Silicon Valley Redneck

Meanwhile.

The financial world seems particularly volatile, not to mention down, so far this week.  Trump lunacy?  Anti-Trump lunacy?  Something else entirely, that's not being reported?  Chinese investors are liquidating everything and buying tickets for the Tesla Martian Express?  Some ill-behaved dog in Zurich has been going around piddling on gnomes?  An international counterfeiting ring has been circulating high-quality^* rolls of false Venezuelan 5-Bolivar notes?

Closer to home: that low-grade probably-virus crud I had for a while?  Joy's had a full-blown case of it for over two weeks now.  Gotta take a couple of hours away from my desk for a medical excursion, 'cause she's not fit to drive, and maybe it's not just the-flu-that's-going-around.

Update: No actual diagnosis yet, but a suspicion that there's an infection going on.  She's got some antibiotic pills that are supposed to be safe for her, and a followup appointment tomorrow.  Lunch got to be 1430-ish, which is a long time after my usual 0630 breakfast.  BY the time we got home, I had a nasty headache (and a cruddy little bit of fever).  Also, a test I'd started at 0655 turned out to have failed at 0738, so that's one proposed bug fix down the drain.  Got to do some serious bug-hunting when I don't have a headache and am not running errands, medical or otherwise.

 

* Extra-plush.  Quilted.

More evidence...

...that the Commerce Clause has been stood on its head:

“California must take action to avoid supporting or financing discrimination against lesbian, gay, bisexual, and transgender people,” states the California law, which was passed in September.

The law prohibits state agencies and universities from using state dollars to pay for travel to states with laws it views as discriminating against lesbian, gay, bisexual and transgender people. There are a few exceptions, such as for law enforcement purposes.

Hello?  Regulating commerce among the several States is a power assigned to Congress.  That's to take it away from the States themselves, and avoid 156-way squabbles. (13 States, 2 at a time, non-symmetric: did I get the math right?)

And yet, somehow, the current interpretation of the Commerce Clause seems to be: anything with the vaguest connection with commerce is subject to action by Congress, while still allowing the States to have their own regulations, even when it specifically involves other States.

That way lies madness.

On a more promising note:

Other states on the list are North Carolina, Tennessee and Mississippi.

California can't send emissaries to Tennessee?  Cool!  Another point in favor of moving there!  But why is there a law-enforcement exemption?  California has far too many of those.  And surely money spent in a forbidden place is no less spent there if it's spent for purported law-enforcement purposes...?

Meanwhile, on the work front

Debugging the firmware for the yonder end of that comm link I mentioned a few posts back - the Linux end seems to be fixed, modulo a couple of not-yet-implemented features that don't involve any cans of worms.

There appears to be a memory-corruption issue.  I figure it's either a stack overflow or a stray pointer.  I turned on stack checking in FreeRTOS, and didn't get anything.  So much for the semi-easy hypothesis.

I already eliminated the local implementation of printf as a culprit, by the simple expedient of replacing the calls that were happening with simpler code to emit constant strings for test purposes.

This leaves a wonky pointer, somewhere in my code (which spans three tasks) as the thing I need to track down.  Hm.  One of the tasks is pretty simple, and can be eliminated quickly.  On the other hand, there are also a few event handlers to be examined.

Ah, the dangers of C.  But just try writing this application in any "safe" programming language, to run on the target hardware (much CPU, little memory, extended temperature range, ¢heap).

And this struggle has gotten to the point where I'm not even tracking most of the time I'm spending on it, because charging the client for chasing down stupid bugs in my code Just Ain't Right.

Update: Turning off preëmptive scheduling doesn't solve the problem.

Update 2: Oh, great.  It seems to be down to a single failure mode (at the moment).  The scheduler tries to pull a TCB pointer out of a list which ostensibly contains a single item, except that the item maybe isn't real, and at any rate the TCB pointer that comes out of it is NULL.

If I put in a test for this pathological case, and attempt to deal with it, the program doesn't work at all, not even a little bit.  Which is weird, because that seems to imply that the clearly-pathological situation routinely occurs at startup, but somehow doesn't cause the failure at that point.

I are confoozed.

Update 2a:  That's what I get for working too late.  Looking at my test for the pathological case, I find that I typed = instead of ==.  So I was causing it every time, instead of testing for it.

Bye-bye, Wired.

I'd pretty much given up on Wired quite some time ago, but anyway just clicked on a link that sounded maybe interesting.

After a moment, I got a page-obscuring popup complaining about my adblocker.

The browser I was using at the time doesn't have an adblocker.

