• (nodebb)

    THIS IS THE BEST WAY!

    Explicit unrolled static conversion easily optimized by the compiler for best performance!

  • MacFrog (unregistered)

    TRWTF is that there are 256 possible error codes.

  • MacFrog (unregistered) in reply to MacFrog

    "No error" is a valid error code as well, isn't it?

  • A_L (unregistered)

    First strike!

  • I'm not a robot (unregistered)

    Better than I was expecting when I saw the mention of the Excel file. I was thinking it would actually open the file at runtime (using the bare minimum hackiest-possible code that "works", of course, not a proper robust Excel reader library) and look up the error number dynamically.

  • DQ (unregistered)

    Ah, Excel as code generator. I must admit I've done it multiple times (usually for creating batch files) ren a1234567.txt b7654321.csv ren b2345678.txt c8765432.csv ...

  • DQ (unregistered)

    Why can't I add a 'newline' in a comment? It always looks so messy without.

  • TheCPUWizard (unregistered) in reply to DQ

    Could it be a lack of Skill?

  • (nodebb) in reply to DQ

    The skill in question being that you add two.

    One after the other.

  • DQ (unregistered) in reply to Steve_The_Cynic

    Have to try that

    Working ?

  • DQ (unregistered) in reply to TheCPUWizard

    That's why I come to this site: You always learn something new ! (note: I said 'new' not 'good')

  • Oliver Jones (google)

    There was the communications software package for the natural gas pipeline control room. It seems there were microwave towers at compression stations along that pipeline's 3 megameter (2000 mile) path. There were two guys in the control room, always alert, not fresh-out-of-school. One of them had a face that showed evidence of having been burned.

    The messages were, if remember right, 16 bytes long followed by a four-byte CRC code. Stuff like "station id, subsystem id, component id, sensor id, measurement. The contractor was sorting out some sort of big-endian / little-endian hassle in the device driver.

    The development coordinator gave the contractor (me) a test message to test the new comm device driver.

    Addendum 2017-07-26 08:10: It said "station 6 / building 1 / heater / thermostat / temperature = 512" It was supposed to be innocuous and obviously wrong.

    The contractor put it into the system through the test port.

    I've never seen two guys move so fast as those control room guys. By the time the coordinator could say "it's a test" they had helicopters in the air and fire trucks rolling from the nearest prairie town.

    I guess, when pipeline compressors catch fire, they CATCH FIRE. And the building thermostat survives a few milliseconds longer than other stuff.

    Moral of the story. PLEASE include real test data points in your monitoring system design.

  • Brian Boorman (google)

    Using Excel to generate code is not always a WTF. I've found it useful for generating static LCD screen bitmap array code for embedded products. Using '0' and '1' in the cell and conditional formatting on the value to color the cell gives you the visual. Then use another tab with concatenate and bin2hex formulas to generate the arrays. Export that tab as a CSV file and copy/paste into the .c file with the rest of the code.

  • Chronomium (unregistered)

    In my experience, Excel's formula filling is great for converting one thing to another thing in bulk when it's a simple conversion (get first three letters, hex/dec swap, stuff like that).

    Unfortunately, in this case, you need to do something like =mid(bin(x,8),y,z). And that requires knowledge of what binary is, which was the problem in the first place.

  • Appalled (unregistered) in reply to DQ

    "Newline". Try View Source and see how everyone else does it

  • Well (unregistered)

    it's not that difficult to generate binary tables, in the lsb column you fill the first two cells with "0" and "1", mark the two values, copy-paste it, mark the 4 values, copy-paste, repeat. (alternatively, just type 0, down, 1, down, ...)

    it gets easier with the higher bits (you can drag the 0 a bunch of times, then the 1 a bunch of times, then start copy-pasting)

    i'd say that table cost him less time than it would have been for him learning bit-magic (although knowing bit-magic might have helped him in other points of his carreer later)

  • Foo AKA Fooo (unregistered) in reply to MacFrog

    Well, it says: "You might do this for 254 rows of data. Which, as a note, the range of possible values is 255, so guess what was causing the error?" In fact, he did it for 255 rows, and the range of possible values (not only error values) is 256.

    So he did present an off-by-one error making another off-by-one error in the posting. (If this is reposted 254 more times, there will be no rows left ...). Kind of like Muphry's law of off-by-one errors. I suggest we call it Murph's law.

    (PS: Wow, only 2 reCAPTCHA screens today! And I also made an off-by-one error in my input there, apparently they don't really care what you click, just keep you busy for a while ...)

  • (nodebb) in reply to Oliver Jones

    "obviously wrong"

    In blatantly hazardous systems like gas pipelines (and, in fact, most industrial systems, but gas pipelines transport potentially explosive materials), there are no obviously wrong values.

    Just like there are no lights that light up to tell you there's a fault. A light that goes out to tell you there's a fault can also tell you that the light itself is faulty (in particular, it can tell you that the filament is broken). A light that lights up to tell you there's a fault cannot tell you that there is also a fault with the light that stops it lighting up.

  • no unit tests? (unregistered)

    Was it unit tested? LOL. That's the real WTF. The problem is that it doesn't work for some reason so a unit test would have caught this.

  • boomslang (unregistered)

    Pause for a moment to consider a problem: you receive a byte containing an set of flags to represent an error code.

    I'm confused. Are you receiving an error code, or a set of flags? They are two different things. Flags are usually checked by using masks, such as if (flags & 0x01) { /* handle flag one / } whereas error codes are checked by value, such as if (err == ENOENT) { / handled noent error */ } I don't see how bitshifting makes sense to determine which bits are set.. it just leaves you with an ugly loop.

