cut

cut -d: -f2 pulls the second field out of each line. It is the smallest tool here and the one whose correctness lives entirely in its edge cases, which makes it a good place to be precise about what “splitting” means.

Split a:b:c on : and you get three fields. The rule that makes every other case fall out is: n delimiters produce n+1 fields, always. No exceptions, no collapsing, no skipping. Adopt that and the awkward inputs answer themselves. Adopt anything else and you spend the rest of the tool’s life patching it.

The program

const std = @import("std");

/// Field `n` of `line`, counting from 1 the way cut does. Returns null when
/// the line has fewer fields, which is different from the field being empty
/// and has to stay different.
fn field(line: []const u8, delimiter: u8, n: usize) ?[]const u8 {
    if (n == 0) return null;

    var it = std.mem.splitScalar(u8, line, delimiter);
    var i: usize = 1;
    while (it.next()) |part| : (i += 1) {
        if (i == n) return part;
    }
    return null;
}

pub fn main(init: std.process.Init) !void {
    var buf: [1024]u8 = undefined;
    var stdout_writer = std.Io.File.stdout().writerStreaming(init.io, &buf);
    const out = &stdout_writer.interface;

    const rows = [_][]const u8{
        "alice:30:london",
        "bob:25:paris",
        // An empty field in the middle: two delimiters with nothing between.
        "carol::berlin",
        // Fewer fields than asked for.
        "dave",
    };

    for ([_]usize{ 1, 2 }) |n| {
        try out.print("--- field {d}\n", .{n});
        for (rows) |row| {
            if (field(row, ':', n)) |value| {
                try out.print("{s} -> \"{s}\"\n", .{ row, value });
            } else {
                try out.print("{s} -> (no such field)\n", .{row});
            }
        }
        try out.writeAll("\n");
    }

    // Splitting never invents or loses a field: n delimiters give n+1 fields,
    // even when some of them are empty.
    var it = std.mem.splitScalar(u8, "carol::berlin", ':');
    var parts: usize = 0;
    while (it.next()) |_| parts += 1;
    try out.print("\"carol::berlin\" has {d} fields\n", .{parts});

    try out.flush();
}

What just happened

carol::berlin has three fields, and the middle one is empty. Two delimiters, three fields, exactly as the rule says. This is the case that separates a correct splitter from a plausible one: it is tempting to treat a run of delimiters as one, because that is what splitting on whitespace feels like it should do. For a delimited data format it is wrong, and it is wrong in the worst way, because it silently shifts every field after it. Field 3 of a CSV row becomes field 2 and the data is corrupted rather than rejected.

An empty field and a missing field printed differently. carol::berlin field 2 is ""; dave field 2 is (no such field). That distinction is the reason the function returns ?[]const u8 rather than []const u8. An empty string is a value that was there and was empty. Null is the absence of a field. Collapse them and a caller cannot tell “this record has no email” from “this record’s email is blank”, which is a difference that matters as soon as anyone acts on it.

No allocation and no copying. Each field is a slice of the line it came from, which is the same trick as the last chapter and the reason cut runs at the speed of the disk.

Check yourself

The real cut treats a line with no delimiter at all specially: cut -d: -f2 prints nothing for it, but cut -d: -f1 prints the whole line. Does the rule above explain that, or is it an exception?

It follows from the rule. Zero delimiters means one field, and that field is the whole line. So field 1 exists and is everything; field 2 does not exist. The program above agrees: dave gives "dave" for field 1 and no field 2. What actual cut adds on top is a flag, -s, to suppress such lines entirely, because “this line was not in the format I expected” is often what you meant.

If you have written C

C has strtok, and it is worth knowing why not to reach for it:

char *field = strtok(line, ":");
while (field) field = strtok(NULL, ":");

Three problems. It modifies the input, writing \0 over each delimiter, so the line you were handed is destroyed and you cannot use the original afterwards. It collapses runs of delimiters, so carol::berlin gives two fields, not three, which is the exact bug described above baked into the standard library. And it keeps its position in a static variable, so it is not reentrant and two loops using it at once corrupt each other.

strtok_r fixes only the third. The collapsing behaviour is by design, because strtok was written for whitespace, and using it on a data format is one of those habits that gets passed down.

Zig’s splitScalar returns slices of the original, changes nothing, and collapses nothing. Same job, and none of the three problems.

Next: printf, where a number has to become text.