GNU coreutils NEWS -*- outline -*-

* Noteworthy changes in release ?.? (????-??-??) [?]

** Bug fixes

cp fixes support for --update=none-fail, which would have been

rejected as an invalid option.

[bug introduced in coreutils-9.5]

ls and printf fix shell quoted output in the edge case of escaped

first and last characters, and single quotes in the string.

[bug introduced in coreutils-8.26]

mv works again with macFUSE file systems. Previously it would

have exited with a "Function not implemented" error.

[bug introduced in coreutils-8.28]

'shuf' generates more-random output when the output is small.

[bug introduced in coreutils-8.6]

'tail -c 4096 /dev/zero' no longer loops forever.

[This bug was present in "the beginning".]

** Changes in behavior

ls's -f option now simply acts like -aU, instead of also ignoring

some earlier options. For example 'ls -fl' and 'ls -lf' are now

equivalent because -f no longer ignores an earlier -l. The new

behavior is more orthogonal and is compatible with FreeBSD.

stat -f -c%T now reports the "fuseblk" file system type as "fuse",

given that there is no longer a distinct "ctl" fuse variant file system.

** Improvements

sort operates more efficiently when used on pseudo files with

an apparent size of 0, like those in /proc.

stat and tail now know about the "bcachefs", and "pidfs" file system types.

stat -f -c%T now reports the file system type,

and tail -f uses inotify for these file systems.

wc now reads a minimum of 256KiB at a time.

This was previously 16KiB and increasing to 256KiB was seen to increase

wc -l performance by about 10% when reading cached files on modern systems.

* Noteworthy changes in release 9.5 (2024-03-28) [stable]

** Bug fixes

chmod -R now avoids a race where an attacker may replace a traversed file

with a symlink, causing chmod to operate on an unintended file.

[This bug was present in "the beginning".]

cp, mv, and install no longer issue spurious diagnostics like "failed

to preserve ownership" when copying to GNU/Linux CIFS file systems.

They do this by working around some Linux CIFS bugs.

cp --no-preserve=mode will correctly maintain set-group-ID bits

for created directories. Previously on systems that didn't support ACLs,

cp would have reset the set-group-ID bit on created directories.

[bug introduced in coreutils-8.20]

join and uniq now support multi-byte characters better.

For example, 'join -tX' now works even if X is a multi-byte character,

and both programs now treat multi-byte characters like U+3000

IDEOGRAPHIC SPACE as blanks if the current locale treats them so.

numfmt options like --suffix no longer have an arbitrary 127-byte limit.

[bug introduced with numfmt in coreutils-8.21]

mktemp with --suffix now better diagnoses templates with too few X's.

Previously it conflated the insignificant --suffix in the error.

[bug introduced in coreutils-8.1]

sort again handles thousands grouping characters in single-byte locales

where the grouping character is greater than CHAR_MAX. For e.g. signed

character platforms with a 0xA0 (aka &nbsp) grouping character.

[bug introduced in coreutils-9.1]

split --line-bytes with a mixture of very long and short lines

no longer overwrites the heap (CVE-2024-0684).

[bug introduced in coreutils-9.2]

tail no longer mishandles input from files in /proc and /sys file systems,

on systems with a page size larger than the stdio BUFSIZ.

[This bug was present in "the beginning".]

timeout avoids a narrow race condition, where it might kill arbitrary

processes after a failed process fork.

[bug introduced with timeout in coreutils-7.0]

timeout avoids a narrow race condition, where it might fail to

kill monitored processes immediately after forking them.

[bug introduced with timeout in coreutils-7.0]

wc no longer fails to count unprintable characters as parts of words.

[bug introduced in textutils-2.1]

** Changes in behavior

base32 and base64 no longer require padding when decoding.

Previously an error was given for non padded encoded data.

base32 and base64 have improved detection of corrupted encodings.

Previously encodings with non zero padding bits were accepted.

basenc --base16 -d now supports lower case hexadecimal characters.

Previously an error was given for lower case hex digits.

cp --no-clobber, and mv -n no longer exit with failure status if

existing files are encountered in the destination. Instead they revert

to the behavior from before v9.2, silently skipping existing files.

ls --dired now implies long format output without hyperlinks enabled,

and will take precedence over previously specified formats or hyperlink mode.

numfmt will accept lowercase 'k' to indicate Kilo or Kibi units on input,

and uses lowercase 'k' when outputting such units in '--to=si' mode.

pinky no longer tries to canonicalize the user's login location by default,

rather requiring the new --lookup option to enable this often slow feature.

wc no longer ignores encoding errors when counting words.

