damit... there's a tuning GUI for megasquirt..
hahaha megasquirt... my neightbour wants me to build one for him, to do his '82 sirroco.
I'm not gonna do it.
he can buy it, when they finish the damn thing.
megasquirt, fuel management system. open sourced. electronics, and software.
kinda like g2002 for car fuel management systems.
creative liscense, and all that.
is paul_c watching?
* paul__c is heading off to bed.
just wanted to mention something... there have been two bug reports in the last two days where the user has identified it as
but from all appearances it seems to be the bdi-4 version
Anders' is HEAD
just wondering about how to distinguish the two
What was the other bug report ?
I wonder what distro he's running under? (HEAD still doesn't compile under 2.6, a few nasty bugs to squash before merging)
"odd behavior in MDI mode"
using apt-get, so yes. That would be bdi-4 branch.
I was thinking about adding something to the tracker page asking folks to clearly identify the version they are talking about
and Rumley committed a fix for the bug to HEAD.
but I'm not sure the average newbie would even know which one
why doesn't that surprise me?
"fix" on HEAD, but bug reported on a branch.
need to move forward on the joining...
kbuild now compiles on all four platforms (bdi2.xx, tng, live, 4.20)
but there is a problem on 4.20 that can lead to lockups, no way I want to merge that into HEAD
A straight join on the common usr space code will cause major breakages across the board.
I know.. it's gonna be a mess
I'm hoping to find and fix the lockup bug in kbuild over the next couple days, then kbuild can merge with HEAD
I tried a BDI-4.21 build this afternoon.
building emc, or building a complete BDI CD?
HEAD doesn't have any kbuild stuff in it
sorry, kbuild branch.
compiled without any significant errors...
that's good to hear
but didn't run I bet
locked up solid as soon as I tried to run.
yeah, alex, martin, and I have all experienced that
I'm workin on it ;-)
keep forgetting to test it on my 2.4 box - I believe it is specific to 2.6, but I'm not sure yet
You are still using rtapi_get_time() despite it not being RT thread safe
you are referring to the version that calls do_gettimeofday?
anonimasu: Go to bed.
But I doubt that is the real cause of the lock.
* jmkasunich looks for CONFIG_X86_HAS_TSC
4.20 is compiled for i686, which certainly has a TSC, so it should be calling rt_get_cpu_time_ns()
most of the BDI-4.xx kernels don't have it enabled.
HAS_TSC is something that is separately configured when you build the kernel, or is it derived from the arch info (i386, i586, etc)?
perhaps the #else in that function should simply return 0 (at least if called from realtime code)
as you say, that probably isn't the cause of the lock, but it should be fixed anyway
on the BDI, if TSC is separately configured, can you turn it on in future builds... ns rather than us times are nice for testing
I don't think there is any code in EMC2 that actually uses the value from rtapi_get_time for anything other than logging and measurements, so a version that always returns zero should be OK
This was pointed out last year...
and I never got around to fixing it
Run through the sequence for running a series of periodic functions under HAL
03jmkasunich 07kbuild-0-1 * 10emc2/src/rtapi/rtai_rtapi.c: Replaced a call to do_gettimeofday() with a thread-safe dummy in rtapi_get_time(). If the kernel is compiled with support for the TSC, then a nano-second resolution function is used, otherwise the dummy is used.
yo mean the code that calls them?
thread_task is the main loop isn't it...
so where does that get scheduled ?
it is a realtime task
but where is the call to rtapi to set the scheduler up ?
* jmkasunich is looking
well, that creates it...
starting it as a periodic task is at 1212
I added some outb's to the "thread_task" loop to set a bit at the start of each thread, and clear it at the end, and did some scoping Tuesday night
OK... Now where are the "threads" added to the list ?
the threads seem to be completing in a few uS each (max 20 for servo, 14 for base)
shortly later... I know what you are thinking... thread runs before it's added
but the "new" structure is already filled out before the task is started
it may not be linked to the list, but it has everything that "thread_task()" needs to run the realtime stuff
no... I was thinking more subtle..
sorry - I shouldn't try to read minds
which function call is used to register an RT component func() ?
hmm, thats kinda dumb... (not crashy dumb, just wastefull dumb)... I grab the mutex before doing some validity checks
are you going down this path because you had a lockup as soon as you started to run?
because I don't think it's a problem in the setup code... Alex and I have experienced lockups ranging from a couple seconds into the run to many minutes
short periods for the base (fastest) thread result in sooner lockups
longer periods (like 100uS) run longer
but the vast majority of our lockups were clearly well after the thread/functions were already set up and no longer being changed
you might have just gotten "lucky" and had it fail in the first 100 milliseconds - indistinguishable from failing while setting things up
(not that I don't appreciate you looking over that code...)
