A critique of "How to C in 2016"
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


A critique of "How to C in 2016" by Matt

Keith Thompson, Sat 2016-01-09, updated Fri 2016-01-15, Sun 2018-06-03

Matt (whose web site does not mention his last name as far as I can tell) has written an article "How to C in 2016". It's been linked to from Reddit and from Hacker News; the latter is where I saw it.

Update: Matt has been kind enough to add a link to this critique to his article.

Update: A couple of people have found Matt's last name from other sites, but since he didn't choose to include it in his article I won't mention it here.

Just to avoid any possible confusion, I am not Ken Thompson, nor am I related to him.

This article has been linked from https://reddit.com/r/programming/, and the link is currently at the top of the page. Lots of comments there, some of them even constructive.

As is inevitable for anything expressing opinions about the C language, there are some things I disagree with. This critique is intended to be constructive criticism. It's entirely possible that in some cases Matt is right and I'm wrong.

I haven't quoted Matt's entire article. In particular, I've omitted some material that I agree with. I suggest reading this critique in parallel with Matt's article.

The first rule of C is don't write C if you can avoid it.

I don't agree, but it's too broad a topic to discuss meaningfully.

C99 is the default C implementation for clang, no extra options needed.

That depends on the version of clang. clang 3.5 defaults to C99. clang 3.6 defaults to C11. I'm not sure how strict it is by default.

If you want to use a particular standard for either gcc or clang, be explicit: use -std=cNN -pedantic.

gcc-5 defaults to -std=gnu11, but you should still specify a non-GNU c99 or c11 for practical usage.

Unless you want to use gcc-specific extensions, which is a perfectly legitimate thing to do.

If you find yourself typing char or int or short or long or unsigned into new code, you're doing it wrong.

Sorry, this is nonsense. int in particular is going to be the most "natural" integer type for the current platform. If you want signed integers that are reasonably fast and are at least 16 bits, there's nothing wrong with using int. (Or you can use int_least16_t, which may well be the same type, but IMHO that's more verbose than it needs to be.)

For modern programs, you should #include <stdint.h> then use standard types.

The fact that int doesn't have "std" in its name doesn't make it non-standard. Types such as int, long, et al are built into the language. The typedefs defined in <stdint.h> are later add-ons. That doesn't make them less "standard" than the predefined types, but they're certainly no more standard.

float — standard 32-bit floating point

double - standard 64-bit floating point

float and double are very commonly IEEE 32-bit and 64-bit floating-point types, particularly on modern systems, but there's no guarantee of that in the language. I've worked on systems where float is 64 bits.

Notice we don't have char anymore. char is actually misnamed and misused in C.

C's conflation of characters and bytes is unfortunate, but we're stuck with it. The type char is guaranteed to be exactly one byte, where a "byte" is at least 8 bits.

Developers routinely abuse char to mean "byte" even when they are doing unsigned byte manipulations. It's much cleaner to use uint8_t to mean single a unsigned-byte/octet-value and uint8_t * to mean sequence-of-unsigned-byte/octet-values.

If you want bytes, use unsigned char. If you want octets, use uint8_t. If CHAR_BIT > 8 then uint8_t won't exist, and your code won't compile (which is probably what you want). If you want something that's at least 8 bits, use uint_least8_t. If you want to assume that bytes are octets, add something like this to your code:

#include <limits.h>
#if CHAR_BIT != 8
    #error "This program assumes 8-bit bytes"

Note that POSIX requires CHAR_BIT == 8.

the C type of string literals ("hello") is char *.

No, the type of string literals is char[]. In particular, the type of "hello" is char[6]. Arrays are not pointers. See section 6 of the comp.lang.c FAQ for more on this topic.

At no point should you be typing the word unsigned into your code. We can now write code without the ugly C convention of multi-word types that impair readability as well as usage.

Many C types have multi-word names. There's nothing wrong with that. Saving keystrokes is not a good reason to use terse abbreviations.

Who wants to type unsigned long long int when you can type uint64_t?

For one thing, you can use unsigned long long; the int is implied. For another, they mean different things. unsigned long long is at least 64 bits, and may or may not have padding bits. uint64_t is exactly 64 bits, has no padding bits, and is not guaranteed to exist.

And unsigned long long is a predefined C type. Any C programmer will recognize it.

You can also use uint_least64_t -- which may or may not be the same type as unsigned long long.

The <stdint.h> types are more explicit, more exact in meaning, convey intentions better, and are more compact for typographic usage and readability.

Yes, the intN_t and uintN_t types are more explicit. That's not necessarily a good thing. Don't specify things that you don't care about. Use uint64_t only if you really need exactly 64 bits, no more, no less.

