How do we tell truths that might hurt? (1975)(cs.utexas.edu) |
How do we tell truths that might hurt? (1975)(cs.utexas.edu) |
That isn't to say that he did not have valid reasoning for his dislikes as a language expert and a mathematician but those reasons are not ones the bulk of practitioners would likely agree are valid.
And it's not like he was a lone crank railing against the 99% who were moving forward with a well-founded idea of how computing could scale indefinitely by composing incorrect programs. Just like today, 99% of programming, including programming for government programs and vital infrastructure, was done by people who were only hoping to make their own projects succeed well enough for the next six months to three years. I'm sure there were brilliant people who made intelligent arguments against the need for correctness, but their arguments didn't carry the day. Complacency and short-term thinking did.
In that context, Dijkstra's pessimism and his use of harsh, attention-getting language makes a lot of sense. How many people at the time really understood that in 2020 every part of our civilization would depend on code compiled by compilers with bugs, linked with libraries with bugs, in virtual machines with bugs, on operating systems with bugs, on CPUs with bugs in their microcode, and yet it would still all mostly work?
That is clearly wrong if one is willing to take a moment to stop gazing at the wonder of pure mathematics and look at the outside world. There is no notion of "correctness" for the pyramids of Egypt, the dykes of the Netherlands, Milan Cathedral, or the world economy and yet those huge-scale systems all function.
> Dijkstra's pessimism and his use of harsh, attention-getting language makes a lot of sense.
It makes sense in terms of Dijkstra's personality, but it makes no logical sense. If you believe mathematical correctness is such a valuable principle, then you should be able to leverage that same principle in one's own arguments. The fact that Dijkstra couldn't dispassionately prove that correctness was vital and had to resort to emotionally-loaded weasel words undermines his own claim.
Of course there are critical software engineering projects as well, Boeing MAX comes to mind.
The BASIC he is describing had no call stack, nothing that would resemble a function in today's languages. FORTRAN at the time handed masses of state around in global variables and the latest version had just added subroutines and functions. This is after ALGOL 60 had established the model that we all take for granted today.
I think the bulk of practitioners today would absolutely agree with him.
About programming languages I don't think there's any question he was right. Sometimes we agree with him even without realizing it, for example, the original argument of GOTO considered harmful was that GOTO statements decreased "linearity" by having the control flow jump to random places. The recent trend in mainstream PLs to adopt functional-style control structures (JS array methods, Java streams, etc.) is predicated on the exact same rationale.
I'm also in broad agreement on his stance about natural language being just flat out wrong in the context of computing.
But the man made an exorbitant amount of claims, many of which are just provably wrong. The bits about BASIC and COBOL are particularly egregious examples of his at times cavalier attitude. After all, arrogance is measured in nano-Dijkstras.
> It is practically impossible to teach good programming to students that have had a prior exposure to BASIC: as potential programmers they are mentally mutilated beyond hope of regeneration.
is at best a hyperbolic joke. I dearly hope he wasn't being serious.
"The second major development on the software scene that I would like to mention is the birth of FORTRAN. At that time this was a project of great temerity and the people responsible for it deserve our great admiration. It would be absolutely unfair to blame them for shortcomings that only became apparent after a decade or so of extensive usage: groups with a successful look-ahead of ten years are quite rare! In retrospect we must rate FORTRAN as a successful coding technique, but with very few effective aids to conception, aids which are now so urgently needed that time has come to consider it out of date. The sooner we can forget that FORTRAN has ever existed, the better, for as a vehicle of thought it is no longer adequate: it wastes our brainpower, is too risky and therefore too expensive to use. FORTRAN’s tragic fate has been its wide acceptance, mentally chaining thousands and thousands of programmers to our past mistakes. I pray daily that more of my fellow-programmers may find the means of freeing themselves from the curse of compatibility."
To this day if I saw 6502 assembly on a resume I'd be intrigued and if I saw BASIC I'd view it as a yellow flag.
> Python —"the infantile disorder"—, by now nearly 30 years old, is hopelessly inadequate for whatever computer application you have in mind today: it is now too clumsy, too risky, and too expensive to use.
> It is practically impossible to teach good programming to students that have had a prior exposure to JS: as potential programmers they are mentally mutilated beyond hope of regeneration.
> The use of Java cripples the mind; its teaching should, therefore, be regarded as a criminal offence.
