How to make fewer errors at the stage of code writing | Hacker News Reader

How to make fewer errors at the stage of code writing

viva64.com

44 pointsGarbage23 comments

Tangentially related (section 3 talks about this): why would you ever disable code using the && false idiom?

You could just remove it (and rely on you source management system to get it back) or you comment it out (where you will at least get syntax highlighting to help you spot the dead code).

It happened multiple times to me that I missed a tucked in &&false while I was porting older code to a new API. As the code was disabled anyways, I could have spared myself the time of patching it up.

Now sure, if you are just testing something during development, this might be a viable way to disable code, but something like this should not be committed. Please use proper comments or better, just remove the code.

In this light I don't understand why the article is even trying to suggest syntactical "improvements" (adding parentheses doesn't make it more readable IMHO), when the complicated expression should have just been rewritten to

    UINT uFlags;
    uFlags = DST_ICON | DSS_NORMAL;

No condition, more readable and less runtime overhead. And whoever committed this got the pleasure of committing "red lines" (git shows removed lines as red and being able to remove stuff always feels good)

I've definitely spent a good while in the past trying to decipher the intent of someone who used &&alwaysFalseExpression - Is it a bug? Am I overlooking some subtle side-effect somewhere? If they just wanted to turn the code off, they would have just commented it out...right?

you have a declaration without initialization there. Variables should always be initialized - this removes another large potential source of bugs.

UINT uFlags = DST_ICON | DSS_NORMAL;

or in C++

UINT uFlags( DST_ICON | DSS_NORMAL );

Also, declare variables const whenever possible.

Understand the problem you are trying to solve. Accept no substitutes.

Get good with your tools (whether that be IDE, knowledge of the language, APIs etc)

Test your assumptions mercilessly

If you write code as though nobody will ever see it and nobody cares, the code will suck. If you write code as though for the best, most respected and most intelligent hacker(s) you know, the then code will be better. Write code as though you have an important audience that cares deeply about the result.

Yes, you can sprinkle the code with comments explaining to your important audience how awesome you are and what a clever piece of coding something is. And this is actually a good thing, since these become focal points for mercilessly testing your assumptions :D

It seems to me that the most common errors in C/C++ are memory leak and/or corruption issues. Coming from someone who's background is in managed/garbage collected languages, doesn't using a managed language help to avoid these common mistakes?

I know that it's still possible to leak memory in C#, like not releasing event handlers, but it's very difficult, if not impossible, to corrupt memory without using unsafe code. In those cases you're probably more likely to be careful with how memory is used or manipulated anyway to avoid causing corruption with the managed code.

Maybe true, but still not much consolation for the embedded systems programmers or maintainers of legacy code.

I would say that the most common errors are because there is no built-in string type (at least in C) and lot of library function expects null terminated strings.

And these are hard to get right, because they are very prone to off-by-one errors, and novice programmers tend to write string manipulation routines from scratch, which leads to corruption.

Use libraries (e.g. bstring or std::string in C++) that do that for you, and you won't have problems.

Regarding memleaks: you can leak database handles, file handles and other resources in managed languages. In C/C++ memory is one of the resources you also take care of. In C++ there are also destructors and you can use RAII to handle most of the cases without thinking.

I wouldn't bundle C and C++ in the same group in memory handling aspects. C++'s destructors and RAII make memory management rather easy (I actually think that it's easier than garbage collection, as I know with some certainty when some memory is allocated and freed. No need to trust some magical black box that GC is.

> Align everything you can in code

Yes. Makes code easier to read and makes it easier to refactor if you use an editor with column select/multi-line caret.

I personally detest people who do this, outside of things like tables (i.e. structured constant initializers) and comments on type definitions. If there's a high degree of regularity in the source that you're trying to expose, there's probably a better way of doing it - extracting out common variables, using a table lookup, adopting acronym conventions for regular parts along any given axis which have the same length, so that the code naturally lines up without ugly random rivers of space down the middle.

Unsigned types, on the other hand, are just evil. Way too many people use an unsigned type because they think it represents an invariant that this value will never be negative, that it has some kind of documentation or safety aspect, but that's just wrong. Unsigned types don't act according to the intuitions of most average programmers, and things as simple as for-loops are incredibly easy to get wrong with them (e.g. think about the most common idiom for iterating backwards in C derivatives). If I had my druthers, mixing signed and unsigned ought to expand the expression to the next larger signed type that can represent the range of both (and error out if there is none), and there ought to be separate subtraction / decrement operators specifically for unsigned types. 'u += -2' where u is unsigned would be a type error - assigning a signed value to an unsigned location.

Sometimes I arrange my variable declarations so that they make a picture of something in profile...

