All 9 entries tagged Philosophy

View all 176 entries tagged Philosophy on Warwick Blogs | View entries tagged Philosophy at Technorati | There are no images tagged Philosophy on this blog

June 06, 2022

Wherfore art thou – variable names matter!

Sometimes, the simplest questions are the hardest to answer. For instance - what is the meaning of the word "the"? If you've never thought about this, have a go. If you'd prefer a non-grammar, or non-English specific example, try to describe the number '1'. Trickier than you'd think, isn't it?

But that is hardly relevant to programming, so lets look at today's deceptively simple question instead. Namely, how long should variable and function names be?

Some people might seem to have an answer to this question, such as "between one and five words, usually two or more". They are wrong. Any answer containing numbers is unhelpful and either sometimes wrong, or too broad to mean anything. For instance, "Supercalifragilisticexpialidocious" [typed from memory... excuse spelling] is one word, and is terrible. And "SolveWaveEquationSecondOrderWithLimiter" is seven and (in the right context) is quite good. "FlagSettingWhetherWeShouldPrintTheAnswerToScreenOrNot" is clearly terrible - but it is thoroughly descriptive.

So why do we struggle so much to answer this question? Because we are in a sort of "tug-of-war" between two competing interests, and which one pulls a little harder depends on a lot of things. Both ends of the rope are anchored in clarity: on the one hand we want maximum clarity for the function - a very descriptive name. This tends to favour longer names, with more detail. On the other hand, we want maximum clarity for the code as a whole - a name that doesn't take up too much mental space in a block of code. This tends to favour shorter names, with fewer, simpler words. To find our optimum, we must somehow balance these two.

A brief aside here into our absolute best weapon in this - make the two ends stop pulling. Find a way to make "descriptive" and "compact" the same thing. Specialised definitions of terms specific to a field of interest, i.e. jargon, really works in our favour here. We must use it with caution, because new jargon can be very jarring and difficult to master, but in general jargon terms are addressing exactly the same problem - descriptive yet compact terms. For instance, a word you may barely think of as jargon - "laptop". Let us expand this definition - perhaps we get "a portable, all-in-one computing device". But we still have some specialised terms here - what is "all-in-one"? What is "computing"? We could continue the expansion almost indefinitely. Or, we can simply use the term "laptop" and in most cases be perfectly understood.

It is important to flag here that word, "most". Is a tablet a laptop? If I strap a monitor to a tower PC and plug in a keyboard, do I have a "laptop"? It's going to depend on context whether those are reasonable (OK, the second one is extremely unlikely to be). But consider similar questions - "do you have a car?" "No, I have a van" - sensible answer, or irritating pedant? "We need to seat 5 people, who has a car?" "I have an MX5" - as it turns out, "car" doesn't always mean 5 seats.

Back from our aside now, new weapon in hand. Careful, selective use of jargon can make our names very descriptive, while remaining compact. If the jargon we use is not universal, we can provide a glossary or add context in our docs. We should be very very careful about making up our own jargon here, because unfamiliar terms can lead to misunderstanding, but if your field provides words, exploit them.

One last important point - sometimes an otherwise optimal name is not useful due to ambiguity. This can be typographical - we should never have names that differ only in characters like '1' and 'l' or differ only in the case (small or CAPITAL) of the letters. Ambiguity can be similarity with some other function name - imagine we somehow tried to have "GetDiscreteName" and "GetDiscreetName" - could you even tell those apart? Ambiguity is an enemy of clarity - avoid it.

OK, actually there is one final point. Clearly redundancy can only harm our compactness of names. But it does have some applications. We might want two functions with very similar names, but different parameters - "SolveTypeAEquation(a, b, c)" and "SolveTypeAEquationNormalised(a, b, param)" for instance. If we can't design the code to make these less confusing, we might deliberately make one name less individually clear, if it makes its usage clearer. So perhaps the second one becomes "SolveTypeAEquationNormalisedWithParam(a, b, param)" which repeats information already in the signature (redundancy), but helps us keep in mind which function we want.

Short posts like this rarely have conclusions, because everything has already been said, usually repeatedly. Since repetition is another form of redundancy that actually works though, lets restate our key point:

We must balance the demands of clarity between our functions and our wider code and try and find a name which is BOTH descriptive and also short and easy to handle. Sometimes one of these will be a bit more important, sometimes the other.


March 18, 2022

Getting a little testy

Testing code is essential to knowing if it works. But how do you know what to test? How do you know you've done enough?

Let's be clear to start here, "testing" as we think of it is some form of comparing a code answer to a predicted one, making sure the code "gets it right" or meets expectations. If there is a canonical "right answer" then it's fairly easy to do - if not then things get difficult, if not impossible, but correctness is the goal.

What to test? Well, all of it, of course! We want to know the entire program works, from each function, to the entire chain. We want to know it all hangs together correctly and does what it should.

So when have you done enough testing to be sure? Almost certainly never. In a non trivial program there is almost always more you could test. Anything which takes a user input has almost infinite possibilities. Anything which can be asked to repeat a task for as long as you like has literally infinite possibilities. Can we test them all? Of course not!

Aside - Testing Everything

Why did we say user input is only almost infinite? Well, all numbers in the computer have a fixed number of bits for storage, which means there's a strictly fixed number of numbers. Technically, for a lot of problems, we really can test absolutely every input. In practice, we can't afford the time and anyway, how do we know what the right answer is without solving the problem completely.

Picking random things to check is an idea that's used sometimes, as are "fuzzers", which aim to try a wide range of correct and incorrect inputs to find errors. But these are also costly to perform, and still require us to know the right answer to be really useful. They can find crashes and other "always wrong" behaviour though.