  • operagost (unregistered)

    You know, if I were sending a test message that was "obviously wrong" for an operation where high temperatures were a VERY BAD THING, I would choose a unlikely value that was very LOW. Like zero, or negative, if the variable is signed. Just a tip!

  • RichP (unregistered)

    As the old saying goes: There are 10 kinds of people in the world, those who understand binary, and those who don't.

  • Ross (unregistered)

    Excel is a good tool for exploring truth-table possibilities.

    But then once you've reduced the problem to the minimum number of OR and AND combinations, you throw away the Excel file.

    You don't use it to GENERATE PRODUCTION CODE. Geez.

  • SyntaxError (unregistered)

    Picks up stone, searching for target...

  • Herby (unregistered)

    TRWTF is Excel. Everybody knows that. It is a good thing that the spreadsheet didn't use VBA, or Access. That would have been a cluster-WTF of the highest magnitude.

  • (nodebb)

    I can hardly wait to see the excel when 32 flags are needed. Or 64...

  • Flips (unregistered) in reply to Foo AKA Fooo
    So he did present an off-by-one error making another off-by-one error in the posting.

    Sounds like one of my biggest problems. I really should implement a 'one-indexed array interface' for every list I encounter

  • Ulysses (unregistered) in reply to Appalled

    Heh, 'p' tags are definitely not how everyone else does it.

    Also, I second the reaction that I certainly wouldn't reach for any bitshift operator. I would use a predefined mask or an outright bitfield.

  • Now I'm scared (unregistered)

    The question is how they tested that to make sure that what it spits out is what they expect.

    I know how I would test it, but then I also know how I would implement a flag check. The question is how someone who doesn't understand bit masking or shifts would do it.

  • Jeffrey Nonken (unregistered)

    Eh, I use Excel (actually LibreOffice) for generating production code all the time. It's great for generating large lookup tables and repetitive code, and copy/paste them into the source. I then check the .xls files in along with all the other documentation I generate.

  • reCAPTCHA sucks (unregistered)

    I'm not a robot.

  • vmp (unregistered)

    +1 for generating embedded graphical LCD code with excel.

  • Mirno (unregistered)

    "At this point, you’re probably reaching for a bitshift operator, because that’s the easiest way to do it."

    Um.... No. The easiest way to do it (because it's self maintaining) is to use bit-fields.

    union { unsigned char raw; struct { unsigned char a : 1; unsigned char b : 1; unsigned char c : 1; unsigned char d : 1; unsigned char e : 1; unsigned char f : 1; unsigned char g : 1; unsigned char h : 1; } fields; } my_err;

    my_err.raw = val; if (my_err.fields.a == 0 && ...

  • Well (unregistered) in reply to Mirno

    of all the attributes i could give to that method, "easiest" is probably not amongst them...

    first, you need already a pretty deep understanding of C to know how unions work

    second, C has so many "undefined behavior" pitfalls that you can't be sure that this works across all platforms, all endiannesses, all bitwidths, all compilers, ...

    third, you're probably on that one platform/endianness/bitwidth/compiler which does things different, and then you're a day deep into debugging why your code doesn't work before you find this spot again. and then you dig another day into barely documented pragmas that make the compiler handle this more to your liking.

  • Idort (unregistered) in reply to Brian Boorman

    I've found the PBM/PPM image formats to be good for this kind of thing. It's got the same format as a C declaration, and gvim has a syntax mode for it so you can visually edit bitmap images with a text editor.

  • eric bloedow (unregistered)

    this reminds me of my College project(s): several programs, and the output from the first one becomes the input of the second, etc. i made a small mistake in the first one, which checked to make sure the data was valid, and a field with "month:32" slipped through. (i checked to be sure it was a number and not letters, but forgot to make sure it was in the 1-12 range) so when the second program tried to look up #32 in a 12-field table...CORE DUMP!

  • idioot in de boot (unregistered)

    wow such captcha much spam

  • boomslang (unregistered)

    Um.... No. The easiest way to do it (because it's self maintaining) is to use bit-fields.

    Using an unsigned char as a bitfield is not supported by the C standard, however, some compilers like gcc allow it as an extension. If you compile with the -pedantic flag, you will receive warning messages for using unsigned char type for bitfields: "warning: type of bit-field ‘a’ is a GCC extension [-pedantic] ..." Also, using "a", "b", "c" for bitfield names is not very descriptive. Look through the Linux kernel source. Flags are always compared to predefined masks.

    Compare fields.a = 1, fields b = 1, fields.c = 0, fields.d = 0, fields.e = 0, ... to flags = VM_READ | VM_WRITE;

    From the 1989 C ANSI Standard: "A bit-field may have type int , unsigned int , or signed int ."

  • bobcat (unregistered) in reply to RichP

    ... and those who know this statement is secretly in base-3.

  • bobcat (unregistered) in reply to Steve_The_Cynic

    Which is why a single light that is green for normal operation and red for a fault is good. Then a dark light indicates a failure of the light.

  • Unnamed (unregistered)

    The horrible thing is that I think I might have worked with this person. I quickly left that place.

  • Axel (unregistered)

    I'm confused. 255 is missing; is that the code for "no errors" or for "every error happened at once"? Seems to me that the former would cause near-continuous problems and the latter shouldn't happen but once in a great while.

    So how was it happening infrequently, but often enough to be bothersome?

Leave a comment on “The Nuclear Option”

Log In or post as a guest

Replying to comment #:

Log Out

« Return to Article