replace_regex: remove bash word boundary when detecting gnused

[Aleksanaa]

Bash linked against C libraries other than GLibc may not support GNU extensions of POSIX Extended Regular Regex. For example,

re='\bx'; [[ 'x' =~ $re ]] && echo "1"

does not output the same result on Linux/GLibc and macOS.

[Aleksanaa]

This fixes tests::stdlib::test_stdlib_src_tests_stdlib_text_replace_regex_ext_ab on macOS with gnused.

[b1ek]

tests pass in CI so i guess its ok.

they dont on my archlinux machine, but it could be a me problem since it is a new install, so i ask the next reviewer to check if it runs on their machine and block this if it doesn't

never mind, bc wasn't installed on my system. you really should check the tests on your machine however

[b1ek]

This fixes tests::stdlib::test_stdlib_src_tests_stdlib_text_replace_regex_ext_ab on macOS with gnused.

is there any difference with bsdsed? i think i heard something about macos being built on top of bsd

[Aleksanaa]

never mind, bc wasn't installed on my system. you really should check the tests on your machine however

It's okay on our Nixpkgs CI with both Linux and macOS now.

is there any difference with bsdsed? i think i heard something about macos being built on top of bsd

It's not related to sed, but the regex of bash itself. The command should be able to detect gnused, but it straightly fails on macOS, so it jumps to "this is not gnused", and try to pass an unsupported argument to it.

The biggest mistake here is \b which doesn't belong to POSIX Extended Regular Regex (ERE).Bash is said to be using ERE, but it actually depends on system libc for that. GLibc goes way more than ERE and implements a set of rules called GNU extensions. I cannot find the most related source but you can refer to findutils manual. For example, the following code will give different result on GNU/Linux and other platforms:

  #include <regex.h>
  #include <stdio.h>

  int main() {
      regex_t re;
      int ret = regcomp(&re, "\\bx\\b", REG_EXTENDED);
      if (ret) { printf("regcomp error\n"); return 1; }

      ret = regexec(&re, "x", 0, NULL, 0);
      regfree(&re);
      printf(ret == 0 ? "Match\n" : "No match\n");
      return 0;
  }

As you can see, there's no equivalent of word boundary in standard POSIX ERE. So the best option is simply not using them, and it's not necessary here, too.