Hopefully more controversial programming opinions
prog21.dadgum.com
prog21.dadgum.com
I have an awesome story that i think I've told before, but I'll tell it again anyway :)
My father-in-law is a pharmasist. back in the early 90s computers were starting to make their way into Pharmacy and the first few pharmacy sortware packages appeared. this increased with time, as you'd expect. At some point he decided it was inevitible but he decided (I have no idea why, except that he's always been quite independent) he'd write his own software. Took himself off to nightschool and learned dBase III. He'd never written a line of code prior to this.
He then proceeded to write an entire pharmacy suite of software tools that are still in use. They've been heavily updated and added to over the last 20 years but it's still all DBase III running on DOS. It's a saggering amount of code over hundreds of files but it all works. I've seen the code and it wouldn't pass a first semester coding course. each programme is a single 10,000 line spaghetti-fest. I've tried to explain functions/methods to him but he didn't understand and in the end I gave up. apparently he doesn't need them!
I can't understate how complex this software is - it's stock control, point of sale (including the usual things like birthday vouchers, discounts accounts etc etv), prescription dispensing, patient records etc. everything for a modern pharmacy. The australian health dept know about it and it's met all their criteria. it does everything.
It's mindbogglingly awfull bad code by any metric imaginable but yet it's robust, appears to be as-good-as bug free and has been able to be maintained (gov regulations for pharmacy change constantly so it's required major changes each year to keep up to date) and in production use for almost 20 years. oh, and it's fast :)
[Edit: looking at the other comments it seems I've replied to the wrong thing. I was actually writing a comment in response to this http://prog21.dadgum.com/87.html (Write Code Like You Just Learned How to Program) which was linked to from the main article]
I find it hard to believe that someone was maintaining it for 20 years without improving the quality (of the code/design) and making it easier to maintain. Have you seen the code base of the initial version or of the version that is currently being used?
The short answer to your question is no, I didn't see any evidence of improved structure or practices at all. All variables global. goto ftw. no procedures/methods/functions at all. He's never even heard of the concept when I tried to explain.
[Edit: managed to get hold of some of the code (see parent) and it looks like I was wrong on the indentation - he did start to use it at some point]
Thanks for sharing!
He thought himself how to write Perl in a week. And then started coding. He wrote a script to automate a very tedious testing process so perfectly, we were able to reduce our test team by like 70% and test more precisely than before.
It was fully procedural code. And it looked monstrous. Our architect challenged him that it won't workout. He even challenged to rewrite it in Java, but that project never finished. He was never able to ever complete the rewrite.
It motivated me to learn Perl. Since then my life has been totally different. I've done mountains of work alone working, hacking during nights.
I learned the following things from that episode:
1. First write the program.
2. Write the correct program.
3. Write the program to run fast and efficient(scalability and all that)
4. Beautify the program.
But most people never cross 2)- The point is there are tons of people who can do this tasks related to micro optimization. The people who win in a ordinary work environment are ones who know the art of converting ideas to sell able products in the fastest way possible.
By the way that manager also wrote database clients in C++ which helped us troubleshoot our in memory databases from remote locations. Again the architect challenged to get it done Java, as before the project never completed :)
Its a fact that majority of the software world is hacked and held in existence with tools like Perl and Php. The people who care about artistic elegance are few, they often fail and generally don't matter.
- Make it work
- Make it right
- Make it fast
My process tends to be:
1. Make it work - Knock up a prototype (lots of assumptions made and noted, limited error checking, etc)
2. Make it right - By starting from scratch and use what was learned in making the prototype
3. Make it fast(er) - Having a prototype, and trying it at scale, should have already given me an idea of what algorithms/structures to use/avoid in Stage #2 but, obviously, a further optimisation stage will almost always be beneficial.
Ideally (if time allows) the prototype is written in a completely different language to the desired target. This is the most important point.