Lastly, in lots of codes, we aren't expected to protect users from themselves, so entering an obviously silly value (like a person's height of 15 ft) needn't give a sensible answer. We can fall back to "Garbage In Garbage Out" to excuse our errors. But our users might be a lot happier if we don't, and in important circumstances we wont be allowed to either.

Back to the Grind

That all sounds rather dreary - writing good tests is hard, and we're saying they're never enough, so why bother? Well, lets back off for a minute and think about this. How do we turn an "infinite" space into a tractable one? Well, we have to make it smaller. We have to impose restrictions, and we have to break links and make more things independent. The smaller the space we need to test, the more completely we can cover it.

Unit testing

Most people have probably heard of unit testing by now - testing individual functions in isolation. It seems like if you do this then you have tested everything, but this is not true. What happens if you call this function followed by that one (interdependency)? What happens if you call this function before you did this other thing (violation of preconditions)?

Unit testing is not the solution! Unit testing is the GOAL!

If we could reliably say that all of the units working means the program works, then we can completely test our program, which is amazing! But in reality, we can't decouple all the bits to that extent, because our program is a chain of actions - they will always be coupled because the next action occurs in the arena set up by the preceeding ones. But we have to try!

Meeting the Goal

So we have to design, architect and write our programs to decouple as much as possible, otherwise we can't understand them, struggle to reason about them, and are forced to try and test so much we will certainly fail. A lot of advice is given with this sort of "decoupling" in mind - making sure the ways in which one part of a program affects another are:

  1. as few as possible
  2. as obvious as possible
  3. as thin/minimal as possible

What sorts of things do we do?

  • Avoid global variables as much as possible as anything which touches them is implicitly coupled
  • C++ statics are a form of global, as are public Fortran module variables (in most contexts)
  • Similarly, restrict the scope of everything to be as small as we can to reduce coupling of things between and inside our functions
    • C++ namespaces, Fortran private module variables, Python "private" variables (starting with an underscore, although not enforced in any way by the language)
  • Avoid side-effects from functions where possible, as these produce coupling. Certainly avoid unexpected side effects (a "getter" function should not make changes)
    • Fortran offers PURE functions explicitly. C++ constexpr is a more complicated, but related idea
  • Use language features to limit/indicate where things can change. Flag fixed things as constant.
    • Use PARAMETER, const etc freely
    • In Fortran, use INTENT, in C/C++ make function arguments const etc
    • In C++ try for const correctness, and make member functions const where possible. Use const references for things you only want to read
  • Reduce the "branchiness" (cyclomatic complexity) of code. More paths means more to test.
  • Keep "re-entrancy" in mind - a function which depends only on its arguments, doesn't change it's arguments, and has no side effects can be called over and over and always give the same answer. It can be interrupted/stopped and started afresh and still give the same answer. This is the ultimate unit-test friendly function!

Overall, we are trying to break our code into separate parts, making the "seams"between parts small or narrow. If we drew a graph of the things in our code and denote links between them with lines, we want blobs connected with few lines wherever possible.

Every link is something we can't check via unit testing, and unit testing is king because it is possible to see that we have checked every function, although we can never see if we have checked every possible behaviour within the function, even with code coverage tools.

Aside - Writing useful unit tests

Even if everything is unit-testing amenable, it doesn't make it actually easy. There's a few things to keep in mind when writing the tests too.

  • Poke at the edge-cases. If an argument of '3' works, then '6' probably will too, but will '0'? What about a very large number? Even (a+b)/2 for an average will break down if a + b overflows! A negative number? Look for the edges where behaviour might change, and test extra hard there
  • If your function is known only to work for certain inputs, make them preconditions (things which must be true). This can be done either in documentation, or using things like assert
  • Check for consistency. If you mutate an object, check it is still a valid one at the end
  • Try to come up with things you didn't think about when you wrote the function, or that no "sensible caller" would do
  • Check that things fail! If you have error states or exceptions, make sure these do occur under error conditions

Dealing with the Rest

If we could actually produce a piece of code with no globals, no side-effects and every function fully re-entrant, unit tests would be sufficient to check everything. Sadly, this is impossible. Printing to screen is a side-effect. All useful code has some side effects somewhere. Mutating an argument makes a function non-reentrant (we'd have to make a copy and return the new one instead, and that has costs). So we seem doomed to fail.

But that is OK. We said unit-testing is a goal, something we're trying to make as useful as possible. We can do other things to make up for our shortfalls, and we definitely should, even if we think we got everything. Remember, all of those bullets are about minimising the places we violate them, minimising the chances of emergent things happening when we plug functions together.

We need to check actual paths through our entire code (integration testing). We need to check that things we fixed previously still work (regression testing). We probably want to run some "real" scenarios, or get a colleague to do so (sorta beta testing). These are hard, and we might mostly do them by just running our program with our sceptical hats on and watching for any suspicious results. This is not a bad start, as long as we keep in mind its limitations.

Takeaway

What's the takeaway point here? Unit Testing works well on code that is written with the above points in mind. Those points also make our code easier to understand and reason about, meaning we're much less likely to make mistakes. Honestly, sometimes writing the code to be testable has already gained us so much that the tests themselves are only proving what we already know. Code where we're likely to have written bugs is unlikely to fail our unit tests - the errors will run deeper than that.

So don't get caught up in trying to jam tests into your existing code if that proves difficult. You won't gain nearly as much as by rewriting it to be test friendly, and then you almost get your tests for free. And if you can't do that, sadly unit tests might not get you much more than a false sense of security anyway.


