From 1613dd5332c66bcebb3e1b51c7d74664d7e186a1 Mon Sep 17 00:00:00 2001
From: reachableceo <charles@turnsys.com>
Date: Wed, 2 Sep 2026 21:45:44 -0500
Subject: [PATCH] =?UTF-8?q?fix(doorman):=20stdbuf=20cannot=20exec=20shell?=
 =?UTF-8?q?=20functions=20=E2=80=94=20buffer=20inside=20decode=5Fstream=20?=
 =?UTF-8?q?[#355]?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Deployment rehearsal on ultix-field 2026-09-02 caught it live: the
listener showed active but decode_stream never ran (stdbuf exec's
binaries only). Line-buffering moved inside decode_stream; the od
stage keeps its own stdbuf.

Rehearsal note: https://projects.knownelement.com/issues/355
---
 bin/doorman.sh | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/bin/doorman.sh b/bin/doorman.sh
index 17b35af..28a8d16 100755
--- a/bin/doorman.sh
+++ b/bin/doorman.sh
@@ -61,7 +61,10 @@ usage() {
 #     behavior); Enter on an empty buffer emits nothing (defensive fix).
 # ---------------------------------------------------------------------------
 decode_stream() { # decode_stream <width> — od -t u1 records on stdin -> IDs
-    awk -v w="$1" '
+    # stdbuf exec's binaries only — it can NEVER wrap this function, so
+    # line-buffering happens here inside (piped awk block-buffers otherwise).
+    # shellcheck disable=SC2016  # awk program is single-quoted by design; '"$SELF"' re-opens quotes deliberately
+    stdbuf -oL awk -v w="$1" '
     {
         if (NF != w) { next }
         ti = w - 7; ci = w - 5; vi = w - 3
@@ -143,7 +146,7 @@ stream_device() {
     log "listening on $dev (record width $WIDTH)"
     # stdbuf: od/awk block-buffer when piped; scans must not lag behind
     stdbuf -oL od -v -A n -t u1 -w"$WIDTH" "$dev" 2>/dev/null \
-        | stdbuf -oL decode_stream "$WIDTH" \
+        | decode_stream "$WIDTH" \
         | while read -r bid; do
             process_scan "$bid"
         done