Sometimes you do need a type with a particular exact width, for example when you need to conform to some externally imposed format. (Sometimes you also need a particular endianness, alignment, and so forth; C's <stdint.h> doesn't let you specify those.) But more often all you need is a particular range of values. For that, you can use either the [u]int_leastN_t or [u]int_leastN_t types, or one of the predefined types.

The correct type for pointer math is uintptr_t defined in <stddef.h>.

This is dangerously wrong.

First off a minor point: uintptr_t is defined in <stdint.h>, not <stddef.h>.

That's assuming it's defined at all. An implementation where void* can't be converted to any integer type without loss of information won't define uintptr_t. (Such implementations are admittedly rare, perhaps nonexistent.)

Instead of:

        long diff = (long)ptrOld - (long)ptrNew;

Yes, that's bad.


        ptrdiff_t diff = (uintptr_t)ptrOld - (uintptr_t)ptrNew;

That's no better.

If you want the difference in terms of the type they point to, use:

ptrdiff_t diff = ptrOld - ptrNew;

If you want the difference in bytes:

ptrdiff_t diff = (char*)ptrOld - (char*)ptrNew;

If ptrOld and ptrNew don't point into, or just past the end of, the same object, the behavior of the pointer subtraction is undefined. Converting to uintptr_t will give you some result -- but that result won't be useful. Unless you're writing low-level system code, don't do any pointer comparison or arithmetic unless both pointers point into or just past the end of the same object. (Exception: pointer == and != work correctly for pointers to distinct objects.)

In these situations, you should use intptr_t — the integer type defined to be the word size of your current platform.

No, it isn't. The concept of "word size" is not well defined. intptr_t is a signed integer type big enough that you can convert a void* to intptr_t and back without loss of information. It might be bigger than void*.

On 32-bit platforms, intptr_t is int32_t.

Probably, but it's not guaranteed.

On 64-bit platforms, intptr_t is int64_t.

Again, probably, but it's not guaranteed.

size_t is defined as "an integer capable of holding the largest array index"

No, it isn't.

which also means it's capable of holding the largest memory index" which also means it's capable of holding the largest memory offset in your program.

It's capable of holding the size of the largest object your implementation supports. (There's an argument that that's not necessarily guaranteed, but for practical purposes you can rely on it.) It can hold the largest memory offset if all offsets are within a single object.

In either case: size_t is practically defined to be the same as uintptr_t on all modern platforms, so on a 32-bit platform size_t is uint32_t and on a 64-bit platform size_t is uint64_t.

Likely, but not guaranteed.

More to the point, size_t can represent the size of any single object, while uintptr_t can represent any pointer value, which means it can distinguish between the addresses of any byte of any object. Most modern systems have monolithic address spaces, so the theoretical maximum size of an object is the same as the size of the entire memory space -- but the C standard very carefully does not require this. You could have a 64-bit system where no object can be bigger than 232 bytes, for example.

By emphasizing "modern" systems, you ignore both old systems (like the old x86 systems that exposed segmented addressing with "near" and "far" pointers) and possible future systems that might be perfectly compatible with the C standard while violating "modern" assumptions.

You should never cast types during printing. You should use proper type specifiers.

That's one way to do it, but it's not the only good approach. (And even you agree that you need to cast to void* for "%p".)

raw pointer value - %p (prints hex value; cast your pointer to (void *) first)

Good advice -- but the output format is implementation-defined. It's usually hex, but don't assume that.

        printf("Local number: %" PRIdPTR "\n\n", someIntPtr);

The name someIntPtr implies a type of int*, but in fact it's of type intptr_t.

There's an alternative which means you don't have to remember the alphabet soup of macro names:

some_signed_type n;
some_unsigned_type u;
printf("n = %jd, u = %ju\n", (intmax_t)n, (uintmax_t)u);

intmax_t and uintmax_t are typically 64 bits. The conversions are going to be far cheaper than the physical I/O.

Notice you put the '%' inside your format string, but the type specifier is outside your format string.

It's all part of the format string. The macros are defined as string literals, which are concatenated with adjacent string literals.

Modern compilers support #pragma once

That doesn't mean you should use it. Even the GNU cpp manual doesn't recommend it. The section on "Once-Only Headers" doesn't even mention #pragma once; it discusses the #ifndef idiom. The following section, "Alternatives to Wrapper #ifndef", briefly mentions #pragma once but points out that it's not portable.

This pragma is widely supported across all compilers across all platforms and is recommended over manually naming header guards.

Recommended by whom? The #ifndef trick isn't pretty, but it's reliable and portable.

IMPORTANT NOTE: If your struct has padding, the {0} method does not zero out extra padding bytes. For example, struct thing has 4 bytes of padding after counter (on a 64-bit platform) because structs are padded to word-sized increments. If you need to zero out an entire struct including unused padding, use memset(&localThing, 0, sizeof(localThing)) because sizeof(localThing) == 16 bytes even though the addressable contents is only 8 + 4 = 12 bytes.