The quotes about languages were always controversial, weren't they? But it's clear now in retrospect what Dijkstra was complaining about. He found FORTRAN to trick people into thinking that programming was merely about specifying arithmetic operations in a certain order, considered BASIC to require mental models which rendered folks memetically blind to actual machine behaviors, and thought COBOL tried to be legible to management but ended up being confusing to everybody.
> Many companies that have made themselves dependent on AWS-equipment (and in doing so have sold their soul to the devil) will collapse under the sheer weight of the unmastered complexity of their data processing systems.
Yep.
> In the good old days physicists repeated each other's experiments, just to be sure. Today they stick to Python, so that they can share each other's programs, bugs included.
Reproducibility is a real problem, and sharing code is just the first step. It's an embarrassment to physics and mathematics that we don't have a single holistic repository of algorithms, but have to rebuild everything from scratch every time. (Perlis would tell Dijkstra that this is an inevitable facet of computing, and Dijkstra would reply that Perlis is too accepting of humanity's tendency to avoid effort.)
> You would rather that I had not disturbed you by sending you this.
Heh, yeah, let's see what the comment section is like.
I mean, if FORTRAN is a mess, then isn't ALGOL too? Is anyone here old enough to remember?
ALGOL 68, on the other hand... "The more I see of it, the more unhappy I become." [0]
[0] https://www.cs.utexas.edu/users/EWD/transcriptions/EWD02xx/E...
This is not the same as saying that EWD thought that ALGOL was good...
how they hurt in 2012 https://news.ycombinator.com/item?id=4926615
and 2011 https://news.ycombinator.com/item?id=2279260
Smaller threads:
I think the second issue, is it is fairly safe to say that the path from Python 2 to 3 was not well thought out and has been a disaster. A lot of people where burned by it, and it left a bad taste in a lot of peoples mouths.
That being said, Python enjoys a huge userbase so I would not worry about the hate, it's just not the language of choice for some people and that is fine.
First, the excellent readability leads directly into hard-to-read code structures. This might seem paradoxical but Dijkstra insisted that the same thing happened in FORTRAN, and I'm willing to defer to his instinct that there's something about the "shut up and calculate" approach that physicists have which causes a predilection for both FORTRAN and Python.
Second, Python 2 to Python 3 was horrible, and created political problems which hadn't existed before. Now, at the end of the transition, we can see how badly it was managed; Python 2 could have been retrofitted with nearly every breaking change and it would have been lower-friction. Instead, there's now millions of lines of rotting Python 2 code which will never be updated again. Curiously, this happened in the FORTRAN world too; I wasn't around for it, but FORTRAN 77 was so popular compared to future revisions and standardizations that it fractured the FORTRAN community.
That said, he did have relatively nice things to say about Haskell [2] and preferred Haskell to Java:
> Finally, in the specific comparison of Haskell versus Java, Haskell, though not perfect, is of a quality that is several orders of magnitude higher than Java, which is a mess (and needed an extensive advertizing campaign and aggressive salesmanship for its commercial acceptance).
I imagine that he would have liked something structured, equational, declarative, and modular; he would have wanted to treat programs as mathematical objects, as he says in [2]. Beyond that, though, we'll never know. He left some predictions like [3] but they are vague.
[0] https://en.wikipedia.org/wiki/ALGOL_60
[1] https://www.cs.utexas.edu/users/EWD/MCReps/MR35.PDF
[2] https://www.cs.utexas.edu/users/EWD/transcriptions/OtherDocs...
[3] https://www.cs.utexas.edu/users/EWD/transcriptions/EWD12xx/E...
The first languages and their compilers were strongly driven by hardware constraints, a kilobyte of memory costing an arm and a leg.
Imagine storing function names in memory to compile the program, it doesn't fit in 1 kB. Imagine storing the whole source code in memory for processing, it doesn't work when there is less than a 1 MB of memory available.
It's ridiculous today but it's real reasons why things were made global back then or why C/pascal split the code between a header and a source file.
And, for all his complaining, I don't know of any language that he authored. He's sure good at telling everyone that they're doing it wrong, though...
Also the migration from python 2 to 3 is actually the best migration ever. Breaking changes on a code that has a massive library around it. All code is migrated along with the libraries. I haven't heard of any migration done as successfully as python. It required movement of the language and the entire community to make it happen.
Most migrations refrain from breaking changes and instead they take the less risky route and make the language backwards compatible at the cost of being more bloated. See C++ 17 and C++ 20.
