Silicon Valley Redneck

Another irritation

The C / UN*X convention for specifying directories to be searched for header files, libraries, and so on.

There's no syntax for "this directory and all its subdirectories" in this world.

DG's AOS did it right, if memory serves, back in the days of 16-bit minicomputers.  There was a wildcard for "all subdirectories" - from fading memory (confirmed by a PDF of the manual), it was the gridlet (#).  Meaning was somewhat context-dependent, but I seem to recall that, when specifying a list of files, foo:bar:# would get all files under foo:bar, and when specifying a search path, it would get all subdirectories of foo:bar.

(This is a vague recollection; I haven't touched AOS since the 1990s, and don't seem to have brought the manuals home with me.)

I seem to recall that Knuth's TeX also has some convention for this sort of thing, and a library to implement it.

In the mess I'm staring at just now, such a convention wouldn't be all that helpful, owing to the unfriendly organization of the source tree, but I keep coming across situations where there's a huge list of -I$PROJECT/include -I$PROJECT/include/sys -I$PROJECT/include/drivers and so on.

I also have yet another rant on IDEs coming on, but that's for another day.

Afterthought: AOS got a lot of things amazingly right, for running on such limited hardware.  The CLI didn't have all the bells and whistles of, say, bash, but it was really easy to do some common things that remain awkward under *NIX-style shells.  Maybe (IMCFT) I should write a sort of hybrid CLI, with the AOS-style nice features massaged to fit the *NIX environment, and of course all the really-commonly-used shell features included.  Not happening this year, though.

...And maybe I should try implementing an Eclipse in an FPGA?  A Nova 1200 would be trivially easy; the Eclipse architecture involved sequencing and microcode and such, but still fit in a 15" square board with 1970s technology, so cramming the whole thing including a couple of I/O controllers into a Spartan 6 ought to work.  But I don't think I have any AOS backup tapes lying around (and I certainly don't have a 9-track, 1600 BPI tape drive to read them on anyway), so maybe there's not much point.

Dinosaur experience required

Just looking at Someone Else's Code... for another RTOS, as it happens.  It's got a couple of abstraction layers, sort of (though nothing along the lines of chardevs and blockdevs).

There's a function to disable interrupts globally, and one to enable them.  They're called in pairs, naturally.

Anyone else seeing the problem yet?

Yup.  On some architectures - 68000 and Data General come to mind - there are multiple levels of interrupt masking, and you don't know, when you write the code, what the interrupt-enable state will be on entry to your function.

Actually, even on the architecture at hand, and granting that the RTOS may only manage two enable states, you still don't know, at the time you write a function, that it won't be called in an interrupts-disabled context... in which case, the disable call at the beginning will have no effect, and the enable call later on will leave the calling code vulnerable.

Guys?  The interrupt-disable function should be returning a magic cookie, which can be passed to the interrupt-restore function later to put things back as they were.

Don't go making dangerous assumptions, and definitely don't go having unexpected side effects (such as unmasking interrupts) in low-level functions that everybody will be using, perhaps indirectly.

The more I look at other people's RTOSes, the more I feel the need to take the time to clean up AGROS and port it to Cortex M, MIPS, and suchlike architectures.  (The architecture porting should actually be pretty easy, but then somebody's gotta write a bunch of device drivers for each chip family.  Hey, it's meant to be open source!  Um, also the config program maybe needs to be re-done in some language other than Ruby, which implies re-doing the architecture/chip config files, too, since they're basically written in Ruby (application config is XML, so not tied to the language).  They might end up in ECMAscript or something.  Or maybe some kind soul will make the Qt bindings for Ruby work with Qt5 and the current version of Ruby...?)

Another thing: Counting semaphores?  Should be able to count higher than 255.  And the limit should be configurable per semaphore.  Mine have 24-bit count and limit fields, an 8-bit flags field, and a lock byte (plus pointers to the first waiting task and to an optional name).  Because some wise guy might just be using a counting semaphore to manage a FIFO or similar, and sometimes those want more than 255 byte capacity.