March 02, 2022

Learning to Program

We (Warwick RSE) love quotes, and we love analogies. We do always caution against letting your analogies leak, by supposing properties of the model will necessarily apply to reality, but nontheless, a carefully chosen story can illustrate a concept like nothing else. This post discusses some of my favorite analogies for the process of learning to program - which lasts one's entire programming life, by the way. We're never done learning.

Jigsaw Puzzles

Learning to program is a lot like trying to complete a jigsaw puzzle without the picture on the box. At first, you have a blank table and a heap of pieces. Perhaps some of the pieces have snippets you can understand already - writing perhaps, or small pictures. You know what these depict, but not where they might fit overall. You start by gathering pieces that somehow fit together - edges perhaps, or writing, or small amounts of distinctive colours. You build up little blocks (build understanding of isolated concepts) and start to fit them together. As the picture builds up, new pieces fit in far more easily - you use the existing picture to help you understand what they mean. Of course, the programming puzzle is never finished, which is where this analogy starts to leak.

Jigsaw image

I particularly like this one for two reasons. Firstly, one really does tend to build little isolated mini-pictures, and move them around the table until they start to connect together, both in jigsaws and in programming. Secondly, occasionally, you have these little moments - a piece that seemed to be just some weird lines fits into place and you say "oh is THAT what that is!". One gets the same thing when programming - "oh, is that how that works!" or "is that what that means".

Personally, a lot of these moments then become "Why didn't anybody SAY that!". This was the motivation behind our blog series of "Well I Never Knew That" - all those little things that make so much sense once you know.

The other motivation for the WINKT series is better illustrated using a different analogy - hence why we like to have plenty on hand.

A Walk in the Woods

Another way to look at the learning process, is as a process of exploring some new terrain, building your mental map of it. At first, you mostly stick to obvious paths, slowly venturing further and further. You find landmarks and add them to your map. Some paths turn out to link up to other paths, and you start to assemble a true connected picture of how things fit together. Yet still there is terrain just off the paths you haven't gone into. There could be anything out there, just outside your view. Even as you venture further and deeper into the woods, you must also make sure to look around you, and truly understand the areas you've already walked through, and how they connect together.

My badly done keynote picture is below. The paths are thin lines, the red fuzzy marks show those we have explored and the grey shading shows the areas we've actually been into and now "understand". Eventually we notice that the two paths in the middle are joined - a nice "Aha" moment of clarity ensues. But much of our map remains a mystery.
Schematic map with coverage marked

And that is one of the reasons I really like this analogy - all of the things you still don't know - all that unexplored terrain. This too is very like the learning process - you are able to use things once you have walked their paths, even though there may be details about them you don't understand. On the one hand, this is very powerful. On the other, it can lead to "magic incantations" - things you can repeat without actually understanding. This is inevitable in a task this complex, but it is important to go back and understand eventually. Don't let your map get too thin and don't always stick to the paths, or when you do step off them, you'll be lost.

This was the other main motivation behind our WINKT series - the moments when things you do without thinking suddenly make sense, and a new chunk of terrain gets filled in. Like approaching the same bit of terrain from a new angle, or discovering that two paths join, you gain a new perspective, and ideally, this new perspective is simpler than before. Rather than two paths, you see there is only one. Rather than two mysterious houses deep in the woods, you see there is one, seen from two angles.


Take Away Point

If you take one thing away from this post, make it this: the more you work on fitting the puzzle pieces together, the clearer things become. Rather than brown and green and blue and a pile of mysterious shapes, eventually you just have a horse.

And one final thing, if we may stretch the jigsaw analogy a little further: just because a piece seems to fit, doesn't mean it's always the right piece...


January 09, 2020

Finding The Solution

New Year, New blog post. Just a short one this time, following on my my post on FizzBuzza few months ago. Even a problem as simple as that can be solved in myriad ways, and as I program more and in more languages, I find myself less often wondering how I can solve a problem, and more often how I should.

Most of the languages I work with let me solve problems using the basic command structures, and, as I wrote about last time in The X-Y problemit can be hard, but is vitally important, not to get confused by your partial solution and miss a better one.

Recently I've been learning Perl to do some complex text-processing, and find it to be a drastically different way of thinking to my C/Fortran background. It's tricky to think in terms of text matches and substitutions when I am so used to thinking of the position of each character in a string and working in terms of "index-of-character-X plus 1" (similar to working in terms of for-each loops when one has only used for). For the processing being done, the proper Perl solution is much shorter, easier to understand, etc, although it takes me a bit longer to produce initially.

A recent Stack Exchange post I saw had somebody asking why his boss didn't appreciate his brilliant coding techniques, because design patternswere second nature to him, and his boss wanted to use far simpler solutions. He probably came back to Earth with a bit of a bump when it was firmly pointed out that "patterns being second nature" was actually a bad thing, because it rather sounded like he trotted out the first "pattern" he could think of, instead of actually thinking about the problem he was solving. Nothing wrong with the patterns themselves, but critical thinking is required to decide whether they are suitable, optimal etc.

The other common mistake people make is demonstrated in Terry Pratchett's description of "Death's Swing" (e.g. https://en.wikipedia.org/wiki/Death_(Discworld)#Home), which mirrors the Sunk Costs fallacy. Trying to build a swing for his Grandaughter, the character of Death plows forward inspite of all problems. He hangs the swing from the two strongest branches. These being on opposite sides of the tree, he cuts away the trunk, shores it up and ... This can easily happen when programming and the trick is never to be afraid to throw away (or file away for future use) a solution, even a good one which took a long time, if it stops fitting the problem. In Death's case, it is less because he is unwilling to throw away the work already done, and more an issue of very linear thinking, but the effect is the same.

