[erlang-questions] illegal guard expression: filelib:is_dir(H)

Michael Turner <>
Tue Feb 16 11:54:25 CET 2010


> He had RUN his program, you see, so he KNEW it was right.

Richard, look, I've run across the same mentality - it was while porting
some C code, where my boss said it had worked on a couple other
platforms, so stop foot-dragging, Michael and move on.  In a meeting
with five other people.  But I knew C and its limits, I knew a bad stack
crash zone when I saw one, and I can smell bogus assumptions about C
type sizes and alignments from 50 paces.  That code got fixed.  We
couldn't ship without it working.  He lost some face over it.

It's not formal correctness VERSUS Real Man Programmers with Grease on
Their Hands.  I think the problem is that formal correctness got
oversold on unrealistic merits alone, and after the dashed expectations,
nobody got in there and sold a practical and credible bill of goods with
more modest goals.

I mean, look at Peter Deutsch -- his dissertation was about the Stanford
Pascal Verifier, but he ended up being sort of anti-formalism about
programming by the mid-1970s, and the exodus continued long after his
departure.  He loved programming.  The formal correctness people didn't
give him much to help him love it even more.  Quite the contrary, at
that point.  So he moved on.  (And his advisor?  I think contributing to
the SPV with Deutsch was the last thing Richard Karp did in programming
language semantics.  He had a good nose for where you can get results.)

Things don't happen in the real world until somebody makes the sale. 
And if you screw up the first sale, especially in technology, it can be
a lo-o-ong time before the idea has a second chance.   The friend of
mine I quoted earlier sold his boss on the idea of letting him go to
that Asilomar conference co-starring Edsger Dijkstra by arguing that
knowledge of formal semantics and tools would help their company be
competitive.  He came back without any tools to recommend and with a
dislike of most of what he saw there, especially the attitudes.

I'm sure you're experienced with the phenomenon of something in
computer science being oversold on unproven merits.  You mention Prolog
quite a lot, for one thing.  Hey, don't get me wrong, I've got nothing
against Prolog.  The only reason I'm not elbows-deep into ECLiPSe right
now is that Erlang soaks up most of the time I have available for
hacking.  But you know what happened.

-michael



On 2/16/2010, "Richard O'Keefe" <> wrote:

>
>On Feb 16, 2010, at 2:36 AM, Michael Turner wrote:
>
>>
>>> The reference to Guantanamo is, um, hyperbolic, to put it mildly.
>>
>> I plead guilty to Argument by Reference to Wikipedia.
>>
>> (A mitigating circumstance: I was once told by a friend that he'd
>> heard
>> Dijkstra say, haughtily, at a conference in Asilomar, "'Run a
>> program'?!  Why, I haven't *run* a program in years!"  Well of course
>> not.  How silly.  Why run programs when you can be proving them
>> correct
>> instead?)
>
>Let's see, you have a second-hand report of *part* of a conversation,
>and you base your judgement on that?  What had the other person said
>to him?  What was the conversation about?  How do you know that the
>next sentence wasn't "I've often had a computer carry out my plan"
>or something of the sort?
>
>I've just been reading an article in CACM which is simultaneously
>inspiring and deeply saddening.  The article is about Coverity.
>The inspiring part is that they've built a wonderful tool for
>checking C code.  The deeply saddening part is the refusal of many
>people to believe that their grossly nonstandard code is anything
>but perfectly legal C.  One example sticks in my mind:
>
>	int a[2], b;
>	/* context: sizeof (int) is 4. */
>	memset(a, 0, 12);
>
>Coverity pointed out the bug.  The customer said "no no, your tool
>is broken, I *meant* that because a and b are side by side."
>The fact that the C standard explicitly denies any relationship
>between the addresses of a and b, and that there have been (and
>for all I know may still be) compilers that sorted globals,
>putting the small ones first so it had a fighting chance of
>being able to address most of them with a single register, which
>compilers would have guaranteed that a and b were NOT adjacent,
>all of this was of no importance to the customer.
>
>He had RUN his program, you see, so he KNEW it was right.
>
>If you want to quote Dijkstra, quote this:
>
>	I would therefore like to posit that computing's
>	central challenge, "How not to make a mess of it,"
>	has not been met.  On the contrary, most of our
>	systems are much more complicated than can be
>	considered healthy, and are too messy and chaotic
>	to be used in comfort and confidence.  The average
>	customer of the computing industry has been served
>	so poorly that he expects his system to crash all
>	the time, and we witness a massive worldwide
>	distribution of bug-ridden software for which we
>	should be deeply ashamed.
>By the way, this doesn't contradict Armstrong's "Let it crash"
>dictum.  Joe is talking about _components_ hidden inside a
>system crashing.  Erlang *systems* aren't supposed to crash,
>which is why Francesco Cesarini and Simon Thompson stress
>early on in their wonderful book that processes should be
>watched by some other process.
>
>Whether Dijkstra personally ever ran a program or not (and he
>devised the implementation techniques at the heart of Java),
>his _notation_ deserves a reasoned responds on its own merits
>for what it is, not what it might be imagined to be.
>
>
>


More information about the erlang-questions mailing list