Getting into the holiday spirit

Yesterday afternoon, Huckleberry was very insistent that Something Under The Sofa Was Drooling.

This morning, he's scampering hither and yon, searching out wee ghosties and ghoulies all around the house.

I hope he's just excited about Halloween, and will be back to normal tomorrow.

I wouldn't want to think that he's been reading the Monster Hunter books, and his young mind is taking them all seriously....

Oh, look: a recording!

Costume idea

Inspired by BC earlier this week, with a Santa suit not being a good choice for a Halloween costume:

Isn't it just about Easter in Australia now?  So how about dressing seasonally as an Easter Bunny with corks dangling from its hat?

Oh, wait.  The Easter Bunny doesn't go to Australia anymore, does he?  Something about Myxomatosis?

So what do they have, an Easter Kangaroo?  Hopping around with a basket of eggs?

Not that I'm likely to assemble such a costume in the next couple of hours (this being my dress-up day).  I won't even try to dress as Teddy Roosevelt (though I really should do that one of these years).  Probably just dig out the khakis and pith helmet and dress as a generic imperialist.

Wondering how this will turn out.

Gradually making my way through Monster Hunter Alpha.  (Almost typed Alpo there.)

Consider the scenario: small town with a dire problem, on top of which it's been magically cut off from electricity and any form of electronic communication.  Various good(-ish) guys are trying to get out, to pass a distress message to the next town.

Um.  Ya think disconnecting an entire town from the power grid, phone network, cellular network, cable-TV network, dish network, and so on, might just set off a few alarms?  The absence of communication / power consumption is itself a message, albeit not a very detailed one.

I find myself wondering whether, as the story finishes up, there'll be a horde of utility-company maintenance vehicles trying to get in.

 

Update: Apparently disconnecting an entire town didn't set off any alarms in utility-company offices.  Curious.

Also: It's firmly established that if you're bitten by a werewolf, you become a werewolf.  Apparently, though, if you get all lacerated and then a werewolf comes up and licks your face, you don't get infected that way.  Does this make any sense?  (Or is this just a case of not being able to believe that seventh impossible thing before breakfast?)

 

And this, children, is why Project Wainwright, back in the 1950s, sought to put a thermonuclear demolition charge in the basement of every city hall in the country.  The cover story about pathogens from outer space never did make much sense, even if Michael Crichton did sell a pile of books with it later.

"Question"

Congressional Republicans, Democrats and the White House are about to strike a compromise that would raise the debt limit, prevent a government shutdown and calm financial markets that fear of — another Washington-created crisis is on the way. Does your opposition to it show that you’re not the kind of problem-solver American voters want?

 Is not a question.  It's a little speech with a question mark tacked on at the end.  Apart from several statements, it carries a large load of embedded assumptions.

This is a typical tactic for advocacy "journalists", and for that matter for politicians.

It calls for a lengthy response, allowing time to research and compose it, because this sort of thing is never a simple question, and the answer needs to be much longer and carefully thought out, in order to expose and challenge the assertions and assumptions.

This is closely related to the tactic of playing "how many lies can we fit into one sentence", in that the response is always much longer and more boring than the snappy and sensational assertion.  See also: "A lie will go round the world while truth is pulling its boots on."

THINK

Some ideas sound totally cool at first.  Like, e.g., this reclining desk.

But...

Yeah, I've done the reclining-with-keyboard thing.  Tilt back in my chair, keyboard on lap, feet on desk.

Tilting the desk surface, though... consider for a moment why there's a desk there at all, and not just the keyboard and some sort of mouse arena.

Right.  Tilt the desk, and suddenly you've got a lap full of schematics, datasheets, business cards, backup disks, notes to call people back, reading glasses, telephone, snacks, pencils, and possibly cat.