Hopefully this is already obvious, and you always think before you code, happily refactor or rework your own code, and have an ever-growing solution bank to call on. I suspect very few people are willing or able to throw away all the false starts they probably should though. Just keep in mind that there is a crucial difference between "what solves the immediate problem" and "the code I should probably actually write", and strive for the latter.


December 05, 2019

The "XY Problem" or how to ask so somebody can answer

There's a lot of things that I feel ought to have pithy names or words, and this is one of them. The classic "XY problem" goes: Person wants to solve problem X. They don't know how. They think about it, and conclude that they will need to do "Y" as part of the solution. They don't know how to do this either. So they ask for help, but omit, or forget, to mention the larger problem. This can lead to bad or not-applicable answers, and an inability of the (presumably) skilled responders to explain how to do X. Since the original asker doesn't know how to do X, there's a good chance "Y" was a poor approach, or irrelevant, or harder than it needs to be etc.

I managed to commit a classic XY swap today while knocking up a very simple bash script. I wanted to take a filename, and strip its extension off. I knew this was going to be ".f90" in context, so I searched for how to strip the last 4 characters of a string in bash. I found this questionand used the substring solution given, before reading a bit more of the page and realising that I (and the original asker) were asking the wrong question. We both actually wanted this solutionto remove the dot and the extension. Funnily enough, this is also the archetype XY problem, as described at e.g. hereor here.

In this case, it isn't a big problem. Both solutions work, and for my actual use I don't need to gracefully handle a string without the ".f90" extension at all. But in more "interesting" (i.e potentially serious and causing of harm) cases on some of the forums I read (e.g. DIY, powerlifting, cybersecurity) the asked question hides a serious misunderstanding and goes off down an unhelpful path. For example, thankfully this guyasked his question openly (why are is breakers tripping after he removed a cooker fan and tied the live to the neutral wire) even if it was rather too late! Somewhere in the hundreds of questions about tripping breakers on that forum, there is probably somebody who's done something just as bad, but hidden it.

Another amusing type is the one where the question as asked can be answered effectively, but the asker would be surprised by the extra information. For instance (summarised from a real exchange):

A: I need a workout program that takes 3 hours every day!
B: Here are some! They're awesome!
C: Hold on, why do you want this?
A: Because I'm bored and I need to occupy 3 hours.
C: Jeepers! Find a hobby buddy! Forcibly spending a fixed amount of time is not going to lead to productive training...


But it Works, Doesn't it?

Even when the question isn't the right one, it is possible to get a solution which "works". But assuming you're trying to improve your programming skills, there's a lot wrong with "good enough" and the "wrong" solutions often have a whole host of problems:

  • They're too specific. For example, the substring solution for my file-extension problem works, but it's less general than it should be. It only works for 3 character extensions (plus one for the '.') and gets the wrong answer if there is no extension, wheras the actual solution works in both of those cases.
  • They're misleading when read back. Again, with the file-extension problem it's not clear what I'm doing by taking 4 characters off - the purpose is much clearer if I split on the dot.
  • They're not actually any quicker/simpler than the "proper" solution. The file-extension thing is a good illustration again - bash substrings are kinda inelegant, while pattern substitutions aren't. The "proper" solution is shorter and much clearer in general.
  • You still don't know how to solve the "actual problem" where you could have learned a much broader-applicable thing and improved your problem solving ability. In a lot of cases, you end up writing "Fortran in every language" rather than actually learning the approch of the one you're in. Again, the file-extension example is a good one. Doing it with a string slice is something familiar to me from other languages, but not a good move in bash, and it makes the problem more fiddly (I need to find the string length because of how substringing works).
  • Last and perhaps most egregiously, it hides from YOU how well you understand what you're doing, and how hard it actually is. My hacky solutions can end up being far more complex than they need to be, or they can end up making a genuinely hard problem look simple, because they don't actually solve it. See, for example, the difference between a genuine AI chatbot and the ELIZA model.

So what's the solution?

Mostly, when asking a question, try to ask the actual real one. This is easy when asking a question (in person or in text), but gets a bit tricky when using a search engine, since you're having to break things down into just a few key words. But the general idea is to look for the overall problem, as well as the partial solution you've thought up, to think very hard about what that actual problem is, taking a step back if required, and to not get too attached to your approach.

Always bear in mind that a complete change of tactic might be required to actually achieve your goal, and that you might not have quite nailed down what that goal is yet. And keep in mind the quote from Henry Ford:

Failure is simply the opportunity to begin again, this time more intelligently.

or as I've heard it paraphrased but can't find a source for:

Finding out you're wrong is great, because you get the chance to become more right.


October 10, 2019

The Basics aren't so Basic

Sorry for the long break! We've been busy with the start of term and busy expanding our training material (link). This week I am just going to talk about something that you should always keep in mind, not just with programming and computers but with a whole bunch of things, and that is, what does it mean to say something is "basic".

There is a quote often attributed to Einstein, although not directly traceable to himwhich goes

Common sense is the collection of prejudices acquired by age eighteen.

Whether the origin is real or not, it's often true that what people think is simple, or "just common sense" is only so because of their background. To somebody who cut teeth on a BBC Micro, programming might seem super BASIC.

Jokes aside, you will probably keep coming across things that are super-basic and feeling a bit awkward that they somehow escaped you until now. Especially if you have learned things in the usual manner, i.e. by necessity, it is very easy to miss some of the basics. You can find yourself doing really quite advanced things, while not knowing something that "everybody else" seems to. This is normal. It is not beneficial, but it is perfectly normal. Frankly, in computing and programming there is a vast, vast sea of "basics", and no matter how much you learn there always seems to be more.

