An awk one-liner that swaps date parts stops on a Mac:

$ printf 'a\nb\nc 2026-10-21\nd\n' > log.txt
$ awk '/-/ { $0 = gensub(/([0-9]+)-([0-9]+)-([0-9]+)/, "\\3/\\2/\\1", "g") } { print }' log.txt; echo "exit $?"
a
b
awk: calling undefined function gensub
 input record number 3, file log.txt
 source line number 1
exit 2

gensub is a gawk (GNU awk) function, according to gawk’s manual; I didn’t run gawk here. macOS ships a different awk:

$ awk --version
awk version 20200816
$ what /usr/bin/awk
/usr/bin/awk:
	PROGRAM:awk  PROJECT:awk-40
	PROGRAM:awk  PROJECT:awk-40

The error comes late, or not at all

Look at the output above again: a and b were printed before the error. This awk did not reject the script when it read it. It failed when line 3 was the first line to reach gensub.

So the same script passes when no line reaches the call:

$ printf 'a\nb\n' | awk '/-/ { $0 = gensub(/x/, "y", "g") } { print }'; echo "exit $?"
a
b
exit 0

A script tested on input without dates would look fine on this Mac. And in a pipe, the exit status you see is the last command’s:

$ awk '/-/ { $0 = gensub(/([0-9]+)-([0-9]+)-([0-9]+)/, "\\3/\\2/\\1", "g") } { print }' log.txt | cat; echo "exit $?"
awk: calling undefined function gensub
 input record number 3, file log.txt
 source line number 1
a
b
exit 0

Output stopped at line 2, and the pipeline still says 0. (set -o pipefail in bash or zsh makes a failing awk fail the pipeline.)

Two other gawk-isms failed the same way here. strftime gives the same “calling undefined function” at run time; the three-argument match(s, re, arr) is a syntax error before any input is read:

$ echo 'c 2026-10-21' | awk '{ print strftime("%Y") }'; echo "exit $?"
awk: calling undefined function strftime
 input record number 1, file 
 source line number 1
exit 2
$ echo 'c 2026-10-21' | awk '{ if (match($0, /([0-9]+)-([0-9]+)/, m)) print m[1] }'; echo "exit $?"
awk: syntax error at source line 1
 context is
	{ if (match($0, >>>  /([0-9]+)-([0-9]+)/, <<< 
awk: illegal statement at source line 1
awk: illegal statement at source line 1
exit 2

No backreferences needed: gsub

If the replacement doesn’t use \\1, \\2…, gsub does the job. It changes its third argument in place (or $0) and returns the number of replacements, so copy first if you need the original:

$ echo 'a.b.c' | awk '{ s = $0; n = gsub(/\./, "_", s); print n, s, $0 }'
2 a_b_c a.b.c

& in the replacement is the matched text, and \\& is a literal &:

$ echo 'cat bat' | awk '{ gsub(/[cb]at/, "<&>"); print }'
<cat> <bat>
$ echo 'cat bat' | awk '{ gsub(/[cb]at/, "\\&"); print }'
& &

Backreferences in awk: a match / substr loop

This awk has no capture groups. Match the whole piece, cut it out with substr, and take it apart yourself (here with split). Save as swap.awk:

{
  out = ""
  s = $0
  while (match(s, /[0-9]+-[0-9]+-[0-9]+/)) {
    split(substr(s, RSTART, RLENGTH), p, "-")
    out = out substr(s, 1, RSTART - 1) p[3] "/" p[2] "/" p[1]
    s = substr(s, RSTART + RLENGTH)
  }
  print out s
}

$ printf 'start 2026-10-21 end 2025-01-02\nno date here\n' > dates.txt
$ awk -f swap.awk dates.txt; echo "exit $?"
start 21/10/2026 end 02/01/2025
no date here
exit 0

match sets RSTART (where the match starts) and RLENGTH (its length). The loop keeps the text before each match, adds the rewritten date, and continues on the rest of the line.

The regex must not match an empty string. If it can, match succeeds with RLENGTH 0, s never gets shorter, and the loop would never end. With a counter capping it at 5 rounds:

$ echo 'ab12' | awk '{ s = $0; n = 0; while (match(s, /[0-9]*/) && n < 5) { n++; print n, RSTART, RLENGTH; s = substr(s, RSTART + RLENGTH) } }'
1 1 0
2 1 0
3 1 0
4 1 0
5 1 0

Use + rather than * for the part that must be there.

Or leave awk: sed -E and perl -pe

If the substitution is the whole job, macOS’s BSD sed does it with -E (unescaped groups and \1 backreferences):

$ sed -E 's|([0-9]+)-([0-9]+)-([0-9]+)|\3/\2/\1|g' dates.txt; echo "exit $?"
start 21/10/2026 end 02/01/2025
no date here
exit 0

and so does perl ($1, $2…):

$ perl -pe 's|(\d+)-(\d+)-(\d+)|$3/$2/$1|g' dates.txt; echo "exit $?"
start 21/10/2026 end 02/01/2025
no date here
exit 0

| as the delimiter avoids escaping the / in the replacement.

gensub’s “replace only the Nth match” form (a number instead of "g") has a sed equivalent, a number after the last delimiter:

$ echo 'x x x' | sed 's/x/y/2'
x y x

Checklist

  • awk: calling undefined function gensub (exit 2): /usr/bin/awk on macOS is awk version 20200816 (awk-40), not gawk.
  • In my tests the error came only when a line reached the call. Lines before it were already printed; input that never reached it passed with exit 0. Test with input that hits every branch.
  • In a pipe the error is hidden behind the last command’s exit 0 — use set -o pipefail, or check the awk step alone.
  • No backreferences: gsub(re, repl[, target]); & is the match, \\& a literal &. It edits in place and returns a count.
  • Backreferences: a while (match(...)) loop with substr, RSTART, RLENGTH. The regex must not be able to match an empty string.
  • Just substituting: sed -E 's|(..)|\1|g' or perl -pe 's|(..)|$1|g'; Nth match: sed 's/x/y/2'.
  • Same family of failures here: strftime (undefined function), and match(s, re, arr) (syntax error).

Tested on macOS 26.5.2 (Apple silicon) with /usr/bin/awk (awk version 20200816, PROGRAM:awk PROJECT:awk-40), /usr/bin/sed (text_cmds-199) and perl 5.34.1, in zsh 5.9 (arm64-apple-darwin25.0). The pipefail line was checked in zsh 5.9 and bash 3.2.57 (pipeline exit 2 in both); everything else ran in zsh only. gawk and Linux were not tested.