But despite this, yeah these are complaints that I've heard, but this is totally unrelated to the hate I see. Like parentheses in lisp is something to complain about but not something to hate lisp for.
The hate I believe is more egotistical than anything. One interviewer told me I could code in any language other than python because python makes things too simple. The hate is because they believe python programmers are stupider than normal programmers.
That being said my favorite language is not python. It's my least favorite language out of all the languages I know well but not a language I hate.
> gotos and state machines (which are considered best practices) are basically equivalent
WTF? State machines can be implemented in a lot of ways: mutually recursive calls, pattern matching, function pointers... In what sense are they "equivalent" to gotos? Also what does it mean that state machines "are considered best practices"?
Wow, couldn't disagree with this more, at least on your examples of civil engineering (buildings, dikes). There are testable, comprehensive physical principles that govern whether any of these engineered products function in their most fundamentally-intended ways. This is the reason most buildings are resilient and don't collapse under load, or that dams keep water from flowing uncontrolled. You can debate "correctness" in the sense of the purpose the product serves, but there is the principle of correctness of construction which your civil engineering examples (and anything truly engineered) satisfy. Correctness in construction is not subjective.
Perhaps, but it's never certain. So things that are "correct" can still be wrong.
Here's how Einstein expressed his disagreement with what you are saying:
> As far as the laws of mathematics are certain, they do not refer to reality, and as far as they refer to reality, they are not certain.
Nothing in physics (or any other science) is certain. All one can do in science (other than mathematics) is disprove. One cannot prove.
To prove you would have to be all knowing. You would have to have taken into account all relevant aspects of all physical characteristics. For example, all known aspects of quantum mechanics, including the uncertainty principle, AND all relevant unknown aspects of quantum mechanics, which is guaranteed to be hit and miss, and can rationally be expected to contain an unknown number of misses that isn't zero.
It's the human propensity to ignore these fundamentals that leads to things like the Tacoma Bridge collapse.[0]
[0] https://en.wikipedia.org/wiki/Tacoma_Narrows_Bridge_(1940)
These are statistical engineering tests of the probability of failure under certain conditions. That is not at all what Dijkstra would consider to be "correctness". Dijkstra is talking about mathematical proof. In mathematics, one does not say "1 + 2 = 3 plus or minus 0.1 with a safety factor of 2".
It's unclear how this would be any different than mathematically formalizing a distributed system, identifying the properties that constitute correctness of operation of that distributed system, and symbolically proving that subject to certain assumptions, the distributed system model does or does not satisfy those properties. This would presumably be consistent with the Dijkstra view of mathematical correctness.
I think it's impossible to come up with a rational judgment of whether the world economy "functions." I mean, those of us who are alive are alive, but there are a lot of things that have happened that haven't been ideal.
Besides the world economy, nothing you mention has the functional complexity of a large piece of software, and none of it was designed and built with much confidence, other than confidence that came from similarity to previous designs, and from an optimistic outlook that issues encountered during construction could be coped with. This wall is cracking, so we tear it down and rebuild it a little thicker. We have a few centuries, after all. I think you're right that modern software projects are analogous to ancient pyramids and medieval cathedrals. That's pretty much where we are. Yet somehow, working at that low level of sophistication, we've built working software that is orders of magnitude more complex than cathedrals.
I'm not saying I know how to improve on the situation, or even that we can; I'm just pointing out how different it is from, say, contemporary structural engineering where we can analyze the design of a novel structure and have a pretty good idea of whether it will be safe based on its geometry and known characteristics of materials and how they're joined. It's amazing that we've reached the scale we have with nothing comparable.
> If you believe mathematical correctness is such a valuable principle, then you should be able to leverage that same principle in one's own arguments.
The idea that mathematical reasoning is on some spectrum with dispassionate discourse has motivated hundreds of years of attempts to improve natural language by making it more mathematical. Result: some interesting contributions to philosophy but very little change in how people communicate. People are still people.
The issue have nothing to do with mathematical correctness, so you can't apply the same principles. This is not math. But, Dijkstra's it's not being logical in arguments, because he is not of dick and likely this style of arguing was working for him in his life.
I had an interviewer tell me that I couldn't code up the solution in python. I think it might be because python is so easy that people look down on it.