When I were a Lad

When I was a PhD student, I was happily using 'ssh' to login to remote machines, but I would always type out the whole host spec, such as "username@machinename.blah". I remember feeling a bit dumb when my supervisor pointed out that I didn't need the "username", and he thought this was somehow basic and obvious. I was frankly a little bit irritated because nobody told me! How was I meant to know?

"Simplicity is the final achievement."

(Quote from Frederic Chopin)

Moreover, just because something is "basic" doesn't mean it is simple. In fact, Merriam-Webster's definitionof the adjective "basic", while perhaps a bit unhelpfully recursive does not say simple anywhere. That thing with the username isn't so simple. It's fundamental, sure, but it's not simple!

Years later, I am still regularly coming across things that are "basic" that I have never encountered before. The whole "learning how to program" thing is far more of a helix than a road. You come across fundamental things all the time, some for the first time, some repeatedly, and often you can understand them better every time. Eventually, you find them simple. Sometimes they feel even elegant, because they arise so smoothly from the things you do know, or perhaps even seem so obviously "the only way it could be".

This is most of the motivation for our "WINKT" blog post series. These are fundamental, mostly "basic" things, but they're mostly not things you could usefully be told about the first go-around. Mostly, they are the basics of how the complicated things work. For example:

  • On the command line: if you use the '*' wildcard, when does this get expanded into the list of matches? Specifically, if you accidentally create a file called "-rf" in your home, and ran the command `rm *` to remove files, how much trouble would you be in? The answer is, _a lot_. * is processed first, by the shell, and unfortunately '-' comes first in the alphabet. You just ran the equivalent of `rm -rf *`. Ooops.


  • Any C/C++ programmers: if you use a variable which is undefined, what is it's value? If you said "whatever is in the associated memory beforehand", you're close, but wrong. An undefined variable is undefined behaviour - it can be given any value, including a different one each time it is accessed. Why? Because the standard says so. But who needs to know that? It is enough to know that its value is unreliable. Using your "basic" knowledge of the C memory model, you would likely guess the above, and it would never matter. [Disclaimer: this is one that I personally only learned a few weeks ago. It's absolutely fundamental, but not at all simple.]


  • For Fortran 2003 people: if you have a function-scoped ALLOCATABLE array, allocate it inside the function and forget to free it before the function exits, what happens? A memory leak? Nope! Fortran will helpfully deallocate the array on exit. If you didn't know this and freed everything yourself, there would never be a problem, but this one often surprises people.


  • For Python people: suppose you give a function a default argument, like `def func(arg, list_arg = []): ...` and suppose inside the function, list_arg gets filled with stuff. If you call the function twice without supplying list_arg, what do you get the second time? If you said "the combined contents from the first and second calls" you would be correct! The default arg. is an empty list, but it is the SAME empty list each time!

Take Aways

What's the point of all this rambling? Just that there is so much often classed as "the basics" that nobodycan know it all, and there is nothing so basic that you wouldn't do well to re-examine it anyway. It gets said all the time, but with computers there really are no stupid questions. Well, OK, there are some pretty stupid questions. But I have never seen one yet that wasn't worth thinking about.


Postscript: any suggestions for things that make you go "Well I Never Knew That"!, email us or comment! We can always use more


June 27, 2019

"And then it just clicked

A bit more of the general "philosophy of programming" today, based on a quote I found on the brilliant "C FAQ", currently here and hopefully there to stay. The quote is from Q 18.9b, on learning resources and says:

A word of warning: there is some excellent code out there to learn from, but there is plenty of truly bletcherous code, too. If you find yourself perusing some code which is scintillatingly clear and which accomplishes its task as easily as it ought to (if not more so), do learn everything you can from that code. But if you come across some code that is unmanageably confusing, that seems to be operating with ten bandaged thumbs and boxing gloves on, please do not imagine that that's the way it has to be; if nothing else, walk away from such code having learned only that you're not going to commit any such atrocities yourself.

It's a very good idea to read other people's code when learning - either completely in the wild, or in the form of snippets on sites like Stack Overflow. But always keep in mind that there is some truly terrible code out there, even in commercial packages. It's easy to fall into the trap of thinking that your problem is just "soooo hard" that it just can't be done elegantly, or readably, or even conforming to standards, at all, and mostly that just isn't the case.

Obviously, different people will find different things to be readable, clear, elegant etc. I have an Undergrad degree including Maths, and I rather like the shorthand notation of sums and sets and implication. To some people though, the symbols are simply "all Greek" (pun thoroughly intended). Writing things in a way everybody will find clear is thus often a losing battle. But there is one very very important thing to avoid at all possible costs : the anti-click.

The Click

The 'click' is that feeling of clarity when you jump from grasping the parts of the thing to understanding the whole. I get it a lot with things like Anagram word puzzles, or even song intro quizzes (the kind where you're meant to guess the title). One moment you're seeing a jumble of letters, trying to hold all dozen in your head, and the next you can 'see' the word it must be. Similarly, one moment you're following along the notes and words and the next the key lyric pops into your head and the song is obvious. Optical illusions do it too - and once you've seen it you struggle to "un-see" it again.

This is the 'click' and it's a real asset that you can develop with practice, by reading plenty of code, and writing plenty of code, until you can 'get the gist' of the thing easily.

Symptoms and Diagnosis

It also comes up a lot in debugging - you just 'get the hang' of the way an error pops up, and know what must have happened. That's why reading issue trackers or mailing lists can feel so confusing - the people answering the questions seem to have some mysterious knack for guessing what's wrong from the most un-intuitive errors.

For example, writing in C you may use a function like 'sqrt', and get a 'undefined symbol' error from the linker, which is saying that it can't work out what that function is. So, being logical, you go back and check that you have absolutely definitely included the "math.h" header, which absolutely definitely contains a sqrt function. What did you do wrong? Well probably you forgot to compile using the '-lm' flag, which means "link against the maths library", because that library contains a lot of compiled code (the code isn't in the math.h file). This error can seem very weird, but actually, once you know the cause, it's "obvious".