So Wired has a broken automated reader driver-awayer.  Lotsa luck with that, guys.

Fantastically bad system design

Seems some evildoers hacked the computers of a hotel in Austria, and made the door locks stop unlocking.

OK, that's bad.  Worse, though:

The guests could no longer get in or out of the hotel rooms and new key cards could not be programmed.

Couldn't get out?

In what universe is this an acceptable design for a hotel door lock?

Occupants should always be able to let themselves out  of their rooms, no matter what's gone wrong with the computers, electrical supply, or anything else.  Basic safety rule, that.

The only case where it's OK to have a failure lock people in their rooms would be a confinement facility - jail, prison, that sort of thing - and then you'd better have a plan (and the manpower) for evacuating the inmates without computer assistance if the fancy high-tech locking system is knocked out by, e.g., a fire.

Great big update: Seems the accounts circulating in the press were wrong.  Revised account says the hotel couldn't generate new key codes, but the locks continued to work, and the room locks have (as any sane fire inspector would demand) straight-up mechanical unlocking from the inside.

Work is fascinating.

I could sit and watch it for hours.

Southmoon watching AvE doing stepper-motor rotary-fourth-axis stuff.

This gives a slightly misleading idea of the relative sizes of Southmoon and the monitor.  It's not a laptop, and she's not a giant atomic mutant.

Arrogance vs. humility

Good article here.  A couple of notes:

Is the work done by financial engineers really more economically productive than the work done by civil engineers?

Now, this one gets interesting.  Not so much the "economically productive" aspect, as the "engineering" aspect.

If a civil engineer makes a mistake - one erring on the side of optimism - the consequences, when they become apparent, can be pinned on him.  Also, since the calculations are objective, and the principles well documented, a civil engineer's work can be checked by another civil engineer before the structure is built.

Spotting the errors of a financial "engineer" is more of a challenge, the consequences of error can be much more widespread, and some financiers seem to have a remarkable ability to fail upward, so that the adverse consequences of their failures fall entirely on others.

 

Everybody can do something better than you can. Imagine how much different our society would be if each of us embraced those words as a daily mantra.

Yup.  There are vast areas of endeavor at which I know perfectly well I'm not very good, and it would be better to hand such tasks off to someone better qualified.  (Later this year, for example, finding a real accountant will become a priority.)

On the other hand, sometimes finding the someone who can do it better can be a challenge, and for small projects the handoff can be more trouble than it's worth.  And sometimes doing it myself, not so well, can be educational, fun, or both.

 

I have a suspicion that religion, as well as social isolation, has something to do with the arrogance/humility divide.  While the coastal irreligious (looks in mirror) tend to perceive the our deity talks directly to us crowd as being arrogant, what are we to say of those who believe we are as gods, and it is our destiny to reshape mankind?

 

A fine example of arrogance was on display in the past week, as the Bulletin of the Atomic Scientists fiddled with its "clock".  Arrogance?  Yup.  Even if we grant the premise that the clock-fiddlers are actual atomic scientists and not mere activists playing dress-up, consider: how does expertise in nuclear physics grant insights into geopolitics?  It's not like they're related fields.

Sure, as with any technical field, the politicians routinely make false assumptions about what the technology can and can't do (see also: encryption and search warrants, or anything an urban Democrat says about guns), and need their magical thinking corrected.  But the ability to point out that the ruling class is talking nonsense in any particular field does not imply omniscience.

By extension, if a famous actor makes proclamations on any subject other than acting, it is entirely appropriate to question his expertise.

Architecture!

No, not the architect sketch.  And I'm not doing the voice.  Well, maybe I am, but that's beside the point.

Been bashing at a program.  It's based on a sort of library thing I wrote a few months ago, that was originally meant to be integrated with an application program that would be using a moderately-simple API.  Among other things, it fires up a bunch of threads to handle many communication channels.

Well, the mission changed, and now I'm trying to adapt the existing code to a very different API, for use with a single-threaded application.  Which means the code that does the adapting ends up sprouting some threads of its own, to handle sundry functions.

Threads, threads everywhere!  This is starting to look a bit Gordian.  Unfortunately, the simple solution won't work; a certain amount of complexity is necessary.

Probably need to back off, take a break, and then do some restructuring, maybe even writing an architecture document first.  Aside from anything else, the layered approach I've been taking results in two sets of per-channel data - one for the library, and one for the interface layer - with one indexed from zero and the other from one (because at the higher level there's a virtual extra channel).  And synchronization is interesting.

