Home / server / Questions / 1161951
Accepted
Zulakis
Asked: 2024-07-03 19:59:01 +0800 CST

How does sort compare strings?

  • 772

I would expect bash sort to compare strings like this:

  • Start at the first char (of both strings)
  • If the chars are equal, proceed to the next char
  • If they are unequal, return greater/lesser result to sort algorithm
  • If there are no more chars, return equals

For some reason, it seems like this is not a case.

Let's take the following input:

a
b
.
-

This is sorted by bash sort as

-
.
a
b

Now, for input

b.de
bb.de

I would expect the following sort result:

b.de
bb.de

Because the first char is equal, and for the second char, . comes before b (as seen in the first test).

For some reason, this is not the case, the strings are sorted like this:

bb.de
b.de

Why is sort behaving this way, and is there a way to make it behave "as expected"?

I have tested the same examples with python, and python sorts as expected.

bash

2 Answers

  1. Best Answer

    Sort by default does a locale aware sort which uses the lexicographical rules for your locale. see strcoll(3)

    ltrace(3) got me this:

    strcoll("b.de", "bb.de") = 20

    locate-aware comparisons seem to split strings into words and sort on that. as words nver start with '.' sort sees a 0 lenfgh words and puts that at the start of the list. however '.' is alloerd in wordd eg: "Jr." "Ph.D"

    if you require a byte-wise comparison instead export LC_COLLATE=C or LC_COLLATE=POSIX

    • 3

  2. I checked the coreutils package and if you dont provide any arguments, it looks as if it (eventually) uses the C strcmp routine. The only case that isn't true is where the values in lines can be interpreted as integers.

    The man page of which says:

    In glibc, as in most other implementations, the return value is the arithmetic result of subtracting the last compared byte in s2 from the last compared byte in s1. (If the two characters are equal, this difference is 0.)

    This means that the strcmp of bb.de and b.de really is down to the last character.

    That is if 'd' < 'e' which (in ascii at least) would be if 100 < 101 which is true.

    • -1