That someone goes so far as to promote a reclining desk as anything more than a goofy concept makes me ask: what color is the ceiling in his office?

Huh. Could sworn this used to exist.

Programming socket adapter for a 10-pad, 3x3mm DFN package, suitable for out-of-circuit programming of ATtiny13A parts (in the itty-bitty package).

I'm not finding it.  Someone's offering me a socket, for a modest $155.  Ugh.

Looks like Microchip makes it easy-ish and not too expensive to program their little MCUs, in the 8-pad, 2x3mm package*.  Which also takes up slightly less board space.  And costs a few cents less.  (Also is much less capable, but it's a trivial application.  Basically a glorified button-debouncer, with a couple of extra outputs.)

Maybe I need to learn to program PICs again.  I guess it's gotten easier since... 1995?  Was that when I last used a PIC?

* Curiously, I'm not finding adapters for the 3x3mm version, in which more MCU models are available.  And the ones for the 2x3mm aren't exactly cheap, but a $60 board beats a $155 bare socket, I guess.

Minor technical note

While it's entirely possible to design a living hinge in ABS...

...Attempting to fabricate it using an FDM printer will not give good results.

(No, I wasn't fool enough to try this.  I did, however, design some rather thin walls, which turned out to be excessively fragile when FDM'ed.)

Oh, this makes all kinds of sense.

I heard this morning (from an informal source, out on the trail, and it's too juicy to verify) that our glorious bought-and-paid-for corporatist city council is proposing a $1000 license fee for therapeutic-massage businesses.

Yeah, a flat amount.  Regardless whether the business is one part-timer working at home or a real business with a storefront and a dozen workers.

Think the individual part-timers are going to stay in business?

Ah, but there's a good and moral reason for this.  It's meant to crack down on the scourge of prostitution.

So... may we assume that any individual, work-at-home part-timer who can come up with the license fee must be selling those extra services?  The existence of the fee will put the non-prostitutes out of business, and remove all ambiguity...?

If we're really determined to crack down on prostitution, maybe we could try a different approach: supply interdiction.  Drugstores already have a system in place to ration pseudoephedrine, thereby putting all the meth labs out of business.  Why not expand it to ration condoms?  Move the condoms back behind the counter, require buyers to show ID and sign the form, and limit purchases to some reasonable number per week.

I'm sure this would be entirely beneficial, with no possible unintended effects.

Mechanical stuff. Phooey.

Got a project that involves a plastic bit of rather complex shape.  I'm designing it using VariCAD.

Once I have an approximation of the design, I make a test article using the hobby-grade 3D printer.  Usual procedure is to export the part as a .stl file, generate G-code with Slic3r (having given up on both Skeinforge and the version of Slic3r embedded in ReplicatorG, as the former generates dreadful G-code and, if memory serves, the latter doesn't work at all), then use ReplicatorG to send it to the printer.

That was working for this part until this afternoon.  I made a fiddly little change - which, owing to the peculiarities of VariCAD's constructive solid geometry engine, involved disassembling a large part of it, changing a few dimensions, moving a few things around, and then reassembling it - and suddenly Slic3r chokes on the .stl (a Perl crash: "Attempt to free unreferenced scalar").

Well, VariCAD has been known to emit some nasty .stl from time to time.  So I try importing it into FreeCAD to clean it up.  FreeCAD's idea of cleaning it up seems to involve reducing it to shredded ribbons.  Oog.

OK, so save as a STEP file, import that into FreeCAD, generate a mesh, and... well, that mesh doesn't clean up much better, and Slic3r still chokes on it.

So now I'm using MakerWare to print.  At least it didn't choke, and the printer is busy making noises (and building up a plastic thingy).  We shall see how the output looks once it's done.

And this is for a live client... but of course all this futzin' around time ain't rightly billable.  I really am most vexed.

