awk on Mac Is From 2020: 17 of 24 Probes Fail Silently

The awk on a Mac isn't gawk. It's Brian Kernighan's awk, often called BWK awk or "the one true awk", and the copy in macOS 26.4.1 says awk version 20200816. That's a snapshot from August 2020. Upstream has kept going since then: a few months later it stopped reading hex and nan strings as numbers, and the 2023 second edition added UTF-8 and a --csv mode. Apple's copy has none of those changes, and it never had gawk's extensions.

Scripts written on Linux assume gawk, so I wanted to know what actually happens when one of them runs on a Mac. On the Mac mini M4 that runs this blog, I ran 24 features that gawk or newer BWK awk have through three awks. Apple's awk got all 24 wrong. Seven stopped with an error. The other 17 exited 0 with wrong output, and those are the ones that worry me.

Which awk does a Mac have?

$ awk --version
awk version 20200816
$ which -a awk
/usr/bin/awk
$ strings /usr/bin/awk | grep PROJECT
@(#)PROGRAM:awk  PROJECT:awk-40

Unlike GNU make on Mac, which is an old GNU build behind an xcrun shim, /usr/bin/awk is a real 302,368-byte binary that ships with the OS. It doesn't come from the Command Line Tools. Apple's build is awk-40, and Apple has patched a few things. toupper("é") returns É, and [[:alpha:]] matches é. It also has the multibyte error that illegal byte sequence on Mac traced. Most of the code is still the 2020 snapshot.

To compare, I installed the two Homebrew options. brew install gawk took 4.0 seconds and installed GNU Awk 5.4.1 as gawk. brew install awk took 1.8 seconds and installed BWK awk 20260426, the current second edition.

24 features, three awks

Each probe is a one-line program with a known correct answer. A probe passed only if both the output and the exit code were right. I ran the UTF-8 probes under LC_ALL=en_US.UTF-8 and everything else under LC_ALL=C. I also ran 14 control probes for features all three should have, such as nextfile, delete arr, regex RS, interval expressions like a{3} and system() return codes. All three passed all 14, so the failures below aren't caused by the test setup.

Probe	macOS awk 20200816	BWK 20260426	Exit (Apple)
gensub(), strftime(), systime(), mktime(), asort(), and()	calling undefined function	same error	2
match(s, re, arr)	syntax error at source line 1	same error	2
IGNORECASE=1	/abc/ didn't match ABC	same	0
FPAT, FIELDWIDTHS	ignored, $2 empty	same	0
BEGINFILE	block never ran	same	0
PROCINFO["sorted_in"]	printed zeta bravo alpha mike	same	0
\s, \< \> in a regex	no match	same	0
0x10 in the program	printed 00	same	0
--csv	warning, then $2 empty	correct	0
length("é")	2	correct	0
substr("héllo",2,1)	half of é (invalid byte)	correct	0
index("héllo","l")	4	correct	0
printf "%c", 233	invalid byte	correct	0
printf "%-4s|" with é	é | (one space short)	correct	0
Input 0x10, nan, inf plus 0	16, nan, inf	correct (0)	0

The table groups probes that failed the same way. Counted one by one, Apple has 7 errors and 17 wrong results with exit 0.

The errors are the easy ones, because you see them. The silent results are a problem because BWK awk doesn't reject gawk syntax. It reads it as ordinary awk. IGNORECASE, FPAT, FIELDWIDTHS and PROCINFO are just variable names it has no reason to treat specially. BEGINFILE is an uninitialized variable used as a pattern, so it's always false and the block never runs. 0x10 parses as the number 0 next to a variable named x10, and awk concatenates them into "00". The --csv case does print unknown option --csv ignored to stderr, but it still exits 0 and splits on whitespace.

The input-number row is the one gawk users wouldn't expect. Apple's awk turns the field 0x10 into 16, and nan or inf into IEEE special values that then spread through any sum. Upstream changed this four months after Apple's snapshot. The BWK awk FIXES.1e log for December 8, 2020 says "Disallow hex data". The 2023 second edition notes in FIXES add the UTF-8 rule that "the length of a string of 3 emojis is 3, not 12".

The locale catch, even with gawk

The UTF-8 rows also depend on the locale. I ran length("é") with an empty environment, the way a launchd job or a bare cron line starts:

$ env -i /usr/bin/awk 'BEGIN{print length("é")}'
2
$ env -i gawk 'BEGIN{print length("é")}'
2
$ LC_ALL=en_US.UTF-8 gawk 'BEGIN{print length("é")}'
1

All three awks count bytes when there's no UTF-8 locale. Installing gawk fixes the missing functions, but a scheduled job still needs LC_ALL or LANG set. The scheduler that runs this blog starts its jobs with no LANG at all. More on what a job's environment does and doesn't have in launchd vs cron.