I'm not looking for any mutex or thread timing issues.
none of the public functions in hal_lib.c are called after the setup is completed
When you were scoping out the timings, did the system lock up at any point ?
yeah, once... seemed to be in the base thread (that does step generation and parport)
but the servo loop continued to run ?
nope, everything was frozen at that point
base is the highest priority
so the stepper task was still running, but not the slower tasks ?
the bit that says "in the stepper task" was stuck on
as if the stepper task never finished up and yeilded the CPU
more likely, the stepper task started, and the system went tits up at that point.
that was the last thing I say Tuesday night (weds morning really)
couldn't work on it yesterday
today I'm gonna do more instrumenting of the base thread
I was thinking about adding an even higher priority and faster RT task, that would serve as a watchdog
if the base task locks up, the WD task could kill it and unfreeze the system... to save time during debugging
hmmmm... emc2kbuild is running chips3d and backplotting on my TNG box, with a compile going on at the same time to cause some disk and RAM thrashing
at 50uS base poeriod (which caused lockups in seconds on this box)
I wonder if there have been any RTAI api changes between RTAI 2.4.xx and RTAI 3?
I see one possible problem area in freqgen.c
You have a union in freqgen_t
st0_t & st2_t (argg... that _t gets to be....)
sorry bout that
Those two structs need to be packed..
03:10 .. time to drive to the airport ...
going to geneva for 10 days ...
I always use one or the other, never both (I'm not doing the "write it as an int, read it as chars" trick
so you are making the assumption that gcc will use the same amount of space regardless of which struct you want to use.
I reckon you need to pad the smaller of the two to be safe.
no I'm assuming that gcc will allocate enough space for the larger of the two structs
Is 50uSec the fastest you can get a thread to run at ?
that's just the default I stuck in the ini file... dates from when I was testing on a P233 and didn't want to have to worry about it
actual execution times for base thread were about 4.6uS avg, 12uS max, servo was 6.2uS avg, 20uS max
(on the BDI-4.20 box, 1.666G Sempron)
the one on my TNG box is still happily running.. the lockups are definitely unique to 4.xx (or rtai3, or kernel 2.6, etc)
yet the bdi-4 branch runs faultlessly - Throw in a null pointer, and it doesn't lock up.
probably a bug in rtapi or hal... neither of those has ever been tested on the new stuff
where "bug" might mean some API change that in the worst case could require rtai3_rtapi.c (but I hope not)
rtapi could be tested with the bdi-4 branch without too much effort.
I was also planning to do some standalone HAL tests (run say stepgen and siggen without the emc motion module)
but then rtapi is just wrappers around rtai/rtlinux calls..
motmod is by far the most complex part - if lockups happen even without it that gives it a clean bill of health
areas to suspect.... multiply or divides on long longs
or even long ints
long ints are the same as ints aren't they (both 32 bits)?
on that note.. one thing that I believe is different is the rt math lib
do a diff between vesuvio and stromboli
TNG build: configure: WARNING: Using glibc libm with mathstubs.
4.20 build: checking for kernel math support... ok, using RTAI's libm kernel module
(although in that case I would have expected the 4.20 to be more reliable in realtime)
I see math.h includes a pair of macros for converting ints to/from floats
Right.. Diff'd the current math lib with Stromboli
No significant differences.
are you saying the version of RTAI on BDI-TNG has a RT math lib that is similar/identical to the one on BDI-4.20?
if so, how come it's using glibc's libm?
double check TNG - It should have a math module.
and I'm off to bed.
I suspect that it has the lib, but configure isn't finding it
two items from harbor freight that may be of interest: 42184-6CAB AIR HOSE, 3/8" X 25', BLACK $3.49, and 42444-5UCB 5 PC. QUICK COUPLER BRASS SET $1.29
i got 10 of each :)
well, we'll see if they're back ordered or not
good day people
* anonimasu is installing solidedge
* Phydbleep is wondering if he wants to bother trying to do anything with this oddball acme thread or whether he should just go buy a piece of 1/2"-10 and some nuts.
It's a 3 start.. WTF were they smoking to make a 3 start?
It' a 5/8"-8 3 start even..
Has anyone used one of these or thought of it for the spindle in a pc board mill?
