Why C Strings Suck And How I Fixed Them

I have always been frustrated about C not having a proper String type that most modern languages have nowadays. I love the simplicity of C, but there are some features that I would really like to have in it, and this is one of them.

Since I made a small library (just one .c file) that you can use to make the experience of working with strings in C just a little better, I would like to show it to you here.

For the impatient, this is the general idea:

struct str_view {
    usize           len;
    const u8       *data;
};

struct str_owned {
    usize           len;
    u8             *data;
};

But before getting into the library itself, I will go over how ā€œstringsā€ work in C and why I find them annoying, what the issues are with this model, and then it will be easier to explain why I decided to take the route that I took.

If you only care about the actual implementation details, you can click here.

How Strings Work in C

As many might already know, strings in C are NUL-terminated. NUL is an ASCII character which, in the context of ASCII encoding, literally corresponds to the value 0. In C, this character is represented by '\0'. Inside a text, it generally does not carry any semantic meaning, so it will never be used inside the text for the sake of carrying information (as opposed to letters like 'a' or 'y').

C does not really have ā€œstringsā€ per se. It only has arrays. These arrays are simply chunks of data where you can put things. To understand where the array ends, one has two options:

  • Store the length of the data somewhere
  • Put something at the end of the array to signify its end

C went with the second option.

Since ā€œstringsā€ (which are not really strings) in C are simply arrays of characters, each string has a NUL-terminator at the end of it.

C is a very old language and back then - when it was first created - this seemed like a good idea.

Storing the length of the data separately required more memory than a NUL-character at the end. A NUL-character in ASCII encoding is 8 bits (or 1 byte). If you tried to store the length of the data in only 8 bits, you could only have strings that are up to 255 characters in length, which is obviously too short. Bigger numbers require more data to store, so weā€™re talking about around 4 to 8 bytes instead of 1.

But this is not even the main reason - the main reason is because of simplicity. Back when C was created, there existed some languages that already had strings with a separately stored length. C just wanted to be as simple as possible, and having NUL-terminated character arrays seemed like a sane decision.

The Problems With This Approach

There are multiple issues with this.

First Issue

First of all, if you want to find out the length of a string, with a NUL-terminated string you have to start at index 0 and start counting until you reach the end of the data. This process has a time complexity of O(n)O(n), as opposed to O(1)O(1) for just reading the separately stored length.

This is fine for short character arrays, but imagine you had to figure out the length of a 1,000,0001,000,000 character text. That would take a considerable amount of CPU cycles that could have been used for something else.

This bothers me especially in a language like C, since it is supposed to give you some of the best performance that programming languages have to offer (without counting assembly and certain exceptions in some specific scenarios). C is often the baseline when comparing languages to gauge their relative performance. Why, if C focuses so much on performance, would you settle for this?

Second Issue

Let us assume that we have some C-string a:

char *some_str = "Hello World";

This is great, but now I would like to work with the second word specifically (World). I do not care about changing the data, just about reading. Maybe I want to pass it to a function or something. Well, I could do something like this:

char *only_world = some_str + 6;

By doing this, we are reusing the NUL-terminator that some_str was using to terminate itself.

Fine, what if I want to access the first word only, though? Now we have a problem. We somehow have to end the string at the fifth index. To do that, we would have to insert a NUL-character at index 6. By doing so, we are changing the data that some_str is pointing to, and we can therefore not use that variable anymore.

So, what do we do? We have to copy the string. Thatā€™s slow, since it involves memory lookups. We have to allocate to memory, and then copy it. We can also use the stack if we know the size of the data at compile time, but that is often not the case. That would look like this:

char *only_hello = malloc(6);
memcpy(only_hello, some_str, 5);
only_hello[5] = 0;

In this simple example, the stack would have been fine, but in most realistic scenarios, we have to use heap allocations.

This problem is solved by so-called ā€œstring viewsā€, which are basically immutable data and a length. You can just give the string view the same pointer as some_str and decrease the length to whatever you need, 5 in this case.

Third Issue

God forbid you forget to NUL-terminate a string. If you pass a pointer to a non-NUL-terminated string to a function in string.h, like strlen, the computer will just keep counting characters until it either randomly finds a NUL-character or - alternatively - until it tries to access a region of memory that it should not touch and the operating system kills the process.

