<sb0>
we're probably arriving late afternoon, will you still be there?
<whitequark>
define "late afternoon"
<sb0>
5-6
<whitequark>
yeah, probably
<whitequark>
unless I run out of refrigerant or something; I finally finished the chiller and starting to fill it
<sb0>
ah, so you found a refrigerant shop eventually
<whitequark>
yes, I went to Mong Kok yesterday
<whitequark>
I had to walk for like a hour and eventually I found just one run-down car AC shop near Yau Ma Tei
<whitequark>
they had what I needed by sheer chance; I bought his last nuts and 1/4" Schrader valves
<whitequark>
that was surprising, car AC shops don't usually have the sizes I need, cars use different fittings entirely
<whitequark>
so I still need to find a proper AC shop
<whitequark>
I'm vacuuming the system now. the hoses are thin, the oil is extremely hygroscopic and I badly soldered one joint
<whitequark>
so it has a lot of moisture, this can take maybe a day to do properly, I fear...
<sb0>
what oil?
<whitequark>
the ether based oil inside the compressor
<whitequark>
(it ships filled with oil already)
<sb0>
heat it?
<whitequark>
how?
<sb0>
with the heating tape?
<whitequark>
hrm
<sb0>
on low setting
<whitequark>
it doesn't have settings.
<sb0>
use the variac
<whitequark>
well, I guess we have a variac.
klickverbot has joined #m-labs
<whitequark>
I'm not sure it's a good idea because the oil hydrolyzes
<whitequark>
which is the reason why you need to remove water from it
<whitequark>
like, if I don't, and I turn on the compressor, the refrigerant-oil mix will get heated to 90°C and hydrolyze into organic acids
<whitequark>
that is not too bad by itself but they are basically goo, that clogs various orifices
<whitequark>
the filter-drier, which is ridiculously oversized for this system (for this very reason), will catch that goo, but I'd rather not exhaust its limited resource
<whitequark>
moisture will freeze inside the metering device as well, stopping the refrigerant from circulating
<sb0>
did you put the diffusion pump oil into your small AC pump btw?
<sb0>
that worked?
klickverbot has quit [Ping timeout: 276 seconds]
<whitequark>
yes, I did, it's not like I have a choice
<whitequark>
it works, sure, rotary vane pumps can run on pretty much anything
<whitequark>
I'm running it for a few hours already and it seems pretty normal
<whitequark>
except the sound that it makes while running at 100 Pa is less like "quiet growling" and more like "intense dog diarrhea"
<sb0>
=]
<sb0>
what did you use to connect to the IMG last time?
<sb0>
some cannibalized ethernet cable?
<sb0>
whitequark, we have female RJ45 connectors too (without integrated magnetics), if that helps
<whitequark>
yes, and it works very badly, the ghetto circuit i have here is very unreliable and noisy
<whitequark>
hence I want to stop doing that
<whitequark>
hm, I could simply leave the pump for a day or two
<sb0>
if that a cable problem?
<sb0>
*is
<whitequark>
I checked that these pumps do not care about such treatment back in RU
<whitequark>
yes, that's a cable problem
<whitequark>
the cable I have here is quite low-quality
<sb0>
if you leave the pump on, please keep it away from anything flammable
<whitequark>
well, maybe not low-quality, but it has this multistrand crap
<whitequark>
whereas the cable I ordered has solid core wires as far as I know
<whitequark>
since we don't have any pinhead connectors, I cannot easily connect that cable to breadboard and it gets torn out etc
<whitequark>
and also torn apart because the wires are so thin
<whitequark>
if you have any pinheads I guess we can do a little less crappy setup
<whitequark>
re flammable: sure
<whitequark>
it won't heat higher than ~85°C though
<sb0>
is there at least one turbopump that is complete and supposed to work?
<sb0>
how does one get windows ReadFile() behave like unix read(), i.e. complete when there is something to read, not when the buffer is full?
<sb0>
_florent_, any idea?
<sb0>
this is the only thing that breaks flterm right now
<sb0>
do I need to set a timeout or what
klickverbot has joined #m-labs
<sb0>
ah, yes, that behavior I want is configured by SetCommTimeouts()
klickverbot has quit [Ping timeout: 240 seconds]
<sb0>
asyncserial integrates very well with asyncio. both are full of gross hacks.