[09:29:47] <Phydbleep> http://www.harborfreight.com/cpi/ctaf/Displayitem.taf?itemnumber=47869
i ordered one of those micro air grinders not two hours ago
possibly for use in a pc board mill
Decided to live dangerously did you?
also got the trim router 44914
wish they made something that took 3/8" weldon shank endmills
I have 4 nice pieces of acme thread i was going to use to build a 'small' 18"x18"x12" milling unit..
i'm just gonna bitch about how 1/2"-10 is $1 a foot
Clean, no rust, no wear..
Unfortunately it's 5/8"-8, 3 start.
so that's effectively 5/8-2.66?
I'd be happy with $1/ft, It's the $5/nut that kills me. :\
bah make your own out of delrin
like you was telling Jymmm the other day
I want phoshor brionze or oilite.
or somebody was
hmm.. you might be able to cut threads into a bushing with a homemade tap
then again maybe not
how many nuts do you need?
I should be able to do it on the lathe.. I'll start the threads with a boring bar and finish them with the tap.
X-4, Y-2, Z-2. :\
Plus I can cheat and cast my own nuts with a former-core so the tap will start easily. :)
you can cast phosphor bronze?
I can cast brass and I can tweak the alloy by adding copper/zinc/tin/etc..
hmmm... i have a "speed start" acme tap that looks like it's about 5/8"
came with an ebay auction.. i'll measure it, one sec
fenn: It looks like you add 1%-8% phosphorus to the alloy to get phosphor bronze..
I wonder if I could get away with substituting TSP (Tri-Sodium Phosphate)?
But it'll be 20 years before I know if it worked. :)
you arent going to use your machine?
I was going to build it for severe abuse and then just use it. :)
well the results are in
fenn: The rabbit died?
it's a three start acme taper tap, major dia .642 minor .566 if my measurements are to be trusted
fenn: Sorry, Couldn't resist.. OLD joke. :)
i'm not that old i guess :/
anyway, i've no use for the damn thing, do you want it?
fenn: You have a 3 start 5/8" acme tap?
thread depth looks a little shallow
not that it matters in the end
What's the pitch.. Lay the ruler on top and count the threads in an inch. :)
ah good point
gotta peel off more wax
The threads will be better defined as you get toward the handle end.
exactly 3 tpi
so that's 1 pitch?
That's not a 3 start then..
there's definitely three threads on it
hm I found a place to shop for servos
[10:08:19] <anonimasu> http://www.allmotion.se/
fenn: Got a webcam/camera?
yeah i'll photo 'er up
anonimasu: I'm having rude thoughts about adding an encoder and real bearings to Chevy Lumia power seat motors. :)
brb, getting a package
They use a cable drive like a flex-dremel or a speedometer. :)
[10:14:57] <fenn> http://fenn.freeshell.org/camera/DCP_0333.JPG
That's just rude looking.. Throw a towel over it. :)
The threads on this line up perfectly with the 1/8" marks on the ruler. :\
oh they do do they
5/8"-3, 3 start would be 1" travel per rev.
fenn: les might be interested in that tap, He was talking about high speed/precision leadscrews.
damn i was hoing my other weirdo tap would be the right one
What's the other one?
6 start .725 major dia
that's 3 pitch right
er no 1.5 pitch
* fenn performs the heimlich maneuver on Phydbleep.
(Hell If I Know) :)
$1.45/ft for 1/2"-10 acme here in town... $193.00 for nuts (only sold in boxes of 100)
like anyone in their right mind would buy 100 acme nuts
Hmmm.. so I should buy 2 boxes?
nah just get some ball screws from techno isel, eh
hello fenn ;)
fenn: Well, Get over there with your credit card and order a set for me. :)
McMaster is out of their fscking gourd... $17.76 for a 1/2"-10 acme coupling nut with a 2g fit. :[
so, what's wrong with delrin again?
alex_joni is now known as aj_away
* anonimasu sighs
fenn: No delrin handy.. I can go out on the mesa with a sifter and a bucket and get brass for free.. $<10 in propane to cast lots of parts.
it costs too much for phyd
* Phydbleep checks his wallet and finds $3 and a moth that starved to death.
you shouldn't have pets if you can't take care of them
That wasn't a pet, it was a cat toy. :)
* anonimasu pokes solidedge
And look.. It's a cat toy again. :)
no thin wall feature in my version..
