aserio changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar.cct.lsu.edu | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | Buildbot: http://rostam.cct.lsu.edu/ | Log: http://irclog.cct.lsu.edu/
bikineev has quit [Remote host closed the connection]
<aserio>
at the moment the philosophy is somewhere between MPI everywhere and MPI+X
<aserio>
though the political waters a still in flux
<aserio>
are*
<heller_>
We'll see then
<heller_>
How was your trip to New Mexico
<heller_>
?
bibek_desktop has joined #ste||ar
ajaivgeorge has joined #ste||ar
pree has quit [Quit: AaBbCc]
<heller_>
aserio: in the mean time, Europe is catching up ;)
aserio has quit [Ping timeout: 240 seconds]
<hkaiser>
heller_: that's brilliant - will force the issue over here
<heller_>
hkaiser: in other news: 106 and now full time writing
<hkaiser>
good
<heller_>
and you have a fun vacation!
aserio has joined #ste||ar
jaafar has quit [Ping timeout: 240 seconds]
<jbjnr>
{what}: parcel destination does not match locality which received the parcel ({0000000900000000, 0000000000000000}), ({0000000200000000, 0000000000000000}:({0000000200000000, 0000000000000000}:component_runtime_support[0]:0):call_shutdown_functions_action): HPX(invalid_status)
<jbjnr>
has anything in master changed recetnly that might cause errors like that? heller_ hkaiser ??
pree has joined #ste||ar
<hkaiser>
jbjnr: heller_ fixed a pp issue
<jbjnr>
that's why I'm asking. just wondering if this might be my fault - or a symptom of something else that got broken
<jbjnr>
(suspecting our stuff at the moment, but if anyone else haad this error, it would be interesting)
<heller_>
i haven't seen this error so far
<hkaiser>
jbjnr: this happens if a parcel is sent to the wrong loclaity
<heller_>
jbjnr: did you merge the latest master changes over?
<github>
[hpx] hkaiser force-pushed P0443R2 from 86b39fc to 4bb1c27: https://git.io/vQeHW
<hkaiser>
ABresting: he said he would get in contact, but I'll try to help if I can
<hkaiser>
ABresting: sorry, he's travelling a lot, currently
<ABresting>
I need your expertise in decision making
<hkaiser>
k
<ABresting>
yes wash is busy
<hkaiser>
let's see if I can help
bikineev has quit [Ping timeout: 255 seconds]
<ABresting>
starting with the project scope and timeline
<hkaiser>
k
<ABresting>
I wish to push major parts of the first evaluation in this week
<hkaiser>
great!
<ABresting>
as project has been stalled enough and this might become bad for me
<hkaiser>
what can I do?
EverYoung has quit [Ping timeout: 246 seconds]
<ABresting>
first major decision is the library(libsigsegv)
<ABresting>
I have understood the parts where stack overflow is being detected
<hkaiser>
I'd suggest to use it, as long as the code is separate from the main repo, that's fine
<ABresting>
yes but won;t it be great if we have our own implementation ? as libsigsegv contains something which we don't even need right now e.g. Handling the satckoverflow
<ABresting>
basic idea is just to detect the stackoverflow which wash has in mind
<hkaiser>
ABresting: if you can solve it without libsegev, even better - I was assuming that in the first step using it would be benfitial
jaafar has quit [Ping timeout: 240 seconds]
<hkaiser>
but I'm fine not using it ;)
<ABresting>
I ca solve it but up to some degree people will say it's derived from libsigsegv
<ABresting>
even I am more than willing to code it up
<hkaiser>
ok, let's do this, then - we'll cross the bridge when we're there
<hkaiser>
if the code you produce has no resemblance to libsegev, why should people complain?
<hkaiser>
everybody will be happy that there is a non-gpl'ed solution to this
<ABresting>
but libsigsegv's technique is the only way around for parallel environment
<hkaiser>
ABresting: it's the code which is gpl'ed, not the technique
<ABresting>
haha
<ABresting>
come'on
<hkaiser>
IANAL
<ABresting>
you been telling me this whole time
<ABresting>
that dont go near it
<hkaiser>
aserio: yt?
<hkaiser>
ABresting: if you don't look at it and create something on your own, who can blame you?
<ABresting>
hkaiser: like you said that technique is not gpl'ed, but wash beg to differ
<hkaiser>
k
<hkaiser>
he might know more than I do
<ABresting>
there is only one bridge i.e. address trace to the page
<hkaiser>
I don't know anything about this technique
<hkaiser>
ABresting: I think we should try to make small steps forward
<hkaiser>
if we use libsegev in the first steps, fine
<hkaiser>
if we find a way to replace it later, even better
<ABresting>
it's a pointer technique, it goes with the size of the stack. when this address gets a reading or writing query, it signal's and handled !
<heller_>
i would strongly suggest to use libsigsev in the first prototype
shoshijak has quit [Ping timeout: 258 seconds]
bikineev has joined #ste||ar
<hkaiser>
heller_: I agree
<heller_>
anything else is just holding you up
<ABresting>
how can we make it an external dependency? like you said include it in user side
<heller_>
do you want to code or worry about potential legal issues?