What people hit on GitHub

To check that these failures show up in real projects, I pulled every GitHub issue and PR that contains the exact error awk: calling undefined function. There were 38 on October 11, from 2011 to October 2026, in 35 repositories. 28 of them mention macOS, OS X, Darwin, Homebrew or /usr/bin/awk. Five don't name a gawk built-in (two are tests in an awk clone, one is a user-defined function and two have no function name), and the other 33 break down like this:

• strftime 10 · mktime 5 · gensub 5 · systime 4 · asort 3 · strtonum 2 · and 2 · or 1 · lshift 1
• 19 of the 33 are the time functions, and 14 of the 38 were opened in 2026

Dates are the usual way in, and that matches what I saw: there's no portable way to format a timestamp inside BWK awk. rocksdb #11494 fixed a release script that failed with calling undefined function strftime because a Mac "does not have gnu version of awk". kakoune #4847 was git blame breaking on macOS on systime. claude-plugins #1 shows how errors turn silent anyway. Its mktime() call sat behind 2>/dev/null and an || echo 0 fallback, and the issue says the scripts reported "plausible-looking wrong numbers instead of failing loudly". It found the same thing with date -d on Mac and stat -c. bash-yaml #36 took the blunt route and calls gawk whenever uname says Darwin.

How to install gawk on Mac without breaking anything

$ brew install gawk           # 4.0 s here, 7.7 MB plus mpfr 3.3 MB
$ gawk --version | head -1
GNU Awk 5.4.1, API 4.1, PMA Avon 8-g1, (GNU MPFR 4.2.2, GNU MP 6.3.0)
$ awk --version
awk version 20200816          # unchanged
$ export PATH="/opt/homebrew/opt/gawk/libexec/gnubin:$PATH"   # only if you want awk = gawk

The Homebrew gawk formula installs gawk and keeps an unprefixed awk in libexec/gnubin, so you choose whether plain awk changes. The awk formula doesn't ask. When I installed it, it linked /opt/homebrew/bin/awk, and which -a awk immediately listed it ahead of /usr/bin/awk. Every script on that PATH would have switched to a different awk. I ran brew unlink awk, called it by its Cellar path for the probes and removed it afterwards. If you want the second edition for --csv and UTF-8, it's a good awk, but install it knowing that it takes over the name.

In a script you share, don't rely on which awk the reader has. Call gawk by name and fail early if it's missing:

AWK=$(command -v gawk) || { echo "needs gawk: brew install gawk" >&2; exit 1; }
LC_ALL=en_US.UTF-8 "$AWK" -f report.awk data.txt

If the script has to run on stock macOS, stick to what Apple's awk does correctly. Use date outside awk to make timestamps and pass them in with -v, use tolower() on both sides instead of IGNORECASE, and split CSV with a real parser instead of FPAT.

FAQ

What version of awk does macOS use?

macOS 26.4.1 ships BWK awk, also called the one true awk, and awk --version prints "awk version 20200816". Apple's build is awk-40 at /usr/bin/awk. It is not gawk, so gawk functions like gensub, strftime and asort are missing, and features such as IGNORECASE and FPAT are silently ignored.

How do I install gawk on a Mac?

Run brew install gawk. Homebrew installs GNU Awk 5.4.1 as gawk and leaves /usr/bin/awk alone. To make the plain awk command run gawk, put /opt/homebrew/opt/gawk/libexec/gnubin at the front of PATH. Scripts started by launchd or cron don't load your shell profile, so they should call gawk by name.

Why does awk say "calling undefined function" on Mac?

The script uses a gawk-only function, most often strftime, mktime, systime, gensub or asort. macOS awk is BWK awk version 20200816, which never had them, so it stops with exit status 2. Install gawk with Homebrew and run the script with gawk, or rewrite the call using functions that BWK awk has.

Method: I ran 24 feature probes and 14 control probes (research/awk-mac-raw/probes.py, with each expected output in the file) through /usr/bin/awk 20200816, Homebrew's BWK awk 20260426 and gawk 5.4.1 on a Mac mini M4 (Mac16,10) running macOS 26.4.1 on October 11, 2026. A probe passed only if both output and exit code matched. UTF-8 probes ran under LC_ALL=en_US.UTF-8 and the rest under LC_ALL=C. Upstream change dates come from the FIXES and FIXES.1e files in the onetrueawk/awk repository. The GitHub figures are all 38 issues and PRs matching the exact string "awk: calling undefined function" on October 11, 2026. I read the function name from the issue text or comments, and "mentions macOS" is a keyword match, not a confirmed Mac report. I didn't test Intel Macs, other macOS versions or mawk.