Doesn't help that, despite the answers one finds on-line, stdio was acting rather non-thread-safe a couple of days ago.  Or... aha.  I believe the problem came in when I was trying to read from, and write to, the same stream at the same time.  That makes sense now: the writing thread locked up because the reading thread had claimed the stream, in the interest of thread safety.  So creating a duplicate stream with the same file descriptor was in fact the right answer, and for an explicable reason.

Anyway.  Got to get this thing sorted out Real Soon Now, but I don't think it'll be this evening.

Time to grill a steak and dig out a silly movie or something.

Not quite comprehending...

I'm looking (again) and Somebody Else's Code for formatting a floating-point number into an ASCII string.

All through the code, there are separate cases for handling positive vs. negative numbers.

Hello?  Surely, for a negative input, you just emit the '-' and then run the rest of the process on -n?  (Yeah, for integers you need a special case to handle minint.  Not the case with floats, unless you're using some really weird representation.)

...And then there's the special case to handle a value of exactly zero by emitting a '0', only, regardless of the format requested by the user.  Well, that does explain why I'm having output that doesn't match the format I was expecting.  Maybe I'll just #if that case out.

Secret identity?

Has anyone ever seen David Brock and Elliot Carver together?

Did someone not get the memo?

Today's mail brought a fat envelope, addressed to me, from someone apparently named "SINGLE-PAYER MEDICARE FOR ALL HEALTHCARE REFORM".

Er.  Which party is in power in Washington at the moment?

And would you want them to be in charge of your centrally-controlled nationalized health care?

Make the Universe yuuuuge again!

Oh, and now the Hubble Constant has suddenly gotten bigger.

I didn't hear about The Donald tweeting mean things about how the Universe wasn't expanding fast enough.

Maybe the galaxies are fleeing our neighborhood?  They don't want to pay for the wall we're gonna build to keep the aliens out?

Or is this the result of the Fed's loose-money policy?  We haven't been seeing the expected consumer-price inflation here because the Universe is inflating instead?

Have I mentioned how much I dislike IDEs?

Integrated Development Environments, that is.

Yeah, they're handy at first, usually.  But then...

Got various projects under one of those Eclipse derivatives.  Using Eclipse can sometimes be a pain.

Now I'm trying to structure a project involving several models of hardware, each with its own firmware, but sharing a base design and a fair amount of code.

I eventually figured out how to include common code from the base project into a specific-model project.  So far so good.

But!  There are some things that the common code needs to know, when it's compiled, about the specifics of the target.  Things like the number of ADC channels configured.

Initially, I put the corresponding defines in the project settings, which more or less worked, though there's still a mysterious problem (presumably related to the shared source code) whereby, when I build the project, the incremental build (whether needed or not) fails with a missing include file, forcing a full build, which then succeeds.

In a vain attempt to sort out that weirdness, I moved the defines into a config.h file in the project-source directory and included it from utils.h in the common-source directory.

Problem is: I can't seem to get the right config.h included.  I play around with search paths, and angle brackets vs. quotes, and... well, sometimes I get the one from the base project instead of the project I'm building (despite the -I list apparently being passed to gcc), and sometimes I get a missing-include-file error, and sometimes I get inconsistent results.

Growf!

I may need to go back to storing configuration details in the project settings.  Which is a structural crock - that stuff should be in plain view in a header file*! - and additionally annoying in that such things get stored in multiple places in the project settings.  Also, the UI for editing the settings is not exactly friendly.

* Better, the IDE's MCU-configuration feature should generate a global project-configuration header file for me, including things like number of channels configured per multi-channel peripheral.  The AGROS configuration utility does this (like, since 2007 or something).  It saves a bunch of oh, dear, I've configured a new peripheral and now I need to figure out where to include its header file and what configuration tables to update.

The wages of isolationism

Up early.  Check world financial indices, because why not.

Huh.  This must be the collapse they all warned us about if we went all America First.

...Or maybe the Investor Caste figures America First is a bargaining position?  That the new model is really The Business of America is Business, and The Business of China is Business, and likewise Japan, and Germany, and Britain, and Lithuania, and...?

Oh, great.

While the mighty mainstream press has been covering attacks on Jeff Sessions over ancient and discredited accusations of racism, there's been a notable lack of attention paid to current and relevant reasons to oppose him.

Like this, for example.

"Encryption serves many valuable and important purposes," Sessions wrote. "It is also critical, however, that national security and criminal investigators be able to overcome encryption, under lawful authority, when necessary to the furtherance of national security and criminal investigations."

No, no, NO!

Encryption that can be overcome by any means short of actual divine intervention no longer serves those valuable and important purposes.