This generally does not happen with actual strings (or string views), since you will use functions to interact with them, and they always keep the length of the string in sync with the data. In that case you always know when the string stops, you do not have to rely on remembering to NUL-terminate all your strings.

Other Issues

There are probably more issues, but these are hopefully enough to convince you. In general, I find strings with a separately stored length more intuitive and easier to work with.

The Solution

Luckily, strings are not particularly hard to implement in C. Here is how I did it.

Types

Just so you are aware when reading the code, I typedef some basic types in C before I start coding. In this case we will only need a few of them, but this is what I paste at the start of every project of mine.

#include <stdint.h>
#include <stdbool.h>
#include <stddef.h>

typedef uint8_t     u8;
typedef uint16_t    u16;
typedef uint32_t    u32;
typedef uint64_t    u64;

typedef int8_t      i8;
typedef int16_t     i16;
typedef int32_t     i32;
typedef int64_t     i64;

typedef float       f32;
typedef double      f64;

typedef size_t      usize;

This makes it more convenient to use the integer types with a fixed size instead of the older int, long, long long, etc.

I generally prefer this since int is not the same size on all platforms. int32_t on the other hand is guaranteed to be exactly 32 bits.

Structs

First, letā€™s define our string types. I donā€™t prefer not to typedef structs, but if you want to you can do so. If you do decide to typedef these structs, I would advise you to use PascalCase as a naming convention (to better differentiate them from built-in types).

struct str_view {
    usize           len;
    const u8       *data;
};

struct str_owned {
    usize           len;
    u8             *data;
};

As you can see, we have two string types: struct str_view and struct str_owned. The difference is that struct str_owned owns the data, and the data can also be changed. Ideally, you should try to only have one str_owned pointing at any given region in memory.

On the other hand, struct str_view should be used as read-only views of a given region of data. It should not be used to modify the data.

Separating these two concerns is nice for multiple reasons. Who owns the data becomes clear immediately. This is especially useful when returning string structs from functions. By using this model, it is easier to understand whether we need to concern ourselves with freeing data or something similar.

As a general rule of thumb, struct str_owned are heap-allocated. If you have a struct str_view viewing some region in memory owned by a struct str_owned, that string view is only valid as long as the underlying owned string is still valid. That means that you have to be careful about not freeing the struct str_owned and after that still using string views that view that data.

Also, you might have noticed that I am using u8s (a.k.a. uint8_ts) instead of chars. That is because, technically - even though in almost all cases it does not matter - chars are not guaranteed to have a size of exactly 8 bits. On certain embedded systems, a char is sometimes 16 or 32 bits large. On very old systems, they used to be 9 bits. This has historical reasons. A char is simply guaranteed to have a size of one byte, but the exact size of a byte is not clearly defined as a definitive size in bits. Knowing all that, using u8 is more consistent across platforms.

As a side note, this is also the reason why the CHAR_BIT macro exists. It tells you how many bits are in a char on the system that the code is run. That is because the size of a char depends on the system that you are currently on.

Also, I do not really like using a redundant type. A u8 holds all the semantic information I need, so u8 it is.

Macros

Next, we create some macros. I try to minimize the use of macros, but in certain cases they are just perfect for the job. I believe that this is one of them.

Mainly I care about the STR macro, that I use to easily create a struct str_view out of a char * literal.

#define LENGTHOF(s) (sizeof(s) / sizeof(*s))
#define STR(s) (struct str_view){ (const u8 *)s, LENGTHOF(s) - 1 }
#define STR_PRINT(s) (i32)s.len, s.data

STR_PRINT is used to print a struct str_view. You can use the %.*s format specifier to manually give it the length of the string to be printed (since string views are not guaranteed to be NUL-terminated, since they can view a region of memory in the middle of a struct str_owned). This can be done as follows:

printf("Here is my string: %.*s\n", STR_PRINT(STR("some_string")));

Here you can also see the STR macro in action, which creates a struct str_view on the fly from a char * literal.