<_florent_>
sb0: sorry, don't know
<rjo>
sb0: for HasEnvironment.set_thumbnail() and output_prefix(), should i funnel those through DeviceManager?
<sb0>
device?
<sb0>
should be datasets, no?
<rjo>
yes
<sb0>
set_thumbnail yes
<sb0>
what does output_prefix do?
<rjo>
i guarantee that people will want to output files by themselves that are not datasets or thumbnails in the DFH5
<rjo>
HDF5
<sb0>
it should be called get_output_prefix() then?
<sb0>
and I don't like this function
<rjo>
i love it
<sb0>
everything in the dataset manager deals with key/values. never with files.
<sb0>
it's ugly
<rjo>
that is why i am asking my original question.
<rjo>
the thumbnail is not a dataset either.
<sb0>
it is, it's a local one
<rjo>
i would keep thumbnail at the same level as artiq_version and repo_version
<sb0>
I haven't heard anyone requesting arbitrary filesystem output from experiments
<rjo>
it is in the SOW
<rjo>
what do i get if someone asks for it?
<sb0>
what about setting the current directory to the final results directory?
<rjo>
you still want the rid-name prefix.
<rjo>
but yes
<rjo>
that sounds fair
<rjo>
but since it doesn't give you the rid-name part of the prefix, does it help?
<sb0>
you can get your RID and format that yourself
<sb0>
_florent_, flterm works on windows now, afaict
<sb0>
can you give it a test on your machine?
<sb0>
rjo, maybe the format function can be part of the library
klickverbot has joined #m-labs
<sb0>
or well. that get_output_prefix() function you want could be at the same level as the experiment info (RID etc.)
<rjo>
sb0: which library?
<rjo>
how does an experiment get its rid?
<sb0>
hm, actually the RID API is self.setattr_device("scheduler") and then self.scheduler.rid, so it doesn't really fit
klickverbot has quit [Ping timeout: 244 seconds]
<sb0>
also, what about execution in artiq_run, that doesn't produce files?
<sb0>
I actually like in-memory runs
<rjo>
you can't keep an experiment from writing files.
klickverbot has joined #m-labs
<sb0>
no, but it should be discouraged
<sb0>
there is no more result structure if they do
<rjo>
discourage it by raising or returning None from get_output_prefix()
<rjo>
artiq_run breaks a lot of things already.
<sb0>
does it?
klickverbot has quit [Ping timeout: 248 seconds]
<rjo>
artiq_run has no results structure. doesn't save what it executed with the hdf5 etc etc.
<sb0>
artiq_run -o file.h5 should save the same file as if it were run on the master
<sb0>
and yes, it doesn't use the results folders, which makes sense as it's meant as an isolated run
<rjo>
thus breaking result structure.
<rjo>
but anyway. that's not my point
<sb0>
the result folder, and the <rid>-<experiment name>.h5 format, is introduced by the master
<sb0>
artiq_run is meant to be independent from the master
<rjo>
we can make that option --hdf5 be --output and thus the prefix for both the hdf5 and whatever people want to save
FabM has quit [Quit: ChatZilla 0.9.92 [Firefox 45.0.1/20160315153207]]
<sb0>
i'd just make the current folder the final results folder, which in the case of artiq_run means doing nothing, and have people format filenames themselves if they really want to create files
<sb0>
if you insist, the <rid>-<experiment name> format could be called the "human readable RID" (or similar) and made accessible from the scheduler
<rjo>
sounds good.
<rjo>
that obj["wd"] in worker_impl is the checked out repository prefix, not the working directory anymore then.
<sb0>
yes
<sb0>
hm
<sb0>
if you change the working directory, that may break local imports
<rjo>
would change just before build()
<sb0>
and messing with sys.path breaks local imports of subprocesses, as we have seen with this bug they made in unittest
<sb0>
ok
<rjo>
well just before making the experiment. that leeps all but lazy local imports intact.
fengling has joined #m-labs
fengling has quit [Ping timeout: 240 seconds]
klickverbot has joined #m-labs
klickverbot has quit [Ping timeout: 246 seconds]
mumptai has joined #m-labs
_rht has quit [Quit: Connection closed for inactivity]
<_florent_>
sb0: just tested flterm on windows, it works fine, thanks!