... don't hate me because I'm beautiful ... :D

----

On a more serious note, something I have noticed working on large projects with other programmers, is that if each programmer has their own distinctive style and conventions, then you can tell straight away by looking at it who wrote it, and more importantly who modified it.

Thus if Billy (who is an idiot) modified something written by Sheila (who is like unto the goddess of inspired poetry in wisdom and understanding), and the whole is broken, you can pretty much bet that it is Billy's modification that has broken it, and you can spot his bits straight away, this makes debugging a lot easier.

I think this is one of the more unfortunate side-effects of things like checkstyle and other forms of overly restrictive style enforcement. The more we sand-paper everybody back to the same (low) level of code appearance, the more information is lost for what I like to call "forensic debugging".

I strongly agree with this as well - however, when I need to overhaul a huge portion of the codebase, I may end up normalizing tabs and spacing somewhat; e.g. tabs between 'if', 'while', 'for' and '(', spaces between function / array name and '(' or '[', etc.

That is to say, I don't mind everyone being free to format code how they see fit, just don't be surprised if in rewriting or modifying it, it gets "fixed" differently.

I'm currently doing this right now at work, hence my vehemence. There's even things like this in class definitions:

    int const * const	FooBar // Yes, that's a tab
        (Foo foo, Bar bar)

which at first glance can end up looking like instance variable declarations instead of methods.

I probably shouldn't have left it at that - if there's much regularity outside of tables, then you're definitely doing something wrong. Within code I see the alignment as a way of highlighting regular areas as potential spots for refactoring. Sometimes it doesn't make sense to turn something into another layer of abstraction yet, but it may later. I can certainly see how the extra spacing might grate on some people's nerve's, though.

Edit: Something I hadn't thought of - you mention "rivers of space" - aligning code might make it harder to read for coders with dyslexia: http://uxmovement.com/content/6-surprising-bad-practices-tha...

Regarding your point about unsigned types - If I'm remembering correctly, "Writing Solid Code" had a section about exactly that - nasty bugs caused by things like looping over unsigned integers in reverse.

I'm one of those few programmers who like to code with a proportional font. I have been declared mad as such, with the reasoning "you can't line up things with a proportional font". Well, guess what, I don't need to. I pay attention to correctly indenting my code, and that works even with proportional fonts. I find monospaced fonts hard and unnatural to read. Almost all books, magazines and newspapers use proportional fonts!

If your code needs lining up to be legible, you're doing something wrong.

A better approach to #5 (Do not copy a line more than once) is to make a "template" copy of the line, which will fail to compile, and then make a bunch of copies of that.

Example:

Template line:

    EnableWindow(GetDlgItem(hwndDlg,),);

Copy/paste:

    EnableWindow(GetDlgItem(hwndDlg,),);
    EnableWindow(GetDlgItem(hwndDlg,),);
    EnableWindow(GetDlgItem(hwndDlg,),);

Final edit:

    EnableWindow(GetDlgItem(hwndDlg, IDC_PRIMARYSTATUS), TRUE);
    EnableWindow(GetDlgItem(hwndDlg, IDC_CYCLETIMESPIN), FALSE);
    EnableWindow(GetDlgItem(hwndDlg, IDC_CYCLETIME),);

Now when you try to compile, but forgot to update something (such as the missing boolean in the last line), the compiler will complain.

For languages that support it, use lambda functions judiciously - too much of them make code really error-prone and hard to debug. (Not the most wide-spread problem, but came across this today)

Also check for null!!! ;)

Your experience is the exact opposite of what I was promised by the functional programming community. Can you please elaborate on how lambdas are giving you trouble?

Don't get me wrong, lambdas are really useful and elegant, I just think they should be used when there is a purpose other than simply making the code succinct - such as limiting scope. I've seen people use them instead purely to reduce the number of lines of code, which I think in the end makes it less readable and harder to debug.

Well, this one is not impressive, but I find it useful for C-like languages, Java included:

    if (5 == somevar) { ....

If you accidentally write "=", you'll notice.

Many compilers (gcc and mxmlc come to mind) simply warn you if you try to do "if (somevar = 5)" and suggest an extra set of parenthesis to shut up the warning, if that's really what you wanted. Python throws a syntax error.

7. Don't write code in C

I know it sound unhelpful, but it isn't. If you can avoid low-level, unsafe languages like C and C++, by all means do so.

Chose a high-level language like Python, Haskell, Lisp… Anything with closures and automatic memory management, provided it let you fall back to C when you really need it.

Outside maintenance, few project need to use C or C++ all the time.

In C++ you should be using the typesafe functions std::fill and std::copy instead of memset and memcpy.

Reply on news.ycombinator.com