Instead, it treats them as non white space.

** New features

chgrp now accepts the --from=OWNER:GROUP option to restrict changes to files

with matching current OWNER and/or GROUP, as already supported by chown(1).

chmod adds support for -h, -H,-L,-P, and --dereference options, providing

more control over symlink handling. This supports more secure handling of

CLI arguments, and is more consistent with chown, and chmod on other systems.

cp now accepts the --keep-directory-symlink option (like tar), to preserve

and follow existing symlinks to directories in the destination.

cp and mv now accept the --update=none-fail option, which is similar

to the --no-clobber option, except that existing files are diagnosed,

and the command exits with failure status if existing files.

The -n,--no-clobber option is best avoided due to platform differences.

env now accepts the -a,--argv0 option to override the zeroth argument

of the command being executed.

mv now accepts an --exchange option, which causes the source and

destination to be exchanged. It should be combined with

--no-target-directory (-T) if the destination is a directory.

The exchange is atomic if source and destination are on a single

file system that supports atomic exchange; --exchange is not yet

supported in other situations.

od now supports printing IEEE half precision floating point with -t fH,

or brain 16 bit floating point with -t fB, where supported by the compiler.

tail now supports following multiple processes, with repeated --pid options.

** Improvements

cp,mv,install,cat,split now read and write a minimum of 256KiB at a time.

This was previously 128KiB and increasing to 256KiB was seen to increase

throughput by 10-20% when reading cached files on modern systems.

env,kill,timeout now support unnamed signals. kill(1) for example now

supports sending such signals, and env(1) will list them appropriately.

SELinux operations in file copy operations are now more efficient,

avoiding unneeded MCS/MLS label translation.

sort no longer dynamically links to libcrypto unless -R is used.

This decreases startup overhead in the typical case.

wc is now much faster in single-byte locales and somewhat faster in

multi-byte locales.

* Noteworthy changes in release 9.4 (2023-08-29) [stable]

** Bug fixes

On GNU/Linux s390x and alpha, programs like 'cp' and 'ls' no longer

fail on files with inode numbers that do not fit into 32 bits.

[This bug was present in "the beginning".]

'b2sum --check' will no longer read unallocated memory when

presented with malformed checksum lines.

[bug introduced in coreutils-9.2]

'cp --parents' again succeeds when preserving mode for absolute directories.

Previously it would have failed with a "No such file or directory" error.

[bug introduced in coreutils-9.1]

'cp --sparse=never' will avoid copy-on-write (reflinking) and copy offloading,

to ensure no holes present in the destination copy.

[bug introduced in coreutils-9.0]

cksum again diagnoses read errors in its default CRC32 mode.

[bug introduced in coreutils-9.0]

'cksum --check' now ensures filenames with a leading backslash character

are escaped appropriately in the status output.

This also applies to the standalone checksumming utilities.

[bug introduced in coreutils-8.25]

dd again supports more than two multipliers for numbers.

Previously numbers of the form '1024x1024x32' gave "invalid number" errors.

[bug introduced in coreutils-9.1]

factor, numfmt, and tsort now diagnose read errors on the input.

[This bug was present in "the beginning".]

'install --strip' now supports installing to files with a leading hyphen.

Previously such file names would have caused the strip process to fail.

[This bug was present in "the beginning".]

ls now shows symlinks specified on the command line that can't be traversed.

Previously a "Too many levels of symbolic links" diagnostic was given.

[This bug was present in "the beginning".]

pinky, uptime, users, and who no longer misbehave on 32-bit GNU/Linux

platforms like x86 and ARM where time_t was historically 32 bits.

Also see the new --enable-systemd option mentioned below.

[bug introduced in coreutils-9.0]

'pr --length=1 --double-space' no longer enters an infinite loop.

[This bug was present in "the beginning".]

shred again operates on Solaris when built for 64 bits.

Previously it would have exited with a "getrandom: Invalid argument" error.

[bug introduced in coreutils-9.0]

tac now handles short reads on its input. Previously it may have exited

erroneously, especially with large input files with no separators.