Hmmmm. Get a piece of acme thread to make a tap, use that to make a die, use that to make core formers for casting nut blanks to run the tap through. :)
* anonimasu sighs
seems like the cad investment went up quite a bit high..
sounds like there is a need for a decent GPL cad/cam program
anonimasu: Have you looked at Open-FX?
[10:50:28] <Phydbleep> http://www.openfx.org
i wonder if "kinematics based animation" means solid model simulation capability
fenn: Yes. :)
fenn: You can build a lathe or mill and move the bits around. :_
yeah but if i build an unbalanced mill will it tip over and fall?
I don't know how gravity/mass work in that.. I used it to model things for another simulator.
Phydbleep: is is good?
Phydbleep: or rather, does it match solidworks?
anonimasu: It's free and pretty easy to use.
* Phydbleep has never used solidworks.
Phydbleep: well, I dont care about "free" :)
so many 3d modelers, so few CAD programs..
The last semi-commercial 3-d package I used was milkshape.. Before that i was using MGI Extreme-3D
Phydbleep: it's for work.. so even if it ends up at a couple of $k's it'd not a big deal..
anonimasu: 44MEG DOWNLOAD FOR THE WHOLE PACKAGE INCLUDING EXAMPLES.
* Phydbleep beats the caps lock key into submission.
Phydbleep: does it match solidworks?
* Phydbleep has never used solidworks.
Phydbleep: every used a modern parametric cad package?
that looks like a 3d modeller not a cad package..
Done what? To Whom? for how many Oreo's?
anonimasu: I've never even used autocad. :)
Phydbleep: well openfx isnt a cad program
it doesn't help that you can't get a parametric 3d cad package for less than $500
if I could find one good under 2000$ it'd be nice..
* Phydbleep learned drafting way back when, when computers were things that filled a 2 car garage and wanted 440V/100A 3-phase..
anonimasu, what do you think of synergy?
fenn: it does not match my drawing style
We used to use this now obsolete material called 'paper' and me made these colored marks on it with 'pencils'.. :)
seems like if anyone ever did make a GPL cad/cam package, it should have the "engine" and then multiple GUI's to make everyone happy
fenn: but the cad package is good.. but well, I cant draw in autocad.. the way you work dosent suit me at all
anonimasu: I use the modelers because 99% of the time I can already see the part in 3-d in my head and I just need to show it to someone else. :)
yeah at least 3d modelers are based on 3d geometry and not 2d drafting stretched into 3d
And I can't run autocad (or any of the clones) to save my ass.. I think in 3+D and autocad is counterintuitive to me.
Phydbleep: I need somthing for making/drawing up parts in.. in a efficient way..
Phydbleep: 3d->3d drafting is a efficient way to work..
Hehehe... It's really funny, I can visualize an N dimensional matrix/manifold that takes a math major 16 hours of code to display on a crt..
hm.. but will that show the people building the mainfold the dimensions they need?
Actually yeah, I was able to look at the display and tell him that the shape/form was wrong and that he had typoed two of the major variables and one of the minors.
FreeCAD sounds like it will be pretty good if it ever gets off the ground
This guy just got his masters and has just been published in Nature as well.
I've been torturing this poor guy for years because I can look at the physical aspects of something and rattle off figures that are within +/-5% of the correct answer..
My last major act of sadism was to ask a question about the fundamental nature of magnets/magnetism that has 3 departments scratching their heads and plotting mayhem.
Phydbleep: well that dosent work when you build stuff..
"Is magnetism an intrinsic part of the material, Or is it an extrinsic force that is focused by the material?"
anonimasu: You'd be surprised how often an educated guess is close enough.. :)
* anonimasu sighs
* Phydbleep yawns..
Phydbleep: you would be surprised on how hard it is to build stuff without a good drawing..
anonimasu: Only if you're trying to get someone else to build it for you. :)
Phydbleep: I work at a place that builds stuff, so yes.
* Phydbleep remembers a 'lawn-mower' that do to a miscommunication ended up as a 'tree-mower'..
opencascade looks even more promising than anything else i've seen yet
1/4" steel for the blade, Not 1/2".. How is that little motor supposed to swing that.. 3hp off, 8hp on..weld in some braces because the cutter deck taco'd..
Pretty soon it was mowing trees better than the tractor and all I wanted to mow was the fscking grass..
Never give 3 redneck motorheads vague directions and expect them to fill in the blanks with common-sense. :)
phydbleep look at http://www.opencascade.org/showroom/demos/
to see what a good cad program looks like :)
i especially like the 2d constraints demo.. wonder if they have 3d constraints too