well, i noticed earlier but thought it was an unrelated bug that had been fixed

then it turned up again

kinda bad

basically it chugs along at 100% CPU until you manually kill that kernel

the good news is that it isn't the FE itself

it is the 'auxiliary' kernel, which is what Evaluator -> "System" refers to (if you've ever seen that, "Local" is your ordinary kernel)

the FE keeps it around as a kind of pet

so you can actually safely kill that guy

so if you're willing to get your hands dirty with top and kill -9

you can recover from it when it happens

that's hardly a satisfying answer, maybe we can do some kind of emergency patch, i don't know

i'm having immense difficulty triggering it by hand, though

it seems completely random

@TaliesinBeynon How should I kill it

with kill

so top -o CPU? is that the flag?

i can't remember

yeah, top -o CPU, read off the PID of that MathKernel that's at 100%, then kill -9 <pid>

@TaliesinBeynon just reading this conversation you are having. Will this problem also affect player pro/CDF users and if so how will they deal with it?

@TaliesinBeynon When I kill it the frontend dies

@MikeHoneychurch i do not know. i do not know the source of the bug. i don't know how to trigger it. it might depend on some particular confluence of circumstances on a particular platform, and those might not be triggered by CDF, or might be.

Let me get a screenshot

@xslittlegrass what was the actual name of the process at 100%?

so... let's have some champaign to celebrate the shipping of 10.1, warts and all!

@TaliesinBeynon The name is "Mathematica"

@xslittlegrass i think top must be grouping by process parent

@xslittlegrass click View > All Processes, Hierarchically

you should see that it's only a particular MathKernel

@TaliesinBeynon it looks like the kernels are OK

@xslittlegrass well in that case it might be a totally different bug from the one i experienced

double click Mathematica

i mean in activity monitor

open files and ports

make sure there's nothing sensitive there

and you can paste that into a pastebin

you can also click Sample

and then paste the call stack

that might give some hints

pastebin.com

OK, I'll do it now. Thanks a lot for your help.

no probs

pastebin.com/qBEQS5Eh

@xslittlegrass looks fine.. and the "sample"?

@TaliesinBeynon here is the sample

pastebin.com/XKKkESRD

one last thing

try "show times"

in the sample window... and paste that. that should be more helpful in seeing what's actually slow

I'm sorry I don't see the "show time"

in the sample window where I copied the sample?

uh

under sample winodw

it says display:

change to percent of parent

OK find it

It looks like it doesn't allow me to copy

screenshot?

the top thread is the one to concentrate on

look down the tree

you should see it terminate in some system calls

well

i'm assuming

I'm taking the screenshot

dropbox.com/s/h2i6ueahxfytkyl/…

apparently dropbox can autoshare screenshots

which is nice

anyway

does it look like that?

yes, here are the screenshot

hmmm

can't say i have any clear ideas

i can give you some simple command line invocations that can tell if Mathematica is writing to disk, or making lots of failed system calls

should we try that? or do you want to give up?

i don't mean give up in a prejudicial way

I'm willing to try if you have time.

i have to go do something for 5 minutes

or 10 minutes

but i'll be back

okay with you?

yes sure

take you time

ok

try sudo errinfo -c, wait a few seconds, then ctrl+c

ok

pastebin.com/HX0Trh1N

ok

thanks

ok

so i think those psynch_cvwaits are a red herring

because 10.0 also produces them when NOT consuming 100%

OK

i give up

i don't know how to go further

i'm not an expert in reverse engineering

i suggest you take the materials you've recorded and include them in an email to tech support

best of all is if you can describe how to reproduce this reliably

OK, I will try tech support.

Thanks a lot for the help

I will use 10.0.2 for now. Thanks again for the time!

ok

@TaliesinBeynon I don't know what I did, but now it's the process named WolframKernel that takes 100% CPU.

@xslittlegrass heh... i think your bug and my bug may be the same

@xslittlegrass in any case... i'm actual now trying to match up the dissassembled x86 code from the stack trace with the Mathematica source code, which of course i have access to

@TaliesinBeynon Looks complicated for me. I'm leaving now, and I'll contact tech support later. Just ping me if you need anything from my end :)

@xslittlegrass sure... thanks

Does anyone else have problems with the Firefox "Wolfram Mathematica" plugin after installing Mathematica 10.1.0?

version 10.1, something happened where M crashed, now my notebook size is 226 MB (before it was 0.1 MB) and each time I double click on the notebook to open it, M hangs and not responding. I am stuck. No idea what to do.

may be if I wait more it will come back

time for coeffee

Strange, my two answers for the spelunking question were downvoted yesterday, this one and the other one based on my code formatter. They might not be perfect answers, but I don't see any good reason for downvotes. May be I just got used to not being downvoted :) - I think there wasn't a single downvote on my stuff for more than a year or even two.

@halirutan your great SE-Editor-Buttons is broken or is it just me?

@LeonidShifrin It happens all the time, I wouldn't worry about it. I don't even look at my rep score anymore.

@Szabolcs As I said, this didn't happen to me not a single time for about two years straight - and this is the only reason why I noticed it, not the rep per se.

@Szabolcs One review on glassdoor.com suggested that WRI did not have a clue about unit tests. Might be an exaggeration by an disgruntled ex-employee, though.

Although I encountered many examples in the documentation that didn't work as advertised. One would expect that those examples would be part of unit testing. Apparently not.

Talking about documentation: mathematica crashed on me two times today while I was browsing through the documentation of the new features of 10.1. Wordcloud was one page that I remember that did this.

my 3 links Robot :

@SjoerdC.deVries I have the same feeling. The big question is why, because I know people at WRI that have a deep understanding of computer science and software engineering. I know e.g. Leonid personally and I know that he has thought very carefully about the things he speaks. When we met we talked all night long and when you look at the code of the formatter you see that he is interested in writing code that is maintainable.

We all know e.g. Daniel Lichtblau and John Fultz and I'm pretty sure they are very good developers. Therefore, one and the only reason I can think of is that it is not forced or even supported by the CEO. This is supported by the fact that I sometimes encounter CSV comments in code while git or mercurial has proven to be much better.

@chris Let me see

@chris It works just fine for me.. can you be more specific?