We had a go at writing a catalogue of programming bugs a while ago, which tried to sort of encapsulate the 'symptoms' of a bug (available here, PDF download). Any bugs, errors or omissions please let us know (rse {@} warwick.ac.uk). Once you see enough bugs, you'll probably start to 'click' and 'just know' what to look for.

The Anti-Click

I mentioned optical illusions up there, and how you can't seem to "unsee" them. Sometimes code has this same sort of thing - an illusory apparent action or cause which is actually not real. This can be awkward and dangerous, because you, and others, will struggle to "unsee" it. It's hard to give an example that isn't really contrived, so I shall fall back onto a classic bit of C: pointer declarations.

In C, you declare pointers using a '*', such as `int * p;`. This is fine. But what about `int * p, q;`? Is q a pointer? No, it's not. Some people argue that the "proper style" will prevent this anti-click, that is, writing that as `int *p, q` and associating closely the * and the p. This is surprisingly hotly debated (e.g. hereand here) with some really tempting anti-clicks (such as here- do read the answers, as the OP there is wrong) but ultimately does need careful thought. You're probably best using whatever you personally find clearest, and changing that only after consideration.

The golden rule - DO NOT write misleading code, docs etc etc. Make it wordy if you have to, but it's not worth the hassle of an obvious, yet wrong, interpretation. As a last resort, just put in a comment explaining what this does not mean, or does not do.

And the flip side - NEVER trust the click - always verify. Always step through line by line and be sure you're right!

Click Immunity

One last thing - there are people for whom some things just never do 'click'. It's very hard to describe, but easy to spot. It's sort of like trying to explain computers to that one relative or quadratic equations to somebody who can't do maths. They're clearly not stupid, but they just don't seem to 'get it'. You explain one thing, like how to print, and it works, but now they need to print from a different program and isn't it just obvious that you do the same thing?

Some people seem to have a knack even beyond that - they don't just "not get it", it's worse! Bug reports display this sometimes - somebody who always gives a lot of information but somehow never includes the vital pieces. They'll post a 100MB log file, but forget to mention which OS. They'll explain exactly what commands they ran but omit some vital parameter.

There's two sides to this - dealing with people who don't click, and dealing when you can't click.

Hell is Other People

Usually you only have one real option to deal with other people who don't get it - guide them through it. Mostly, it just takes longer, not happens never. You probably want to try explaining things different ways, because what feels neat and obvious to you isn't to them. The Socratic style is handy - ask them what they're trying to do, or what they feel they should do, rather than telling.

I just don't GET IT

I expect everybody, sooner or later, finds something that just doesn't click right. You don't get it - it's unintuitive, your gut feeling is always wrong etc. So what do you do? Treat yourself the same as you would somebody else - read different explanations until you find one that fits you. Ask yourself "what am I trying to achieve". Ask yourself "when I ask this question, what might somebody need to answer it" - and if in doubt, ask them! Stop assuming anything, and walk through every step of the problem.

Walk away and do something else - give your brain time to work things through. Never stop asking questions and always learn from the answers. You don't need to 'click' to do something, you just need to watch for wrong intuition. Keep notes - add comments in your code for you to come back to.

Remember - intution is a skill like any other. It can be wrong. And while you might not have a talent for this thing, but with effort, you can make it work.


May 30, 2019

Black and white

Quick 'philosophy of programming' entry this time. General solutions to common programming probelms are often called 'design patterns' (e.g. the wiki articleor the book which started the name). The idea of these is to have language independent (as far as possible) 'patterns', like clothing patterns, which can be tweaked to fit a specific situation. A lot of these patterns seem obvious, which is good, and since they're developed and tested by many people they can be very valuable in showing you questions you hadn't even thought of.

This weeks topic is perhaps too simple to really call a pattern, but it is a very useful thing to keep in mind when doing anything that deals with restricting function which exists, but should not be allowed. For example, forms which take user information often disallow anything except numbers in a 'telephone' field. A code I work on has a lot of user-specifiable options, but as the programmer I know that some are incompatible where it might not be obvious to the user - and I want to either warn or abort if these are used together.

There are two general approaches to things like this, and which you choose depends on many things. You have to maintain some kind of list to check against, but you can choose to use either the "blacklist" or the "whitelist". The former, the "blacklist" is a list of the things which aren't allowed, and anything not in the list is OK. The "whitelist" approach means keeping a list of the things which are allowed, and anything not in the list is excluded.

Sometimes the choice is fairly easy, because one method is a much, much simpler list. For instance, in the phone number example, it is fair easier to use the whitelist, allowing only '1234567890', but nothing else. If you try the other way, you might think to exclude letters, but what about Greek or Cyrillic characters? On the other hand, this is a source of deep annoyance if you forget any needed character - in the example I just gave, one could not put any spaces in, which is annoying, nor brackets or the '+' symbol.

A classic example of the poorly-thought out whitelist is in name fields which often exclude characters like the apostrophe, annoying the Scots and the Dutch for eternity. And what about accented letters, or the German ess-tsett. With a whitelist, you need to be sure you've caught everything, or people will be, rightly, upset. For a user-name on a website, and for a password, it is probably fine to allow any ASCII or Unicode character and set up your systems to handle them, leaving far less upset without any real cost to you.

On the other hand, with a blacklist, anything not forbidden is permitted. These are generally used in cases where certain characters have a function and so must be excluded, even if this annoys. So, for instance, in most programming languages variable names may not start with a number, nor contain a comment character.

