The problem had to do with the "inplace" extension for gawk that enables in-place editing of files, ala perl or GNU sed -i. On BSD systems, I was getting failures from the test suite. E.g., on a regular system, inplace1.ok looks like: ----------------------- before gawk: inplace:47: warning: inplace_begin: disabling in-place editing for invalid FILENAME `-' stdin start is bar replaced? stdin end after ----------------------- On a BSD system, I get: ----------------------- before gawk: inplace:47: warning: inplace_begiafter abling in-place editing for invalid FILENAME `-' stdin start is bar replaced? stdin end ----------------------- There's some kind of buffering problem going on. The problem only occurs when stdout and stderr are redirected to the same file: command line here > _out 2>&1 I tried adding all kinds of calls to fflush in all kinds of places, but no luck. Finally, I had an "aha!" moment (an epiphany, if I'm using the expensive word correctly :-), and made this change: diff --git a/main.c b/main.c index 55983789..f505c71f 100644 --- a/main.c +++ b/main.c @@ -246,6 +246,10 @@ main(int argc, char **argv) if ((cp = getenv("GAWK_LOCALE_DIR")) != NULL) locale_dir = cp; + int flags = fcntl(fileno(stderr), F_GETFL, NULL); + flags |= O_APPEND; + (void) fcntl(fileno(stderr), F_SETFL, flags); + #if defined(LOCALEDEBUG) initial_locale = locale; #endif Forching stderr to be in append mode did the trick. WHY does that make it work? Beats me. It's not needed on any other system. Adventures In C Programming ... Thanks, Arnold arnold(a)skeeve.com wrote: