequark>
same place
<sb0__>
ah, yes
<sb0__>
apt-get detects modifications to sources.list without apt-get update now...
<whitequark>
oh it does? cool
<sb0__>
i had commented out the sid repository, and it did not find cargo
<whitequark>
sb0__: makes no sense... those pass locally
sb0__ has quit [Quit: Page closed]
fengling has quit [Quit: WeeChat 1.2]
sb0___ has joined #m-labs
<sb0___>
whitequark: could it be that the buildbot tester imports some stale old version of artiq laying around somewhere on the machine? that could explain both test failures
<whitequark>
that's not out of question as I changed that code but shouldn't have happened...
sb0___ has quit [Ping timeout: 252 seconds]
<whitequark>
ok, I will fix that a bit later
evilspirit has quit [Ping timeout: 256 seconds]
evilspirit has joined #m-labs
sb0__ has joined #m-labs
<sb0__>
whitequark: what would you recommend for a remote shell on windows, with scp capability?
<whitequark>
sb0__: can you plesae pull rust-vnc and rebuild? I switched the default to request a non-exclusive VNC session, this way QEMU won't kick either of us when the other one connects
<whitequark>
I'm astounded that it actually does what it's supposed to for once
<GitHub196>
[artiq] sbourdeauducq pushed 1 new commit to master: https://git.io/vzHze
<GitHub196>
artiq/master 5076c85 Sebastien Bourdeauducq: worker: Windows VMs are slow, increase send_timeout
<whitequark>
sb0__: log in to buildserver, then sudo -u buildbot -s; cd ~/slaves/debian-stretch-amd64-3/artiq/build; export PATH=~/slaves/debian-stretch-amd64-3/miniconda/bin:$PATH; python -m unittest artiq.test.pipe_ipc.IPCCase
<whitequark>
er
<whitequark>
also source activate buildbot-artiq-179
<whitequark>
sb0__: sebastien.
<whitequark>
you are using asyncio.
<whitequark>
did you add asyncio to conda deps?
<sb0__>
no, it's part of the python standard library
<whitequark>
oh
<whitequark>
right
<whitequark>
why the hell cannot it import pipe_ipc then?
<whitequark>
the file is there, I verified it
<rjo>
whitequark: did you compare the -1 builder with the -2 and -3?
<whitequark>
no
<whitequark>
that doesn't matter, because it will not lead to any insight anyway, I can do that import from command line
<whitequark>
er, python rEPL
<whitequark>
so, "from artiq.protocols import pipe_ipc" fails from a test, but succeeds from REPL
<whitequark>
this is idiotic
<whitequark>
even further, if I go up one level and run REPL, then I can import pc_rpc but not pipe_ipc from artiq.protocols
cyrozap-ZNC has quit [Quit: Client quit]
cyrozap has joined #m-labs
<whitequark>
ohh, I figured it out
<whitequark>
sigh
<whitequark>
bb-m-labs: force build
<bb-m-labs>
you must provide a Builder, try 'force build [--branch=BRANCH] [--revision=REVISION] [--props=PROP1=VAL1,PROP2=VAL2...] <WHICH> <REASON>'
<whitequark>
bb-m-labs: force build artiq
<bb-m-labs>
build #182 forced
<bb-m-labs>
I'll give a shout when the build finishes
<whitequark>
in a nutshell, I had to copy the git checkout to miniconda/conda-bld/work.
<whitequark>
i don't know why that worked. i don't know why that didn't work before. i don't know why sometimes conda-build works and sometimes it needs this hack.