Update: There's a new stable release of Slic3r.  It doesn't run on my system, because I haven't done the dist-upgrade yet and the libraries are a couple of years out of date.  So I grab the latest source code, and build, and... it woiks!  Or, at least, it doesn't choke on the file at hand.  Haven't checked the quality of the output, because it's past Tinga's bedtime so the lab is closed for the night.  A print made using Skeinforge (while I was looking up and updating Slic3r) is of characteristically crappy quality, but is good enough to see that the recently-changed dimensions are right, so there's that.

Update 2: The latest Slic3r does a much better job on the raft, and on holes through the bottom of the part.  Surface quality of the bottom is still pretty rough, though; maybe I need to re-examine my settings.  The top surface is decent; unfortunately, the top is the inside of the assembly, and the bottom is the outside, where surface finish will be visible.

Isn't this a bit meta?

Spotted in someone's front yard this holiday-decorating season: a tombstone with a picture of a tombstone on it.

Er...

What would be buried there?

A dead tombstone?

An undead that's been re-killed?

...And with strange æons even death may die...?

Getting' Kinetis-ic

Yeah, I always get Kinetis mixed up with Kinect and K'Nex, but in fact it's a line of Cortex-M microcontrollers from Freescale.  I've been pondering some of those parts for a while, having a variety of cool ideas for good uses to which to put them.

Having had a bit of fiddling-around time the past week, and having bought the little eval board for the KL27Z (I already had the eval boards for sundry other models, but this one is of particular interest), and having installed KDS 3.0 and KSDK 1.3.0, I started messing around with the tools to try to get a sort of USB "hello world" demo app to run.  Eventually, I figured out (this is not well documented) which combination of platform libraries I needed to import into the Eclipse workspace to let me build the USB CDC demo app, with which running the board enumerates nicely as /dev/ttyACM1 (the debug interface appearing as ACM0).

So, I now have an example, more or less, of how to do it.

And what's so interesting about the KL27Z, specifically?  Well...

• It has the USB interface.
• It has the ROM bootloader, which should make user-installable, open-source firmware a reasonable thing.
• It's available in a selection of small packages, down to a 32-pad QFN.
• Power consumption is reasonably low.
• If I understand this correctly, it can be run (including USB operation) with no external clock components.
• It's cheap.  With 64K of flash, it costs less than an ATmega328.  With 256K, it's still around $4 in onesies.
• Oh, and it has the usual assortment of I/O that you'd expect on a modern MCU.

So, having confirmed that I can in fact build firmware for it that speaks virtual-TTY, it's maybe time to start creating nifty boards.  Got one ready to send out to fab, modulo final sanity check: a USB-to-thermocouple bridge.  Three chips (MCU, thermocouple interface, and an EEPROM on general principles), two connectors (mini USB, and a great bulky thermocouple jack that takes up 1/3 of the board), and a bunch of passives, many of which may turn out not to be needed.  If this turns into a product, it'll be radically cheap compared to any other thermocouple-to-computer interface I've encountered so far.

Got several other things sketched out, along somewhat similar lines.  Small, cheap test & measurement modules, mostly.

For those times that you wonder: why, exactly, does this manufacturing test station include a bunch of bulky, $1000-or-so, general-purpose benchtop instruments?

The Door Into Winter

Completely unrelated to Heinlein, though a cat is involved.

Oddity: "chicken tractor"

Why is a mini coop on wheels, so it can be dragged around, called that?

Surely the fundamental nature of a tractor is that it pulls things.  A chicken tractor, on the other hand, is a thing that can be pulled.

So shouldn't it be called a chicken trailer?  Or does no one want the eggs of chickens that live in a trailer?

cat foo.c > /dev/usb/lp0

Is not how printing is done these days.  Really.

Huckleberry, though, hasn't caught up with the times.  He's firmly convinced that one mustn't print without cat.

So when he hears the laser printer spooling up, he comes a-running and jumps on top of it.

The printer, not being designed to deal with obstructions and/or paperweights in the output tray (never mind being jumped on by 20 pounds of inquisitive kitten), promptly jams.