You might as well translate all your confidential documents to Pig Latin and call them "secure".

But we can't expect our betters in the Senate to understand this, I suppose.  You know, the ones who are supposed to be screening these job applicants.

...And then there are the fiddly details.  Remember the dark days of the Clinton Administration, and having to download your strong crypto software from Finland?  How do you keep people from doing that again?

And how do you tell if someone is using forbidden, un-backdoored crypto without going around trying to decrypt people's files at random?

Oh, I get it: we'll just have a mandatory 30-year prison sentence for using unauthorized crypto.  Catch some suspicious character with files that you can't make sense of, and there's no need to gather any more evidence - you can just lock him up under the crypto law.

Incidentally, the same trivial trick I used to plant a large image of a dead skunk in your browser cache a couple of weeks ago can be used to plant an "unbreakably encrypted" file.  Or the source code for a forbidden algorithm.

Sink the Clipper!  See also this piece from way-back-when.

And just to make things more and more wonderful

There's some nasty virus making the rounds.  Symptoms are a mashup of cold (e.g. upper respiratory crud) and flu (notably whole-body muscle pain).

Joy has a bad case of it.  I've been having vague symptoms the past several days.  If it's some close relative of swine flu, maybe I won't get the full-blown affliction.  If it's something else, ugh.

Update: Crud reworsified in the afternoon.  Pain behind the left eye, and a case of the dizzies that might be crud-related or might just be the result of not enough sleep the last few nights.  I think I'll snort some steroids* and zonk out extra early.

* Is Nasacort a 15^th century French battlefield or a truncated space agency?

Linux mystery

Got this here embedded gadget...

There's a little ARM-based System-On-Module running Linux.  Everything important is connected via USB, in particular an Imperial bargeload of FTDI serial ports (actually a quarter-barge of quad UARTs).

I fire up a test program, previously used (successfully, if memory serves) with a lash-up connected to a regular old PC running Linux.  The test program opens all the serial ports, one after another.  When one is configured for communication (and not before), it launches a thread to listen for incoming characters on that port.

Well!  In what's supposed to be the production configuration, i.e., the SOM plugged into the board with all the ports, it gets Interesting.

Communication works on the first eight ports.  The rest... well, modem control works (at least for output; guess I should try wiggling a modem-control input or two and confirm what happens - back in a moment - yes, modem-control inputs are working too, though now I notice that there's another logical-to-physical mapping function needed).  If I configure one for communication and send a message to it, outgoing data goes out, but I get no receive data.

It's not the hardware: if I tell the test program the world starts at port 8, then ports 8..15 work, just as ports 0..7 did when the world started at zero.

So!  The first eight FTDI serial ports to be opened work; those opened later fail to receive data.

Guess I'll be trying various combinations of search terms, before I go diving into the kernel (and driver) source tree in search of answers.

Hm.  I suppose before that I should unplug the SOM, cut the end off a USB cable, and try hooking the bus up to my workstation to see if that gives different behavior.  [Oh: I already have such a cable prepared, from the time between the arrival of the prototype board and my readiness to plug in the SOM.]

Update: With the gadget's USB connected to my workstation, all behaves exactly as it should.  So, something to do with the Linux/driver configuration on the SOM.  Hm. If I diff the SOM's memory card against the workstation's hard drive and see what's different... no?

A note regarding Nazis and the punching thereof

So there's this question making the rounds: Is it OK to punch a Nazi in the face?

Except... hm.  Actual Nazis are in short supply nowadays.

The actual question being asked is: Is it OK to sucker-punch a lamer who spouts hateful rhetoric?

Answer carefully, for there an an awful lot of lamers spouting hateful rhetoric, and it seems that not all of them have mirrors.

Ah, library functions...

I was about to code up a basic set of linked-list-FIFO-queue functions for some inter-thread communication in a highly-threaded program.

Hey, look!  There's this <sys/queue.h> thing.  Canned goodness!

Er, wait.  Thread-safe?  No.  So I'd need to implement my own thread-synchronization wrappers.  (Do I even need to worry about this?  What are the conditions for a thread switch, anyway?  Looks like I need to do some more learnin'.)

Maybe I'll just roll my own.  Probably easier.  But I still need to take a somewhat deeper look at how threads work.

Does anyone make stupid pills for cats?

Huckleberry has now figured out how to open the door of the cabinet where the treat box is stored, pull down the treat box, and pry open the lid.  (Consumption of the treats thereby liberated is left as an exercise for the student.)

Growf.  Guess it's back to screw-top jars for treat storage.

Or maybe I can make him watch the television.  It'll stupid him up good.