Freeing Memory

It is trivial to free a struct str_owned, since you only have to call free(s.data). In any case, I still wanted to create a utility function for it.

void str_owned_free(struct str_owned s) {
    free(s.data);
}

Conversion Functions

We need some way to convert between struct str_view and struct str_owned and back. Converting an owned string to a string view is trivial; you simply use the same pointer and length of the owned string in a struct str_view:

[[nodiscard]]
struct str_view str_owned_to_str_view(struct str_owned s) {
    struct str_view str_out = {0};
    str_out.data = s.data;
    str_out.len = s.len;
    return str_out;
}

Also, as you might have noticed, I like using [[nodiscard]]. This is a compiler directive which makes the compiler issue a warning if the return value of the function flagged with this directive is disregarded. In scenarios like this one I find it to be very useful, since it is completely useless to call this function if you are not going to do anything with the result.

We also need to convert string views to owned strings. This is a bit more tricky and should not be done unless strictly necessary, since it requires memory allocation. An owned string can be mutated, which means that we have to give it its own region in memory to avoid damaging other owned strings that might be using the same memory (recall when I said that we should try to avoid having multiple owned strings on the same chunk of memory).

[[nodiscard]]
struct str_owned str_view_to_str_owned(struct str_view s) {
    struct str_owned str_out = {0};
    str_out.data = malloc(s.len + 1);
    memcpy(str_out.data, s.data, s.len);
    str_out.len = s.len;
    str_out.data[str_out.len] = 0;
    return str_out;
}

Utility Functions

I also wrote a bunch of functions that implement common string operations and hopefully make the experience of working with strings much more ergonomic than the standard library.

Finding a Substring in a String

When I have to return multiple values, I like to return structs as opposed to using output variables. In fact, I have come to really dislike output variables. So here is my solution:

struct str_find_out {
    usize index;
    bool found;
};
[[nodiscard]]
struct str_find_out str_find_substr(struct str_view s, struct str_view sub_s) {
    struct str_find_out out = {0};
    usize match_index = 0;
    for (usize i = 0; i < s.len; i++) {
        if (s.data[i] == sub_s.data[match_index]) {
            match_index++;
        } else {
            match_index = 0;
        }
        if (match_index == sub_s.len) {
            out.found = true;
            out.index = i - match_index + 1;
            return out;
        }
    }
    return out;
}

Determining Whether a String Contains a Substring

For this, I simply use the str_find_substr function defined above and extract the found field from the struct str_find_out output.

[[nodiscard]]
bool str_contains_substr(struct str_view s, struct str_view sub_s) {
    return str_find_substr(s, sub_s).found;
}

Determining Whether Two Strings Are Equal

This is straight forward. Check the lengths and then loop over the strings comparing each character.

[[nodiscard]]
bool str_eq(struct str_view s1, struct str_view s2) {
    if (s1.len != s2.len)
        return false;
    for (usize i = 0; i < s1.len; i++) {
        if (s1.data[i] != s2.data[i])
            return false;
    }
    return true;
}

Here is the same thing, but it ignores the casing. Adding 32 to an uppercase letter character converts it to the lowercase equivalent. This, of course, only works with ASCII encodings.

[[nodiscard]]
bool str_eq_ignore_case(struct str_view s1, struct str_view s2) {
    if (s1.len != s2.len)
        return false;
    for (usize i = 0; i < s1.len; i++) {
        u8 c1 = s1.data[i];
        u8 c2 = s2.data[i];
        if (c1 >= 65 && c1 <= 90)
            c1 += 32;
        if (c2 >= 65 && c2 <= 90)
            c2 += 32;
        if (c1 != c2)
            return false;
    }
    return true;
}

Checking Whether a String Starts With a Substring

This is similar to str_find_substr, but I didnā€™t want to use that, since implementing it from scratch would have been more efficient, because I can stop the search after finding the first character that is different.

[[nodiscard]]
bool str_starts_with(struct str_view s, struct str_view start) {
    if (s.len < start.len)
        return false;
    for (usize i = 0; i < start.len; i++) {
        if (s.data[i] != start.data[i])
            return false;
    }
    return true;
}

Splitting Strings

This one is a bit more complicated, since it involves heap allocations. At first, I implemented it in a way that required a memory allocation for each new split that was found. I later changed it and created this helper function:

u32 str_count_splits(struct str_view s, struct str_view delim) {
    u32 out = 0;
    usize delim_index = 0;
    for (usize i = 0; i < s.len; i++) {
        if (s.data[i] == delim.data[delim_index]) {
            delim_index++;
            if (delim_index == delim.len) {
                out++;
                delim_index = 0;
            }
        }
    }
    return out + 1;
}

This counts the number of splits before splitting, which is useful for allocating all the required memory upfront.