Python annoys those people because it's both relatively easy to get started with and far, far more successful than whatever their current favorite language is, and this is portrayed as people not getting it rather than having a more insightful discussion about whether other people might reasonably make decisions based on different needs, background, and resources rather than stupidity.
He helped create the ideas of Structured Programming, which most of us now take for granted, since pretty much every language in popular use these days are based on these ideas.
Zig doesn't do either of those things. There are a fair amount of criticisms of the mental model of the author that I've seen voiced - some including security.
What's worse, the community surrounding Zig (in particular, the Discord community) operates more like a cult - any negative questioning gets you shunned.
I was personally a huge fan of Zig until a number of questionable design decisions and dismissed bug reports lead me to believe it will forever remain a toy language. I can't imagine Dijkstra approving.
The cult-like attitude that many programmers have about languages certainly supports Dijkstra’s claims about the immaturity of the field.
I've read countless anecdotes from perfectly capable programmers who not only learned, but first discovered their interest in the field, via BASIC. No language "mentally mutilates programmers beyond hope of regeneration". And BASIC probably isn't even the worst one for learning purposes.
Remember, Dijkstra was a professor and he was an expert on teaching programming. He would have far more experience with the effect he claims than I do. I hazard to guess that is true with respect to you too.
And yes he’s obviously engaging in rhetorical hyperbole in the text you quote.
EWD's quote is from 1975 which is just barely the start of the microcomputer era (the Altair 8800 came out in '75 and the Apple II in '77), so he's not referring to those. Most people writing programs would have had access to mainframes and therefore had access to multiple language interpreters and compilers I would think.
This doesn't make any logical sense. Your saying take the breaking changes in python 3 and put it into python 2? That's just a version number. You can call it version 2.999999.8 and do all the changes in there and the outcome is identical.
No. Every breaking change must have a downstream change in every library that uses that breaking change. That's the reality of breaking changes. No way around it.
Tell me of such a migration as huge as python 2->3 that was as successful. For sure there were huge problems along the way and it took forever. However I have heard of very very few migrations in the open source world that ended up with an outcome as successful as python.
>First, the excellent readability leads directly into hard-to-read code structures.
I don't agree with this either. You refer to fortran but most programmers here haven't used it so you'll have to provide an example for readers to see your point.
* Perl 5 to Perl 6: So disastrous that they rolled back and Perl 6 is now known as Raku
* PHP5 to PHP7: Burn my eyes out, please! But of course PHP has unique user pressures, and a monoculture helps a lot
* Python 2.4 to Python 2.7: Done in several stages, including deprecation of syntax, rolling out of new keywords, introduction of backwards-compatible objects and classes, and improvements to various semantic corner cases
* Haskell 98 to Haskell 2010: GHC dominated the ecosystem and now Haskell 98 is only known for being associated with Hugs, which knows nothing newer
* C++03 and earlier to C++11: Failed to deprecate enough stuff, but did successfully establish a permanent 3yr release cadence
* C99 to C11: Aside from the whole Microsoft deal, this was perfect; unfortunately Microsoft's platforms are common in the wild
Now consider how many Python 3 features ended up backported to Python 2 [0] and how divisive the upgrade needed to be in the end.
On readability, you'll just have to trust me that when Python gets to millions of lines of code per application, the organization of modules into packages becomes obligatory; the module-to-module barrier isn't expressive enough to support all of the readable syntax that people want to use for composing objects. If you want a FORTRAN example, look at Cephes [1], a C library partially ported from FORTRAN. The readability is terrible, the factoring is terrible, and it cannot be improved because FORTRAN lacked the abstractive power necessary for higher-order factoring, and so does C. Compare and contrast with Numpy [2], a popular numeric library for Python which is implemented in (punchline!) FORTRAN and C.
[0] https://docs.python.org/2/whatsnew/2.7.html#python-3-1-featu...
Did you by any chance observe any of the folks involved officially saying that deprecating things was one of the goals? I thought keeping working code working has always been of the C++'s official goals.
It's the kind of ranting you do with your colleagues in a pub after-hours, not something you publish in a professional context and frame as "unpleasant truths", as if they've been proven through scientific measurement.
They did not permit recursion (direct or indirect). Almost by design they encouraged spaghetti code (the thing Structured Programming was meant to work against). They encouraged (or required) global state. They discouraged modularity. They were effective languages (in the sense that people got stuff done in them), but they were poor languages when compared to the capabilities of their contemporaries.