This gets a bit tricky. Usually there's no reason to care about padding bytes. If you do, then yes, memset is the way to zero them. But zeroing a structure with memset, though it will set any integer members to zero, is not guaranteed to set floating-point members to 0.0 or pointers to NULL (though it will on most systems).

C99 allows variable length array initializers

No, C99 doesn't allow initializers for VLAs (variable length arrays). But Matt isn't actually talking about VLA initializers, just about VLAs.

VLAs are controversial. Unlike malloc, VLAs provide no mechanism for detecting allocation failures. So if you need to allocate N bytes of data, then this:

    unsigned char *buf = malloc(N);
    if (buf == NULL) { /* allocation failed */ }
    /* ... */

is at least in principle safer than this:

    unsigned char buf[N];
    /* ... */

Certainly VLAs are dangerous when used incorrectly. The same is true of just about every feature in every language.

But old-fashioned fixed-length arrays have exactly the same problem. As long as you check the size before creating the array, a VLA of some variable size N is no more dangerous than a fixed-length array of the same size. And it's common to define a fixed-length array that's bigger than it needs to be to ensure that you use part of it to store your actual data. With a VLA, you can allocate just what you need. I agree with Matt's advice here.

Except for one thing: C11 made VLAs optional. I doubt that many C11 compilers will actually decide not to implement them, except perhaps for small embedded systems, but it's something to keep in mind if you want your code to be as portable as possible.

If a function accepts arbitrary input data and a length to process, don't restrict the type of the parameter.

So, do NOT do this:

        void processAddBytesOverflow(uint8_t *bytes, uint32_t len)

Do THIS instead:

        void processAddBytesOverflow(void *input, uint32_t len) 

(I've omitted the bodies of the functions.)

I agree, void* is the right type to use for a parameter pointing to an arbitrary chunk of memory. See the mem* functions in the standard library. (But len should be size_t, not uint32_t.)

By declaring your input type as void * and re-casting inside your function, you save the users of your function from having to think about abstractions inside your own library.

A small quibble: There's no cast in Matt's function. There's an implicit conversion from void* to uint8_t*.

Some readers have pointed out alignment problems with this example.

Some readers are mistaken. Accessing a chunk of memory as a sequence of bytes is always safe.

C99 gives us the power of <stdbool.h> which defines true to 1 and false to 0.

Yes, and it also defines bool as an alias for the predefined Boolean type _Bool.

For success/failure return values, functions should return true or false, not an int32_t return type with manually specifying 1 and 0 (or worse, 1 and -1 (or is it 0 success and 1 failure? or is it 0 success and -1 failure?)).

There is a widespread convention, particularly in Unix-like systems, for functions to return 0 for success and some non-zero value (often -1) for failure. In many cases different non-zero results denote different kinds of failure. It's important to follow this convention when adding new functions to such an interface. (0 is used for success because typically there's only one way for a function to succeed, and multiple ways for it to fail.)

A function that tests whether some condition is true or false should return true or false. But success vs. failure is often a different thing.

A bool function should have a name that's a predicate. In English, it would be something that answers a yes/no question. Examples are is_foo() and has_widget(). A function that tries to do something, and that needs to let you know whether it succeeded or not, should probably use a different convention. In some languages raising or throwing an exception is the right approach. In C, you should follow some existing convention -- and zero for success is the most common one.

The only usable C formatter as of 2016 is clang-format. clang-format has the best defaults of any automatic C formatter and is still actively developed.

I haven't used clang-format myself. I'll have to look into it.

I have my own fairly strong opinions about C code layout:

  • Opening brace goes at the end of the line;
  • Spaces, not tabs;
  • 4-columns per level;
  • Always use curly braces (except in rare cases where putting a statement on one line improves readability).

These are just my own personal preferences, which can be overridden by one important rule:

  • Follow the conventions of the project you're working on.

I don't often use automatic formatting tools myself. Perhaps I should.

Never use malloc

You should always use calloc.

I disagree. Initializing allocated memory to all-bits-zero is somewhat arbitrary, and it's typically not a meaningful value. If you write your code correctly, you won't access any object unless you have first assigned a meaningful value to it. Using calloc means that a bug in your code will give you a value of zero, rather than some arbitrary garbage. This isn't necessarily an improvement.

Zeroing memory often means that buggy code will have consistent behavior; by definition it will not have correct behavior. And consistently incorrect behavior can be more difficult to track down.

Of course there's no such thing as bug-free code. But if you're trying to program defensively, you might consider initializing allocated memory to some value that's known to be invalid rather than one that might be valid.

On the other hand, if all-bits-zero happens to be a reasonable initial value, calloc might be a good approach.