Maybe I need to show him a *NIX manual more recent than V7.

Or get a nice sturdy printer with maybe a top cover to keep cats out of the output tray.

Or put up a "This is a cat-free workplace" sign on my office.  Yeah, that'll do the trick.

Another one completely misses the point

From this essay:

Meyer makes a joke of drones, of free college tuition, and of a $15/wage. Do you think that people who fear for their lives in the tribal borderlands of Afghanistan, or who are being crushed by student loan debt, or who labor to provide for their families at a fast food restaurant would ever make similar jokes?

Um.  Ever consider that those low-paying fast-food jobs were never intended for people trying to provide for their families?  Those jobs are for low-skilled workers, who have the opportunity to earn a few bucks while picking up a bit of workplace discipline and thereby, perhaps, becoming better prepared for slightly higher-paying jobs.  Think: teenagers.  Like the ones behind the counter at the local Burger Clown.  The ones who work there for a year or so and then vanish, to be replaced by a new crop of teenagers.

deBoer, being himself completely isolated from this reality, is free to advocate sawing the bottom rung off the employment ladder.  And now what of the plight of the low-skilled worker, who can't deliver enough value to an employer to get any paying job at all?

He does make some good points elsewhere, but, as an academic, can afford to be comfortably disconnected from the economic reality of small businesses and the workers they might employ.

(And, thinking back to my previous post: ya wanna talk about under-represented groups in government?  Try: people with small-business experience.  Or, indeed, any sort of business experience, but those of us who spend most of our careers in small business know what it's like to live close to the cash flow.  The people who make the rules have no clue.)

Addendum: College-grad trust-fund brats can get high-prestige work experience with unpaid internships, which is the professional-class version of an entry-level job these days.  Poor kids who gotta earn something, even if it's not much?  They need some sort of paying jobs that fit the value of their labor.

Predetermined outcomes

Insty links to this story: a school holds a student election, and the principal decides that the results aren't diverse enough.

This reflects the fundamental problem with the notion that an elected government must statistically match the skin tone, genitalia,  accent, and religion of the population at large.  "Vote for the candidate of your choice, but the overall result of the election must match this set of pre-ordained* criteria."

Which is the same problem as, e.g., electric-car mandates.  "Every tenth customer who comes through the door must buy an electric runabout, even if he needs something to haul construction materials up dirt roads."  The mandate throws out customer preferences in favor of central-planner preferences.

Oh, and the whole women-in-tech thing.  We must force (a) more women to go into tech fields, whether they want to or not, and (b) employers to hire women for tech positions, regardless of availability of qualified candidates.  Yeah, that makes all kinds of sense.

And proportional representation throws out democracy in favor of enforced Balkanization.  Hello, institutional racism and sexism!  (And doesn't Lebanon have exactly this structural problem?  Not quite; Lebanon's problem is simpler - not as many types of division.)

 

* Spellcheck doesn't like this, and suggests "pee-ordained".  I Don't Want To Know.

Could be worse. Could be bagpipes.

Someone's dragging horse hair over cat gut in the park.

The resulting tones are sufficiently odd that I wasn't sure at first that he wasn't making blubbing sounds through a brass pipe instead.

The Empire strikes back?

Ol' Itty-Bitty Machines is going after the server business again, looking to rebuild its ancient empire.

With its Power Architecture.  Remember that?

Back in the 90s, I designed a board that included one of Motorola's early PowerPC chips.  'Twas a right Pain.

Then I spent a few years in telco space.  Boards with processors had PowerPC, either as processor chips or as cores embedded in expensive FPGAs.  The processor chips, as I recall, ran conspicuously hotter than ARM parts, for not much more performance (though at least they ran cooler than the x86 chips of the day).

Oh, well.  ARM has evolved a lot since the 90s.  So has x86.  MIPS has been making a comeback.  Maybe Power has improved too?