<ABresting>
I want to use first thing first upon which everyone agrees
<heller_>
we are no lawyers, it's your call. we can just give recommondations.
<ABresting>
wash: do you agree with using libsigsegv for the first prototype?
<hkaiser>
wash[m]: ^^
<ABresting>
heller_: trust me no one needs to deliver this more than I do ;)
david_pfander has quit [Ping timeout: 246 seconds]
denis_blank has joined #ste||ar
pree has quit [Ping timeout: 258 seconds]
david_pfander has joined #ste||ar
<heller_>
hkaiser: When writing the HPX chapter ... there are so many things popping up my mind that I want to get rid of ;)
<heller_>
http://goo.gl/4hxPef <--- vote for FAU - Friedrich-Alexander University Erlangen Nürnberg
bikineev has quit [Ping timeout: 255 seconds]
Matombo has joined #ste||ar
david_pfander has quit [Ping timeout: 240 seconds]
<aserio>
hkaiser: here now
jaafar has joined #ste||ar
<hkaiser>
aserio: hey
<hkaiser>
aserio: those guys are about to buy the wrong replacement drive
<jbjnr>
heller_: yes, we merged master in from yesterday - interestingly the errors come from the tcp parcelport and not the LF or MPI one, but we are using highly dodgy executors etc that might be riddled with bugs etc.
<zao>
I have from good sources that all problems in HPX are completely unrelated to jbjnr's innovations.
<zao>
jbjnr: Internet armchair lawyers say a lot of dumb stuff.
<zao>
If you're properly kosher, do a ReactOS/wine-style white-room thing where one party documents something, another party implements from that documentation.
<jbjnr>
yes. the white room approach is the only safe one
<zao>
Implementing the spirit of a library is rather from public knowledge, presentations and most documentation is "fine". Can't copyright ideas, etc.
<zao>
(IANAL, etc.)
<zao>
The "look at code and you're hosed" attitude is very defensive and theoretical.
<jbjnr>
but aspeople write - once you've read an algorithm and it was implemented in gpl code - does that mean you can never again use that algorithm in non gpl code? obviously horseshit.
<zao>
You also have the bits and bobs about how interface layers are less sensitive as they're about interoperation.
<jbjnr>
so if ABresting looks at the code, see's what it does and then writes his own version - as long as he passes the plagarism test ....
<ABresting>
jbjnr: Thanks, at least after looking at the library code I know what not to do in "my version". I am sure if I look more closely I won't need someone else's idea.
<zao>
And of course, anything you read authored by GNU affiliates and the EFF will be leaning toward bolstering the GNU position.
<zao>
So it's hard to find any sane and level discussion about the subject.
<zao>
Integration at the end user machine via code designed to interface with a more generic endpoint in the software definitely falls under sane usage.
<ABresting>
maybe that's why Richard Stallman doesn't like open source code ;)
<zao>
The key to understanding GNU motivation is that they don't care about the OSS movement in general. They care about GNU world domination to achieve their vision.
<zao>
Not going to rant about how they're actively squashing integration with properly free alternatives, as it'd erode the GNU position.
<zao>
(kind of did there, but hey...)
<ABresting>
i.e. Free Software(safe if possible - no monkey business)
<zao>
Commercial scientific code is way more fun. Got people at work who cannot ever publish or contribute to any other software packages in that particular field, due to having source access to particular codes.
<ABresting>
what are they supposed to do ? reinvent the wheel, and name it something else ;)
<zao>
When the licenses for the software is in six figures, you're sure as heck going to protect your IP :D
<ABresting>
now I wish it were "Free Source"!
aserio has quit [Ping timeout: 240 seconds]
<jbjnr>
well we're all agreed that the gpl is pants then :)
<jbjnr>
makes you wonder how it lasted so long
jaafar has joined #ste||ar
shoshijak has joined #ste||ar
<heller_>
ABresting: it gets very very tricky from time to time, yes. Convincing big companies to release their work as part of open source is hard, even though it shouldn't since it is in the contract they signed :P
<ABresting>
yes we can't have it both ways !
<heller_>
I think it was stallman who argued once that the beginning of the end was when you suddenly had to protect your data with passwords and shit
<heller_>
NB.: Hurd got released: now supporting 3GB of RAM
<ABresting>
heller_: I personally have talked to stallman and he is a bit extreme when it comes to transparency, but his faith of free and safe don't go hand in hand !
<ABresting>
and yeah licenses are shades of grey :P
<zao>
heller_: Tihi :)
<zao>
Do you fine people know of any decent tools to track memory use on clusters? Kind of wish I had something I could throw at users to get an an extrinsic profile of their use over time.
<heller_>
zao: darshan does this, IIRC
<heller_>
what about nagios?
<zao>
More towards I/O, isn't it?