Most of the production code I write is C and I tend to do my prototypes in Perl. The helps me for several reasons:
a) ADTs (lists, hashes/dictionaries, etc) are just easier/quicker in Perl/python/ruby than in C (I do have my own C library that I reuse for ADTs so it's not as if it's a great chore writing them each time, but ADT instantiation is just easier in Perl/python/ruby than C).
b) It's also easier to play around with different algorithms if you can switch between ADTs more easily.
c) I can use the prototype stage to continue to learn new languages (it used to just be Perl but has extended to include languages like Python and ruby). This list will keep on growing.
d) It prevents me (or the business) stopping at Stage #1 and using the prototype code for production. (This could be even worse if the prototype was written in C.)
e) A prototype, being quicker to knock up, means faster to getting it in-front of someone else to look at and give feedback.
f) It prevents me just copy-and-pasting chunks of the prototype code as production code so I, hopefully, don't carry over any of the assumptions I'd made in creating the prototype.
g) "Knock up a prototype" doesn't mean no design/testing/etc. I usually end up with design notes and some automated black box testing that can be reused in the later stages.
Disclaimer: I like C/C++/Python.
The real issue I think is heavy OO programmers(read java) can't live without.
1. Design patterns- The art of bloating already heavily bloated code.
2. Avoiding meta programming of any kind.
3. Love towards getters/settters and absolutely anything that lends to code bureaucracy.
Why get/set variable directly when you can write tens of classes of methods to set them?
4. Avoid learning the command line. And write bad implementations of sub functionalities of tools like awk/sed to achieve the same tasks.
5. Lengthen anything and everything as much as you can. thisIncludesTheVariableNames, Class names, methods names, package hierarchies, class inheritance. You name it they can pointlessly lengthen it.
6. If you write code in file at least 40% of the lines MUST be try/catch statements. If you can't be satisfied with that write your won exception classes and invoke point 5.
7. Always write code which can't be read or figured out unless you have auto complete, intellisense and other IDE goodies.
8. Use XML as much as you can so that you can dogfood your insanity even more.
9. Make the code so verbose that even simple programs sound like very complicated ones that can be written only by you.
and many more...
If you don't follow these points Java programmers/architects think you are writing bad code.
100 lines of python can be turned into a hundred men-hours / 6-digit budget project in Java
Try building a (safe) web app in C and see how long it takes.
I'll give you a short story that approaches this issue from both sides. A client walks into a consulting shop with a piece of really bad client/server software written by a different consulting shop. The server is written in Ruby/Rails and the client in Java (Android). Both client and server code is horrible even though the server is written in Ruby (a supposedly beautiful/compact/expressive language) and the client in Java (super ugly long winded grandpa language, or whatever). The client paid the original consultants about $60k in total. One Ruby engineer and one Java engineer rewrote the entire thing in a few days for 1/20 the cost without really reusing an ounce of the original code. Ruby didn't stop the original server engineer from delivering horseshit in too much time, and Java didn't stop the second client engineer from delivering a clean/functional/performant app in just a few days.
But I've never seen a "software consultancy company" do a project except in Java
Ruby and Python can be beautiful yes, but in the hands of average Java programmers it just becomes a horrid mess
The issue is with a culture that thrives on complicating relatively simple concepts by abusing those tools, not with the tools themselves.
Design patterns don't make for bad code, bad coders do.
Design patterns- The art of bloating already heavily bloated code.
Design patterns are to OO code what salt is to food. Put just the right amount, and you get a tasty meal. Put too much, and you get an inedible pile of crap. Design patterns are not the problem, they are a solution. Bad developers who don't understand when and how to use them are the problem.
Love towards getters/settters and absolutely anything that lends to code bureaucracy.
It's true that Java is more bureaucratic than most languages, but there is a reason why getters and setters are the standard. And the reason is encapsulation. I can expand on this further if you want me to. For what it's worth, I think that this is one of the things that Scala got right [1].
Lengthen anything and everything as much as you can. thisIncludesTheVariableNames, Class names, methods names, package hierarchies, class inheritance. You name it they can pointlessly lengthen it.
CnfNfEx. ConfigurationFileNotFoundException.Which one of these two names do you think describes what the class does better?
