From 9704f18343df02c98ca48efc96f8e51f8db8e1d6 Mon Sep 17 00:00:00 2001 From: Collin Funk Date: Sun, 8 Nov 2026 16:57:38 -0800 Subject: [PATCH] split: don't stop reading if a single filter stops accepting input * src/split.c (of_t): Add the done member. (lines_rr): Set the done member for the file if the filter has an ignorable error. Don't attempt subsequent writes if the done member for the file is set. * tests/split/filter.sh: Add a test case. * NEWS: Mention the bug fix. Discovered by Anthropic's OSS Scanner, as bug ANT-2026-NO8YC6GS. --- NEWS | 5 +++++ src/split.c | 22 ++++++++++++---------- tests/split/filter.sh | 14 ++++++++++++++ 3 files changed, 31 insertions(+), 10 deletions(-) diff --git a/NEWS b/NEWS index ffe134e76..9960c2299 100644 --- a/NEWS +++ b/NEWS @@ -205,6 +205,11 @@ GNU coreutils NEWS -*- outline -*- with a newline. [bug introduced in coreutils-9.2] + 'split --number=r/N --filter' no longer stops reading as soon as any one of + the filters stops accepting input. Now it continues until all filters stop + accepting input or the entire input is processed. + [bug introduced with --filter option in coreutils-8.13] + 'stty' again rejects empty speed arguments. Previously such arguments were accepted as value 0. [bug introduced in coreutils-9.8] diff --git a/src/split.c b/src/split.c index 71a097264..ea410dc9f 100644 --- a/src/split.c +++ b/src/split.c @@ -1131,6 +1131,9 @@ typedef struct of_info int ofd; FILE *ofile; pid_t opid; + + /* True if the filter no longer accepts input. */ + bool done; } of_t; enum @@ -1250,6 +1253,7 @@ lines_rr (intmax_t k, intmax_t n, char *buf, idx_t bufsize, of_t **filesp) files[i_file].ofd = OFD_NEW; files[i_file].ofile = NULL; files[i_file].opid = 0; + files[i_file].done = false; } i_file = 0; file_limit = false; @@ -1299,22 +1303,20 @@ lines_rr (intmax_t k, intmax_t n, char *buf, idx_t bufsize, of_t **filesp) { /* Secure file descriptor. */ file_limit |= ofile_open (files, i_file, n); - if (unbuffered) + if (! files[i_file].done) { /* Note writing to fd, rather than flushing the FILE gives an 8% performance benefit, due to reduced data copying. */ - if (full_write (files[i_file].ofd, bp, to_write) != to_write - && ! ignorable (errno)) + if (unbuffered + ? full_write (files[i_file].ofd, bp, to_write) == to_write + : fwrite (bp, to_write, 1, files[i_file].ofile) == 1) + wrote = true; + else if (ignorable (errno)) + files[i_file].done = true; + else error (EXIT_FAILURE, errno, "%s", quotef (files[i_file].of_name)); } - else if (fwrite (bp, to_write, 1, files[i_file].ofile) != 1 - && ! ignorable (errno)) - error (EXIT_FAILURE, errno, "%s", - quotef (files[i_file].of_name)); - - if (! ignorable (errno)) - wrote = true; if (file_limit) { diff --git a/tests/split/filter.sh b/tests/split/filter.sh index ba8f5e447..90ee3a5f2 100755 --- a/tests/split/filter.sh +++ b/tests/split/filter.sh @@ -71,6 +71,20 @@ for mode in '' 'r/'; do done done +seq 100 >in && +echo 1 >exp-xaa && +sed -n 'n;p' in >exp-xab || framework_failure_ +for mode in '' -u; do + rm -f x?? || framework_failure_ + split $mode -n r/2 \ + --filter='case $FILE in xaa) head -n1 ;; *) cat ;; esac >$FILE' in \ + >out 2>err || fail=1 + compare /dev/null out || fail=1 + compare /dev/null err || fail=1 + compare exp-xaa xaa || fail=1 + compare exp-xab xab || fail=1 +done + # Ensure that "endless" input _is_ processed for unbounded number of filters for buf in 1000 1000000; do returns_ 124 timeout .5 sh -c \