The actual splitting happens here:

struct str_split_out {
    struct str_view *splits;
    usize n_splits;
};
[[nodiscard]]
struct str_split_out str_split(struct str_view s, struct str_view delim) {
    struct str_split_out out = {0};
    if (delim.len == 0) {
        out.n_splits = 1;
        out.splits = malloc(sizeof(struct str_view));
        *out.splits = s;
        return out;
    }
    out.n_splits = str_count_splits(s, delim);
    out.splits = malloc(out.n_splits * sizeof(struct str_view));
    usize current_split_index = 0;
    usize last_split_index = 0;
    usize delim_index = 0;
    for (usize i = 0; i < s.len; i++) {
        if (s.data[i] == delim.data[delim_index]) {
            delim_index++;
            if (delim_index == delim.len) {
                struct str_view split = {0};
                split.data = s.data + last_split_index;
                split.len = i - last_split_index - delim.len + 1;
                out.splits[current_split_index] = split;
                current_split_index++;
                delim_index = 0;
                last_split_index = i + 1;
            }
        }
    }
    struct str_view split = {0};
    split.data = s.data + last_split_index;
    split.len = s.len - last_split_index;
    out.splits[out.n_splits - 1] = split;

    return out;
}

You do not have to understand it. In the final code I documented some parts with comments.

An empty delimiter does not split the string at all, it returns one split which contains the entire string. The delimiter can also be made of multiple characters (which is why I used a struct str_view instead of a char).

Slicing Strings

This should behave similarly to the String.slice function in JavaScript.

struct str_slice {
    i32 start;
    bool has_start;
    i32 end;
    bool has_end;
};
struct str_view str_slice(struct str_view s, struct str_slice slice) {
    struct str_view out = {0};
    if (slice.has_start) {
        if (slice.start < 0) {
            slice.start = s.len + slice.start;
        }
    } else {
        slice.start = 0;
    }
    if (slice.has_end) {
        if (slice.end < 0) {
            slice.end = s.len + slice.end;
        }
    } else {
        slice.end = s.len;
    }
    out.data = s.data + slice.start;
    out.len = slice.end - slice.start;
    return out;
}

I decided to use a struct for the input parameter, because this makes it easier to support optional arguments. You can simply initialize it like this

struct str_slice slice = {0};

And then, by default, both boolean fields are false. You can then set whatever fields you want. If you do not provide a start, the default is 0. If you do not provide an end, the default is the length of the string view.

start is inclusive, while end is exclusive. I could not use negative values for representing ā€œunsetā€ values. For one, I really dislike this practice, and also I wanted to give the user the option of passing negative values which reference the end of the string.

What I mean is that if you pass -1 as start, the function will start slicing at the last character of the string. -2 would be second to last.

This is really ergonomic whenever you only want to shave off some characters at the end (for example, the \r\n\r\n at the end of the head of an HTTP string).

A Final Remark

Another thing to note is that struct str_owned are usually NUL-terminated. The len field does not include the NUL-character, but it is there. This is a natural consequence of creating a struct str_owned from a struct str_view created with the STR macro. Also, I try to keep this convention in my utility functions.

This makes it very nice to work with functions in the standard library that require you to use NUL-terminated strings. Just pass the data field of the struct str_owned, and it acts as a normal C-string.

Conclusions

All the code is available on my codeberg repository.

I plan to add more utilities to this repository as time goes on, which is why the name has nothing to do with strings, but with libraries in general.

I also wrote some tests for the string functions.

The idea is that you can simply copy the code that you need and put it in your own projects. Feel free to use it however you please.

Also, I have recently started to employ so-called ā€œunity buildsā€ for my projects. Basically, I do not use header files for my own code. Everything lives in .c files and I #include them directly.

For very large scale projects, this is not good practice, but for personal the scale of almost all personal projects this is completely fine and has the nice benefit of significantly speeding up compilation times, since it completely skips the linking step, which takes the most time.

If you do not like unity builds, you can create your own header file to hide the API of the string library.

I will create another article if I ever add something to this repository.