If you write code in file at least 40% of the lines MUST be try/catch statements. If you can't be satisfied with that write your won exception classes and invoke point 5.
I don't know which APIs you worked with, but those that I use don't force me to put 40% of my code into try-catch statements.
Always write code which can't be read or figured out unless you have auto complete, intellisense and other IDE goodies.
I can see how autocomplete would make it easier to write code, but to read it? Nope, Java code is perfectly readable in a plain old editor, actually. But it's definitely not as easy to write using a plain old editor, to be perfectly honest.
Use XML as much as you can so that you can dogfood your insanity even more.
I agree with this one. XML madness needs to stop. Fortunately, it seems to be stopping already. For example, most modern DI frameworks don't require you to write your configuration in an XML file.
Make the code so verbose that even simple programs sound like very complicated ones that can be written only by you.
Verbosity is an unfortunate trait of Java. But nobody writes code so that it would seem that only they can write it. If they do, that's called over-engineering and is, believe it or not, considered to be a bad practice.
[1] http://www.dustinmartin.net/2009/10/getters-and-setters-in-s...
Design pattern: I wouldn't be surprised if they are used too much.
Love towards getters/setters. Not the fault of the language, that's just bad practice. And no, you don't have encapsulation. Rarely, getters and setters do some verifications or filtering, but most of the time they don't, in which case you just have raw access to the object's mutable state. Be honest and make your variable public, that's less code. If you ever need to put checks or filters (almost never), then just refactor.
Long names: my jury is out for this one. In your example, I'd say the overly long name comes from an overly narrow functionality. For something that specific, you should use a description string.
Try/Catch: If your QA section tells you to check all exceptions, you may have quite a bit of those statements. (Also, maybe he included the body of the try/catch statements?)
Code that requires an advanced IDE to approach: a good IDE let you manage and tolerate higher levels of complexity. You'll also be less encouraged to simplify your code. You may not even notice when it becomes too complex for a humble vi user.
Verbose code: People tend to stop as soon as it works, without simplify their code further. It's not the language's fault, nor is it done on purpose. I hate when I see code like that. I often have to apply various correctness preserving transformations before I stand a chance at understanding it.
"a good IDE let you manage and tolerate higher levels of complexity." - Which is a good thing.
"You'll also be less encouraged to simplify your code." - Simplifying code is a discipline that can just as easily be avoided by people writing in Notepad. Simpler code comes with experience, not through tool abstinence.
"You may not even notice when it becomes too complex for a humble vi user." - If your tools don't get the job done, switch tools. If you find yourself saying "My tools would work fine if only OTHER PEOPLE would ...", you're using the wrong tools.
2) I agree
3) Thinking about it… you're probably right.
4) I agree
My point was, IDEs are double edged. Like debuggers. Without such fancy tools, you are forced to think before you write nonsense to the compiler. Good programmers will do the thinking anyway, but many others need some "encouragement". Optimizing for good programmers is a good heuristic, but sometimes, you also need to prevent the bad ones to make too many mistakes.
A similar argument can be made about functional languages: I can avoid side effects in C++ just fine, and I mostly do because its plain simpler most of the time. But many programmers need at least an Ocaml straight jacket.
However, even then I think it's not quite so simple. In most cases simply putting in a text description for the exception would be sufficient, given language support when catching should it be needed. But if the application at a higher level needs to know what went wrong in order to inform the user in a friendly way, you'd need some kind of mapping between the exception and what to display on the screen. And if your app is internationalized, simply displaying the description of the exception would be even more problematic, especially if you have the same exception being thrown from different places (in which case you need to make sure none of their descriptions differ by even a single character lest you destroy the i18n mapping). I couldn't say definitely that subclassed exceptions would be superior in this case, but one could make an argument in favor.
"PtgSL. PatagonianSeaLion. Which of these names describes the class better?"
Now whatever you do, long names should be reserved to infrequently used things. If a long name litters your program, you'd better use a shorter one. Conversely, short names belong to frequently used things. This is because a long name is a cognitive burden. On the other hand, one needs to learn what is behind each label, and don't want to forget it.
