Byte conversion in my Caesar's cipher

Question

I was wondering if I am doing any useless conversions byte←→int, for example:

byte((int(ch-'A')+shift)%26 + 'A')

Converting ch-'A' to an int is because the shift argument can be negative (to implement the decode function). I couldn't figure out a simpler way to negate the operation using bytes.

package main

import (
    "bufio"
    "flag"
    "fmt"
    "os"
)

func main() {
    shift := flag.Int("shift", 13, "Cipher shift")
    decode := flag.Bool("decode", false, "Decode input")

    flag.Parse()

    scanner := bufio.NewScanner(os.Stdin)
    for scanner.Scan() {
        if *decode {
            fmt.Println(Decode(scanner.Text(), *shift))
        } else {
            fmt.Println(Encode(scanner.Text(), *shift))
        }
    }
}

func Encode(s string, shift int) string {
    return cipher(s, shift)
}

func Decode(s string, shift int) string {
    return cipher(s, -shift+26)
}

func cipher(s string, shift int) string {
    var line string
    for _, ch := range []byte(s) {
        if ch >= 'A' && ch <= 'Z' {
            ch = byte((int(ch-'A')+shift)%26 + 'A')
        } else if ch >= 'a' && ch <= 'z' {
            ch = byte((int(ch-'a')+shift)%26 + 'a')
        }
        line += string(ch)
    }
    return line
}

Answer 1

Flags

It's not very problematic for very short programs like this, but better get good habits: flags should be outside of all code blocks; so that you detect immediately if you have a flag naming conflict between different files.

var (
    shiftF = flag.Int("shift", 13, "Cipher shift")
    decodeF = flag.Bool("decode", false, "Decode input")
)

func main() {
    …
}

Code organization

You're checking for the value of the -decode flag every time you scan a line, and recomputing -shift+26 every time as well: not very efficient, nor super readable. As you noticed, the only difference between encoding and decoding is the value of shift. Why not do something like:

func main() {
    flag.Parse()

    // we could also directly change the value of the flag, but I find it
    // less readable — it's better to treat flag values as immutable.
    var shift int
    if *decode {
        shift = 26 - *shiftF
    } else {
        shift = *shiftF
    }

    scanner := bufio.NewScanner(os.Stdin)
    for scanner.Scan() {
        cipher(scanner.Text(), shift)
    }
}

Type conversions

bufio.Scanner holds bytes internally: the doc for Scanner.Bytes says:

The underlying array may point to data that will be overwritten by a subsequent call to Scan. It does no allocation.

while Scanner.Text indicates that it returns:

a newly allocated string.

So here, you allocate a string to copy bytes; and then you allocate a new slice of type []byte to copy this string, which is also an expensive operation.

So you're saving two conversions by not dealing with strings at all:

func main() {
    …
    for scanner.Scan() {
        fmt.Printf("%s\n", cipher(scanner.Bytes(), shift))
    }
}

func cipher(bytes []byte, shift int) []byte {
    var line []byte
    for _, b := range bytes {
        if b >= 'A' && b <= 'Z' {
            b = byte((int(b-'A')+shift)%26 + 'A')
        } else if b >= 'a' && b <= 'z' {
            b = byte((int(b-'a')+shift)%26 + 'a')
        }
        line = append(line, b)
    }
    return line
}

Note the slight change in the way you print the result: using fmt.Println on a []byte variable will print the numeric value of each byte, something like [116 104 101 32 103 97 109 101], so you have to use fmt.Printf to tell it to print it like a string.

You're also wondering whether the conversions from byte to int are inefficient: kind of, yes. Arithmetic operations work on bytes, so you could do instead:

shiftB := byte(shift)
b = ((b-'A')+shiftB)%26 + 'A'

However, this is just a simple type conversion: the performance gain from it will be much lower than when avoiding memory allocations, so you shouldn't worry about it too much =)

Optimizations

Initializing a []byte variable and appending char by char to it is not great: the slice is going to be re-sized every time it hits its maximum default size. It's even worse in your code: you're initializing a string variable and appending char by char to it, which will reallocate the string each time. In cipher, you know in advance which size line is going to be: the same as the original slice. So you can initialize directly with the correct allocated capacity:

    line := make([]byte, 0, len(bytes))

okay, now let's go even further: calling append at each iteration means that under the hood, there will be a check to see if the slice over capacity. We know it's not going to happen, because the sizes are the same: why not always allocate the exact amount of memory that we need and write at the right place directly?

func cipher(bytes []byte, shift int) []byte {
    shiftB := byte(shift)
    line := make([]byte, len(bytes))
    for i, b := range bytes {
        if b >= 'A' && b <= 'Z' {
            b = ((b-'A')+shiftB)%26 + 'A'
        } else if b >= 'a' && b <= 'z' {
            b = ((b-'a')+shiftB)%26 + 'a'
        }
        line[i] = b
    }
    return line
}

We can go even further by noticing that the bytes argument won't be used after we call cipher. So we can re-use this slice instead of allocating a new one!

func cipher(bytes []byte, shift int) []byte {
    shiftB := byte(shift)
    for i, b := range bytes {
        // shift b…
        bytes[i] = b
    }
    return bytes // optional: callers could simply reuse the argument
}

Final words

The code we obtain is more readable, definitely more efficient; and has the same behavior as yours. However, there are still two potential issues with it:

• Encoding: we're considering every byte of the input separately, as if it were ASCII; so it will probably not do what you want if the input is in UTF-8.
• Input size: if the input has very long lines, then using bufio.NewScanner is a bad idea; see for example this SO question.