BEGINFILE/ENDFILE vs. FNR==1: Choosing Per‑File State Logic in Awk

Short answer

Use BEGINFILE/ENDFILE only if you can guarantee gawk on every target system. If awk might be POSIX awk, mawk, or BusyBox awk, use the FNR==1 idiom — it is the only option that works everywhere. The performance difference is negligible; this is purely a portability and error-handling decision.

What is confirmed

• BEGINFILE and ENDFILE are gawk extensions (since gawk 4.0). They do not exist in POSIX awk, mawk, or BusyBox awk — a script using them fails with a syntax error there.
• FNR resets to zero at the start of each input file in every standard awk, so FNR==1 is fully portable.
• For an empty file, FNR never reaches 1, so FNR==1 never fires — per-file initialization is silently skipped unless you add explicit handling.
• When a file cannot be opened, gawk still runs BEGINFILE/ENDFILE with ERRNO set, giving you a hook to report the failure. POSIX-style awks typically just skip the unreadable file with no per-file hook at all.

Decision by scenario

Guaranteed gawk

If the deployment environment pins gawk (for example, a controlled container image), the extension version is clearer and safer:

BEGINFILE {
    if (ERRNO != "") {
        printf "skipping %s: %s\n", FILENAME, ERRNO > "/dev/stderr"
        nextfile
    }
    count = 0
}
{ count++ }
ENDFILE { printf "%s: %d records\n", FILENAME, count }

You get initialization, teardown, and open-failure reporting without extra bookkeeping.

Unknown or mixed awk

If the script must run anywhere, use FNR==1 and emulate the teardown in END or by tracking filename changes:

FNR == 1 {
    if (NR != 1) report(prev, count)   # close out previous file
    prev = FILENAME; count = 0
}
{ count++ }
END { if (NR > 0) report(prev, count) }
function report(f, c) { printf "%s: %d records\n", f, c }

The costs: empty files produce no report at all (add a record-count check if that matters), and unreadable files are skipped silently unless you wrap the awk call in shell logic that checks exit status or pre-validates the file list.

One pitfall to avoid

Do not mix both mechanisms in one script. Under gawk, BEGINFILE and FNR==1 both fire for the first record of each file, causing double initialization or double counting.

Verify before committing

Build a test set with a normal file, an empty file, and a file with read permission removed, then run the script under each awk you might encounter:

mkdir t && printf 'a\nb\n' > t/ok.txt && : > t/empty.txt \
  && printf 'x\n' > t/denied.txt && chmod 000 t/denied.txt
for a in gawk mawk busybox\ awk; do
  echo "== $a =="; $a -f script.awk t/*.txt
done

Confirm: no syntax errors under any variant, the empty file behaves as intended, and the unreadable file is either reported (gawk + ERRNO) or acceptably skipped.

Bottom line

The deciding question is the first one in your list: is gawk guaranteed? If yes, take the cleaner hooks and free error reporting. If no — or if you cannot answer it with certainty — accept the extra verbosity of FNR==1 plus explicit empty-file handling. Maintenance overhead is a one-time cost; a syntax error on a production minimal system is not.