Fortunately, good long names (such as "PatagonianSeaLion") tend to be about very specific things. So specific that you rarely need them. But there can be bad long names, like "System_Out_Println". Something this pervasive should never have such a long name.
Zen tip: When there is a reason for a bad thing, that is called a bad reason.
Some automatic tools could help (although I doubt thay'll work on DBase III): Static analysis to see what's there, version control to start at the top and log your way through and be able to rollback to a previous working version.
But it's at the very least weeks of pain.
If you can keep all interactions of your app in your head, then it doesn't matter how badly designed the code is. Architecture is only important because it makes it possible to change things without understanding the whole thing.
And complete understanding with bad architecture beats good architecture with incomplete understanding.
And for that you need to understand the architecture, ofcourse.
You always need to understand something_(n-1) to understand something_n, ad infinitum
Originally by Takeo Kanade. ( http://www.cs.cmu.edu/interviews/kanade/ )
CS programs should be burnt to the ground. And in their place we should build up three separate things. First, software trade schools that are actually good (e.g. not ITT). Second, for reals software engineering majors at colleges, that are heavy on things like practical programming, tools (version control, issue tracking, automated build systems), refactoring, teach multiple languages (javascript, python, ruby, SQL, etc.), and only delve into theoretical underpinnings as warranted (compare electrical engineering vs. physics programs). Third, legitimate Computer Science programs that are contractually limited to about 5% of the current CS student capacity for at least the next two decades and which teach a very mathematics heavy and science focused CS program and have zero expectation that the graduates of the program will go on to write software in industry after graduation.
All those software tradespeople can do the jobs that require connecting pieces together. A tradesperson can be taught enough to know which algorithms to apply in which situations.
Computer scientists become academics inventing new algorithms, and analyzing and perfecting the ad hoc algorithms created by software tradespeople.
Aside: I wish the English language would just decide that "man" and "men", when used as part of a compound word like "tradesman", is gender neutral.
> All those software tradespeople can do the jobs that require connecting pieces together. A tradesperson can be taught enough to know which algorithms to apply in which situations.
Is that a circular argument? If we had enough people cranking out smart enough algorithms (like the automatic programming example upthread), would we need so many tradespeople?
If you go ahead and blindly change your bubble sorts for quick sorts then at best you may be wasting time doing something that has no effect on performance, and at worst you may be making your program slower.
Quicksort is slower than bubble sort for nearly sorted input after all.
With superficial knowledge you might stick to certain rules without really understanding them. That bubble sort example you gave is a perfect example.
A profiler is not good at speed up an algorithm, but a profiler is THE tool you need to speed up a system, and most optimization in the wild is system optimization, not algorithm optimization.
This particular example is a perfect case study in how excessive focus on computer science can lead you astray in software engineering.
Most coding doesn't require it - certainly not the kind of thing that most large companies want coders for.
The problem is not that Universities produce poor CS majors. They don't. The problem is that everyone else expects that a CS major is going to be a good commercial developer. Some are, but that's just the odds.
If you want to be a programmer, do an software ENGINEERING degree. It deals with the practical issues and you actually do lots of real, actual, programming. Or go to a media design school, where you do lots of actual web programming.
Expecting every CS major to be a great programmer, is like expecting every Physics major to be a good baseball player. Sure, (s)he knows the optimal angle to strike ball to achieve a home run, but actually doing so requires a whole of experience and real world context.
(PS. I have a physics degree, and I'm a software developer, not a baseball player. But that's because I programmed a lot for fun and profit before going to university, and CS seemed like a big backwards step. And physics is more fun.)
If you hold an opinion, look at the evidence backing it up. If it's not strong, reduce your confidence. Or even better: Gather evidence, then form your opinion. And remember: Anecdotes don't count. I wish more of my colleagues would do this, but it seems most of them haven't heard of very many software-related studies.
If you want to learn more, I recommend http://www.neverworkintheory.org/ as a starting point. After reading some papers, you'll be surprised how limited our evidence-based knowledge is. Looking at software engineering studies made me realize that I'm not allowed to poke fun at psychology anymore. Even that field is more evidence-based than ours.