Apart from the length of the lists, the two methods trade off this annoyance to your user, who must wait until you fix the omission (with a whitelist) against potential unknown failures and security risk (with a blacklist). Imagine the 'incompatible features' problem with both methods. If I use a whitelist, and forget to allow some pairing, my worst case is that I will likely be asked (somewhat irately) why X and Y can't be used together. I realise they can be, I update the code and I make a new release version to fix the omission and everybody is happy. If I use a blacklist and forget that some X and Y don't work together, my worst case is that one day I have to tell somebody that their last n years of research is all invalid, because the simulation they ran didn't work as expected, and since nothing actually went wrong they didn't know. Worse still, would be having to tell them that their fascinating effect is just a code error, and it's my fault.

In some cases, only one or other list type is really viable. For instance, virus scanners keep a list of 'tells' for malicious code, because even though they let things slip through until their lists update, they could never describe all of the 'allowed' code. App permissions (on better, more granular systems) are a whitelist - you give an app the permissions you choose, and only those.

So as a general rules of thumb:

  • If only one method is viable, obviously use that
  • If one or other list is going to be much much shorter, you have better chances of getting it right, so use that method
  • If it is really important not to let things slip through, use a carefully managed, kept up to date, whitelist. If possible, put it into a file or something, so that updates just require sending out new definitions, not modifying the entire code
  • If it's really important not to get accidental exclusions (false positives) use a, similarly carefully managed, kept up to date etc, blacklist
  • In some cases, combine the two. Programming languages generally have a set of allowed characters (a whitelist) and small blacklists for specific contexts such as the first character of a name.

As well as the literal 'blacklist' and 'whitelist' there is a more general principle here - do I selectively forbid, or selectively allow? Do I stop somebody doing this thing here, here and perhaps here, or do I only permit them to do it there and there. If you find the 'here's' or 'there's' proliferating, re-examine whether you're doing it the right way around. In safety or security critical situations, you almost always must allow only what is permitted. If you find yourself trying to plug up security holes with ever growing blacklists, you should probably change tack and think about what should be allowed instead.


November 28, 2018

Experience Errors to Excel

No, not the Microsoft spreadsheet program. Warwick RSE don't think Excel is evil, or that it makes errors more likely, but most of what we work on or talk about is beyond the point where Excel, Numbers or any other spreadsheet software is really efficient. Anyway, this post is going to be a little bit of programming and research philosophy, as well as some practical bits about how to make errors in your code and research less likely.

Tools of the Trade

First, a few easy rules to make serious errors a lot less likely.

Have some Standards

Most of the "Old Guard" programming languages, the ones you find in banking software or aeroplanes, have very strict standards dictating what is valid in the language. They are usually revised every 3-5 years, and controlled by some professional body, often ISO. A working compiler or interpreter must obey these standards or it is wrong. In practice, support for new versions of standards takes some time to develop, but this is usually well documented. Some major packages, such as MPI, also have standards. If your language (or library) has standards, follow them! If it doesn't have formal standards (e.g. Python), be as conservative as you can in what you assume. For instance, try the calculation 2/3 in Python 2 versus Python 3.

7 Ps

Proper Planning and Preparation Prevents Piss Poor Performance. Supposedly a British Army adage, and one of my Mum's favorites. Plan before starting! Work out what you're going to do. Prepare - read up on any new techniques, do some "Hello World" examples with any new packages etc. Check all the bits of your plan can work in practice - can they be fast enough? Is your chosen language/package a good choice? Can you handle the amounts of data/computation/other needed?

Borrow your Wheels

There is a crucial balance between reinventing the wheel everytime you write some code on the one hand, and importing a package for every non trivial task on the other. Neither of these are a good idea, and both tend to lead to more and worse errors. Borrow widely (with proper attribution), but don't code yourself into a morass of dependencies that you can never hope to test. Especially don't borrow from things that are no better tested or verified than your own code - this is a recipe for disaster.

Document Everything

It's very easy to assume some feature or limitation of a code snippet is "obvious" when you're writing it, only tocompletely forget about it when you come back. The only way to avoid this is to document. Do this a few ways.

Restrictions on a code snippet, that don't impact anything else, so don't fit either of the next two paragraphs, should be commented as near the line as possible. Try and make it so that the comment still makes sense even if the code changes a bit - e.g. don't use line numbers, don't use "the following line" etc.

Restrictions on a function (parameter x must be +ve etc) are best done using one of the documentation libraries, so the restrictions are in comments near the function code.

Restrictions that run beyond a single function (don't use this code for more than 1 million elements at a time!, NaNs must be removed from data before input) should be mentioned both on the functions where they arise and in your examples and any user docs. User docs are vital if you're sharing you code with other people.

Check yourself before you Wreck yourself

Sometimes, you have known preconditions (things which must be true when a function is called), invariants (things which should stay the same throughout), and postconditions (things which are true after a function ends). You should consider adding code to check (some of) these things. Sometimes a compiler can do some of this for you, such as Fortran's array bounds checking, or C++'s ".at()" to get a vector element, rather than using "[]". Since these can be costly in run time, they may be better as a debugging option rather than a normal part of the code.

Test your Limits

The last essential element is testing. This is a very broad topic, and really isn't easy. At its core though, it means working out what parts of the code should do and checking that they do it. Smaller parts are easier to handle, but (remember your time is precious) can mean writing lots of trivial tests where you could easily absorb them all into one test. If it is at all possible, have at least one test case of the entire code - something you can run and know the answer to, that isn't too trivial. We talk a little about testing in our workshopsand there are millions of books, blog posts, papers etc on how to do it. Just make sure you do something. The rest of this post should convince you why.