<heller_>
right
<zao>
heller_: I can get some information about overall node load, but nothing that's accessible to users and we share nodes between jobs.
<heller_>
hmm
<heller_>
other than manually monitoring /proc then?
<zao>
Users don't have SSH access to nodes either, so they'd have to pervert their submit scripts with something I guess.
* K-ballo
needs to check whether the current existing meaning of "funject" is compatible with that of a function object
<heller_>
(I don't like it ;))
<zao>
I don't know nor care, I blame the idiots on slack :)
<zao>
invokable!
<K-ballo>
with a c
<heller_>
what's wrong with callable object?
<heller_>
or even functor, stupid pretentious people
<zao>
invoke-able leaks a lot of silly internal detail no human should care about, IMO.
<mcopik>
I never understood the hate towards functor. sure, it has a different meaning in math but it's not the only concept which is called by the same name in different branches of science
<K-ballo>
just use the proper name, is not that hard...
<zao>
Which is?
<zao>
Changes every single time someone talks, it seems.
<K-ballo>
depending on what you mean, could be function object or callable object or other
<K-ballo>
I'm currently reviewing some stuff where the author says he zero-initializes a variable via the assignment operator
<K-ballo>
he obviously means "int v = 0;"
<K-ballo>
none of the words he use means what he thinks they do
<K-ballo>
is so frustrating, why did he feel the need to write "zero-initialize", a word of power, rather than "zero initialize"
<mcopik>
we use the term "function" in all major programming languages even though it's not compatible with the mathematical definition of a function
<mcopik>
is there a difference between "zero initialize" and "zero--initialize"?
<K-ballo>
"zero-initialize" (single dash) is a standard term, the name of one of the many initialization mechanisms
<K-ballo>
"zero initialize" I suppose would be any initialization that sets the value to zero?
<K-ballo>
in particular, "int v = 0;" is not zero-initialization
<K-ballo>
and it isn't using the assignment operator either
<mcopik>
cctor?
<zao>
How many ways I hate the amount of pedantry you need to talk about C++.
<K-ballo>
if there was no other meaning for "zero-initialize" there would be no room for confusion
<zao>
Not all occurences of the = in expressions is assignment, indeed.
<K-ballo>
but if you are pretending to talk in standardeese then don't make stuff up
<zao>
mcopik: Any implicit unary ctor, IIRC.
<K-ballo>
reminds me of when vinnie wanted to change C++'s "strict aliasing rules" to support some proxy iterators in the networking ts
<zao>
C++ has strict aliasing?
aserio has joined #ste||ar
<K-ballo>
of course, he didn't mean "strict aliasing rules" at all, he meant something else he thought was related to aliasing, and being strict
<zao>
Thought that was just a bunch of people wishfully thinking that C things hold in C++.
<K-ballo>
uhm? didn't C took strict aliasing rules from C++?
<mcopik>
I think something else might have happened here. it's a common thing for non-native English speakers to put a dash in verbs consisting of two words
<zao>
Don't remember.
<zao>
Might be conflating with restrict.
<mcopik>
he might have not been aware about the existence of 'zero-initialize'
<K-ballo>
restrict is a world appart
<K-ballo>
he is aware, he is writing a technical book on this stuff.... I'm reviewing
<mcopik>
I see
<mcopik>
I just see myself making such error
<K-ballo>
but I get that if you don't know what "zero-initialize" means then you wouldn't understand the room for confusion, and thus be annoyed at the pedants correcting you
<mcopik>
in the context of language error
<K-ballo>
the safe sane path is to not make terms up
Matombo has quit [Remote host closed the connection]
jaafar has joined #ste||ar
Matombo has joined #ste||ar
aserio has quit [Ping timeout: 246 seconds]
aserio has joined #ste||ar
zbyerly__ has joined #ste||ar
zbyerly__ has quit [Remote host closed the connection]
zbyerly__ has joined #ste||ar
bikineev has quit [Remote host closed the connection]
Matombo has quit [Ping timeout: 240 seconds]
Matombo has joined #ste||ar
Matombo has quit [Remote host closed the connection]
pree has quit [Quit: AaBbCc]
Matombo has joined #ste||ar
Matombo has quit [Remote host closed the connection]
akheir has quit [Remote host closed the connection]
jaafar has quit [Ping timeout: 240 seconds]
<heller_>
K-ballo: but it is so easy to make stuff up ;)
<heller_>
K-ballo: see pm please
eschnett has quit [Ping timeout: 258 seconds]
<denis_blank>
Hi, my PR failed in CircleCI, while doing "$ docker run -v $PWD:/hpx -w /hpx ${IMAGE_NAME} ./build/bin/inspect --all --output=./build/hpx_inspect_report.html /hpx" (https://circleci.com/gh/STEllAR-GROUP/hpx/7041). Does anyone know, where I could retrieve the log of this step from?
<K-ballo>
you probably won't have permissions, let me see
<K-ballo>
you have to be logged with github, and then if you have permissions you see an "artifacts" tab at the top