It's a mistake to introduce new programmers to OOP before they understand the basics
of breaking down problems and turning the solutions into code.
2) Early on, you learn that ints and chars and floats and some_structs are fundamentally different data types. Then suddenly you're told that int%s, char%s, float%s, some_struct%s, and even void%s are fundamentally the same. Huh?
3) The fact that C uses % both to create a pointer and to dereference. These are conflicting meanings, and the unrelatedness of those two concepts is not sufficiently explained.
EDIT: Silly HN parser. I now replaced asterisks with %s.
int *i;
int a;
int *i;
Makes sense once you understand it, but I can definitely understand how a beginner would find it confusing.
Now, algorithms can be difficult to understand and many of them use pointers... Do some people confuse the 2?
To tie this back in to the beginning of the thread, being able to understand indirection (such as in pointers and algorithms) is a far more important skill than understanding OOP.
I haven't heard anyone argue against it, but for a very long time every single curriculum out there started with either Java or C++, and had to talk about object quite early on (you can't print "hello world" in Java without it, and you can't use C++ streams either).
And arguments I've heard about that practice as a proxy for your question, indicate that it is (or at least, was between the late '90s and the late '00s) controversial.
Having seen 3000-line single-class Java monstrosities, I can assure you that this is not only true in theory.
CS should be offered as a major by itself. All the most interesting stuff is CS-specific with indirect applications. Working on something like automatic programming is far more exciting than working on biology or art or what have you. (I can't think of anything more awesome or more CS-only than automatic programming.)
It is a mistake to introduce programmers to OOP.
A complex compiler is awesome. A sufficiently smart compiler may be a myth, but it is a utopian myth; we should strive for it. However, I would take it even further: program synthesis is better still. I'm in the business of telling the computer what to do, not how to do it, so there should be no obvious but unnecessary correspondence between what I write and what the computer executes--they just have to have the same semantics.
You shouldn't be allowed to write a library unless you have a thorough understanding of programming languages and some relevant math. There is always relevant math. Your functions should be accompanied by useful and verifiable laws others can depend on. Or maybe everyone should be encouraged to write libraries regardless of skill level and then the libraries could be ranked a posteriori. Any other guidelines make less sense.
Pretty code is readable and readable code is pretty. If you can render your code as a nice pdf and distribute it as a paper, it's about as readable as it will ever be. Even if you can't, remember that aesthetics aren't random--there is a reason why pretty code is pretty.
Purely functional programming is a straw man. Even Haskell lets you write code that at least acts impure. Haskell is a local maximum. On the other hand: a purely functional spec that the computer uses to generate a potentially impure program should work. But I've already talked about that :P.
I don't know what a "software engineering mindeset" is. It sounds like something a manager would say. Don't do stuff a manager would say. This is unfair to good managers but still a useful guideline. Have as much fun as you can unless people's lives are on the line.
I should note that I don't even think all these opinions are true. But a belief does not have to be true to be useful. If I could boil it down to a single sentence, it would probably be: math and CS theory aren't scary and you should reject conventional "wisdom". But that would be somewhat cheap--two independent and rather unrelated clauses joined with "and" may as well be two sentences :P.
Also, there's something very appealing about throwing out intentionally extreme opinions. I can certainly see why this guy keeps on writing his blog.
We're not there yet.
I'll restate this: "Every programmer should know more than a little bit about typography". I was amazed at how much learning design fundamentals improved me as a programmer. I've learned to communicate through code much more effectively than I ever had before. Thinking about grouping, spacing, and the like leads not only to more readable ("pretty") code but usually more efficient code as well.
Oh yea, and inserting breakpoints and print statements not only take months, but also change your code in a case-by-case fashion.
It's like reverse engineering for masochists.
- Computer science should only be offered as a minor. Good point, assuming that you don't consider theory an end in it self. I consider theory mind opening, even if it doesn't make money per se.
- It's a mistake to introduce new programmers to OOP before they understand the basics.. Fully agree.
- You shouldn't be allowed to write a library for use by other people until you have ten years of programming under your belt. Another good point. Writing a good library (or designing an API) is one of the more challenging things you can do, because it requires a good understanding of the problem, simple design, and sensitivity to conventions (which only come from long experience).
- Superficially ugly code is irrelevant. Somewhat true, but the main problem with ugly code is that nobody wants to touch it. So if its functionally correct, readability remains irrelevant until you need to make a change.
- Purely functional programming doesn't work Agree, in the sense that there exists problems where a purely functional approach is not the best. I think you can run pretty far with purely functional though.
- A software engineering mindset can prevent you from making great things. Strictly true, but it should be said that the opposite mindset will eventually destroy the things you made.
I disagree with this one. Clear, readable code is... clear and readable. It's like not bothering to format text in a textbook because "the meat of the matter is in there, so who cares?"
Of course substance is more important than style in programming, but style also helps and is it really that much effort to make sure your code is readable for the next guy who comes along?
I just wish all languages had this kind of support.
The other nice thing about automated formatting is that everyone can edit code in their preferred format, as long as they convert it back before checking in (for readable diffs)
As an autocrat, I'd also add another few, what I conceive to be contemporaneously controversial point of view, to the discussion:
* Code Coverage matters. Dead code is broken code. Always.
* Programming is Always in Service To The User. The User is the only way your creative, artistic, amazing, junk of spaghetti-code crud, is going to ever get Used. Use is where your software is alive. Non-use = Dead. Thus, the USER is YOUR MASTER. Serve them.
* Pretty tools are one thing, ugly tools another thing entirely. NO! WRONG! ALL TOOLS ARE TOOLS. Use what works. If you're using something because you want to, even though it sort of doesn't work, its no longer a tool, but instead an .. ingredient .. of something. Something else, perhaps something creative. Do that shit on your own time: use the tools which work, at work.
* Discussion is the only way things ever get resolved. If you hate on something about someone, discussion is the only way the problem will ever get solved, ever. Ignoring something and being afraid to discuss really secretly means 'do not want' to solve the problem. Even vile words are still yet but words, words eventually work it out. Developers who do not use words are not the scribes they're meant to be ..
The way university is structured, it would be nearly impossible to minor in CS and learn anything low-level or advanced like computer architecture or operating systems. Perhaps you can push out a desirable employee with a practical knowledge of programming, but could you really educate true "software engineers" via a minor?
AFAIK, automated pretty formatter is not controversial. I wonder if your opinion about pure functional language is controversial.
Thank you for bringing to light some real controversy. In particular the old one about compiler optimisation. See for reference http://books.google.fr/books?id=6kHs4s-79bkC&pg=PA43&...
Programming skills are important, but a deep knowledge of the problem domain is sometimes far more critical to being able to actually solve the problem.
[0] http://www.codinghorror.com/blog/2012/06/the-php-singularity...
stunning.
(how's that for controversial?)
I'd rather work with someone who has five years of experience than someone with one year of experience repeated twenty times.
(I understand the combined experience thing isn't perfect logic but that's not really my point.)
Just keep in mind that a library takes a surprising amount of time to make "good enough".
The most frequent sin in library design is to knock something up in a weekend, check that it sorta works for you, and then consider it done.
On the "good enough" aspect the least I try to do is to unit test everything I publish.
That isn't controversial nor is it an opinion. It's just the truth. Purely functional code has no side effects. The entire point of a program is to have side effects.
This is one of several ways that Haskell worked before there was an IO monad [1], all allowing pure functional useful programmes.
[1] S. Peyton Jones. Tackling the awkward squad: monadic input/output, concurrency, exceptions, and foreign-language calls in haskell. Technical report, Microsoft Research, Cambridge, 2010.
My purely functional programs actually do things! Amazing that my ivory tower allows me to say "Hello World".
For non-interactive use (like a compiler) you could implement the program as a function taking and returning a string.
Of course, various programs (web apps) are IO-bound. And gcc -O0 should still outperform Python/Ruby/etc.