The Ideal World

In an ideal world, any code you write has a well thought out, complete specification. Its purpose is fixed and perfectly known; there's a wealth of known results to compare to; the computations are all nice and self-contained; nobody else is really working on it, so there's no rush; and yet you can still do something novel and interesting with it. In this case, avoiding errors is as easy as it ever gets. You split the code into self-contained modules, each doing one thing and doing it well. You write careful error checking into every function, making sure they are never called with invalid parameters. You test each function of every module carefully in isolation and as you put the modules together, you test at various stages. Finally, you run some problems with known results and verify that you get the correct answer. As you write, you document all the functions and the assumptions they make. Finally, you write some user documentation describing what the code is and does, what its limitations are.

That's all great in theory, but for it to be enough it really is essential that everythingof importance be in the specs. I am not sure that has ever happened, anywhere, ever. It gets close in things like Aerospace and Medical engineering, but even there mistakes are made.

Code or Computation

The second best situation, from a coding point of view, is the one where the Code is trivial, but the Computations hard. For instance, a simple program to count which words occur near to some target word is fairly easy to write, although far form trivial (capitalisation, punctuation, hyphenations etc), but text concordance analysis is widely used and is able to find new, novel results if correctly analysed. Or imagine coding up a very sophisticated algorithm, with minimal support code. From a code perspective this is also quite easy - you just make sure the code contains the right equations (and hope that you wrote them down correctly).

Of course, in these case you still do all the things from the ideal-world - you modularise the code, you test it carefully, you document the assumptions. I'm not going to discuss all the details of this here: we cover a lot of it in our training, especially here. You do all this, and you'll catch some errors, but it probably wont be all of them.

Sadly though, with many pieces of code you may write, you don't even have this level of "simplicity". The code is hard - it uses difficult techniques, or has many interacting pieces, or there aren't any good test cases you can use. Testing, verification and documentation as important now, if not more so. But you wont find all the potential errors.

What's the Point?

But at this point, it seems a bit pointless. You make all this effort, spend all this time on testing and verification, and its still probably got bugs in. Your code still probably has errors somewhere no matter what you do.

Well firstly, and most practically, the more you check and test, the smaller your errors are hopefully getting. You catch any really big glaring errors, you fix lots of things, and you absolutely have a better piece of code. The time you're spending is never wasted, although its important to try and direct it to the places you get the most return. Keep the 80-20 rule in mind. You get 80% of the gains from the first 20% of the effort in many endeavours, but getting the final 20% takes the remaining 80% of the time. This doesn't mean you should fix 20% of the bugs though! It means you finish all the easy 80s before moving on to the more finicky checks.

Secondly, more philosophically, you will always know that you did it. You can point to the time you spent and say "I tried. I did the things I was supposed to. I wont make this mistake again." That is a very nice thing to be able to say when you find you've made a mistake that's going to cost you time and/or effort.

Sh*t Happens

So, if you've published a paper, you've probably published a mistake. Hopefully, this isn't a important msitakes, just some minor grammar, a misprinted formula, or a miswritten number that doesn't affect results. It's pretty important to keep perspective. Everybody makes these mistakes. The important thing is not to make them because you didn't do things how you know you should.

I like to read Retraction Watchon occasion. Much of what they publish is serious misconduct from serial bad actors, but sometimes there's an interesting post about honest mistakes, and interesting commentary on how it's meant to work. For instance, when should a paper be retracted rather than amended? How do you distinguish stock phrases from plagiarism? What's so bad about salami slicing?

I particularly like things like this retraction, where an error was found that needs more work to address, so a correction wasn't suitable; or this one where the error meant the result wasn't interesting. The errors probably happen a lot more often than the retractions or corrections. That's not really in scope of a blog post, but it's good to see the process working well.

$600,000 Mistakes

To summarise briefly, everybody makes mistakes. If you're lucky yours will be minor, caught before they go "into the wild", and make funny anecdotes rather than tragic tales. The worst thing you can do, as a researcher, is try to hide your errors. You should fix them. Whether this can be done in a follow-on paper, or needs a correction or retraction, it should be done!

Errors in code, rather than papers, are a tiny bit different - they're probably more likely to be harmless. But don't be fooled - if the error invokes "undefined behaviour" (anything outside of, or not conforming to, the standard) you will never know whether the result was affected unless you redo everything. It's very tempting to run a few test cases, find they're OK and assume all the others are fine too. The definition of undefined behaviour means you can't really know - its perfectly valid to work 99 times and fail the 100th. If you find any, you have to fix it and redo all results. Hopefully they're unchanged. That's great. You fix the code, you upload the fixed version - admitting the error and the fix, and you add the error to your list of "things to be careful of". You don't need to do anything about your results - they're still solid.

If results are affected, talk to your supervisor, or PI, or trusted colleagues and work out what's an appropriate solution. Make sure you've learned from this and you're not going to make the same mistake again. The former CEO of IBM, Thomas John Watson Snr said it best:

Recently, I was asked if I was going to fire an employee who made a mistake that cost the company $600,000. “No”, I replied. “I just spent $600,000 training him – why would I want somebody to hire his experience?”

Don't be scared of bugs. They happen and they always will. But the more you find, the better you develop your "sixth sense" and can almost smell where they might be, and the better your code and the research you do with it, will become.


February 2023

Mo Tu We Th Fr Sa Su
Jan |  Today  |
      1 2 3 4 5
6 7 8 9 10 11 12
13 14 15 16 17 18 19
20 21 22 23 24 25 26
27 28               

Search this blog

Tags

Galleries

Blog archive

Loading…
RSS2.0 Atom
Not signed in
Sign in

Powered by BlogBuilder
© MMXXIII