[This bug was present in "the beginning".]

'uptime' no longer incorrectly prints "0 users" on OpenBSD,

and is being built again on FreeBSD and Haiku.

[bugs introduced in coreutils-9.2]

'wc -l' and 'cksum' no longer crash with an "Illegal instruction" error

on x86 Linux kernels that disable XSAVE YMM. This was seen on Xen VMs.

[bug introduced in coreutils-9.0]

** Changes in behavior

'cp -v' and 'mv -v' will no longer output a message for each file skipped

due to -i, or -u. Instead they only output this information with --debug.

I.e., 'cp -u -v' etc. will have the same verbosity as before coreutils-9.3.

'cksum -b' no longer prints base64-encoded checksums. Rather that

short option is reserved to better support emulation of the standalone

checksum utilities with cksum.

'mv dir x' now complains differently if x/dir is a nonempty directory.

Previously it said "mv: cannot move 'dir' to 'x/dir': Directory not empty",

where it was unclear whether 'dir' or 'x/dir' was the problem.

Now it says "mv: cannot overwrite 'x/dir': Directory not empty".

Similarly for other renames where the destination must be the problem.

[problem introduced in coreutils-6.0]

** Improvements

cp, mv, and install now avoid copy_file_range on linux kernels before 5.3

irrespective of which kernel version coreutils is built against,

reinstating that behavior from coreutils-9.0.

comm, cut, join, od, and uniq will now exit immediately upon receiving a

write error, which is significant when reading large / unbounded inputs.

split now uses more tuned access patterns for its potentially large input.

This was seen to improve throughput by 5% when reading from SSD.

split now supports a configurable $TMPDIR for handling any temporary files.

tac now falls back to '/tmp' if a configured $TMPDIR is unavailable.

'who -a' now displays the boot time on Alpine Linux, OpenBSD,

Cygwin, Haiku, and some Android distributions

'uptime' now succeeds on some Android distributions, and now counts

VM saved/sleep time on GNU (Linux, Hurd, kFreeBSD), NetBSD, OpenBSD,

Minix, and Cygwin.

On GNU/Linux platforms where utmp-format files have 32-bit timestamps,

pinky, uptime, and who can now work for times after the year 2038,

so long as systemd is installed, you configure with a new, experimental

option --enable-systemd, and you use the programs without file arguments.

(For example, with systemd 'who /var/log/wtmp' does not work because

systemd does not support the equivalent of /var/log/wtmp.)

* Noteworthy changes in release 9.3 (2023-04-18) [stable]

** Bug fixes

cp --reflink=auto (the default), mv, and install

will again fall back to a standard copy in more cases.

Previously copies could fail with permission errors on

more restricted systems like android or containers etc.

[bug introduced in coreutils-9.2]

cp --recursive --backup will again operate correctly.

Previously it may have issued "File exists" errors when

it failed to appropriately rename files being replaced.

[bug introduced in coreutils-9.2]

date --file and dircolors will now diagnose a failure to read a file.

Previously they would have silently ignored the failure.

[This bug was present in "the beginning".]

md5sum --check again correctly prints the status of each file checked.

Previously the status for files was printed as 'OK' once any file had passed.

This also applies to cksum, sha*sum, and b2sum.

[bug introduced in coreutils-9.2]

wc will now diagnose if any total counts have overflowed.

[This bug was present in "the beginning".]

`wc -c` will again correctly update the read offset of inputs.

Previously it deduced the size of inputs while leaving the offset unchanged.

[bug introduced in coreutils-8.27]

Coreutils programs no longer fail for timestamps past the year 2038

on obsolete configurations with 32-bit signed time_t, because the

build procedure now rejects these configurations.

[This bug was present in "the beginning".]

** Changes in behavior

'cp -n' and 'mv -n' now issue an error diagnostic if skipping a file,

to correspond with -n inducing a nonzero exit status as of coreutils 9.2.

Similarly 'cp -v' and 'mv -v' will output a message for each file skipped

due to -n, -i, or -u.

** New features

cp and mv now support --update=none to always skip existing files

in the destination, while not affecting the exit status.

This is equivalent to the --no-clobber behavior from before v9.2.

* Noteworthy changes in release 9.2 (2023-03-20) [stable]

** Bug fixes

'comm --output-delimiter="" --total' now delimits columns in the total

line with the NUL character, consistent with NUL column delimiters in

the rest of the output. Previously no delimiters were used for the

total line in this case.

[bug introduced with the --total option in coreutils-8.26]

'cp -p' no longer has a security hole when cloning into a dangling

symbolic link on macOS 10.12 and later.

[bug introduced in coreutils-9.1]

'cp -rx / /mnt' no longer complains "cannot create directory /mnt/".

[bug introduced in coreutils-9.1]

cp, mv, and install avoid allocating too much memory, and possibly

triggering "memory exhausted" failures, on file systems like ZFS,

which can return varied file system I/O block size values for files.

[bug introduced in coreutils-6.0]

cp, mv, and install now immediately acknowledge transient errors

when creating copy-on-write or cloned reflink files, on supporting

file systems like XFS, BTRFS, APFS, etc.

Previously they would have tried again with other copy methods

which may have resulted in data corruption.

[bug introduced in coreutils-7.5 and enabled by default in coreutils-9.0]

cp, mv, and install now handle ENOENT failures across CIFS file systems,

falling back from copy_file_range to a better supported standard copy.

[issue introduced in coreutils-9.0]

'mv --backup=simple f d/' no longer mistakenly backs up d/f to f~.

[bug introduced in coreutils-9.1]

rm now fails gracefully when memory is exhausted.

Previously it may have aborted with a failed assertion in some cases.

[This bug was present in "the beginning".]

rm -d (--dir) now properly handles unreadable empty directories.

E.g., before, this would fail to remove d: mkdir -m0 d; src/rm -d d

[bug introduced in v8.19 with the addition of this option]

runcon --compute no longer looks up the specified command in the $PATH

so that there is no mismatch between the inspected and executed file.

[bug introduced when runcon was introduced in coreutils-6.9.90]

'sort -g' no longer infloops when given multiple NaNs on platforms

like x86_64 where 'long double' has padding bits in memory.

Although the fix alters sort -g's NaN ordering, that ordering has

long been documented to be platform-dependent.

[bug introduced 1999-05-02 and only partly fixed in coreutils-8.14]

stty ispeed and ospeed options no longer accept and silently ignore

invalid speed arguments, or give false warnings for valid speeds.

Now they're validated against both the general accepted set,

and the system supported set of valid speeds.

[This bug was present in "the beginning".]

stty now wraps output appropriately for the terminal width.

Previously it may have output 1 character too wide for certain widths.

[bug introduced in coreutils-5.3]

tail --follow=name works again with non seekable files. Previously it

exited with an "Illegal seek" error when such a file was replaced.

[bug introduced in fileutils-4.1.6]

'wc -c' will again efficiently determine the size of large files

on all systems. It no longer redundantly reads data from certain

sized files larger than SIZE_MAX.

[bug introduced in coreutils-8.24]

** Changes in behavior

Programs now support the new Ronna (R), and Quetta (Q) SI prefixes,

corresponding to 10^27 and 10^30 respectively,

along with their binary counterparts Ri (2^90) and Qi (2^100).

In some cases (e.g., 'sort -h') these new prefixes simply work;

in others, where they exceed integer width limits, they now elicit

the same integer overflow diagnostics as other large prefixes.

'cp --reflink=always A B' no longer leaves behind a newly created

empty file B merely because copy-on-write clones are not supported.

'cp -n' and 'mv -n' now exit with nonzero status if they skip their

action because the destination exists, and likewise for 'cp -i',

'ln -i', and 'mv -i' when the user declines. (POSIX specifies this

for 'cp -i' and 'mv -i'.)

cp, mv, and install again read in multiples of the reported block size,

to support unusual devices that may have this constraint.

[behavior inadvertently changed in coreutils-7.2]

du --apparent now counts apparent sizes only of regular files and

symbolic links. POSIX does not specify the meaning of apparent

sizes (i.e., st_size) for other file types, and counting those sizes

could cause confusing and unwanted size mismatches.

'ls -v' and 'sort -V' go back to sorting ".0" before ".A",

reverting to the behavior in coreutils-9.0 and earlier.

This behavior is now documented.

ls --color now matches a file extension case sensitively

if there are different sequences defined for separate cases.

printf unicode \uNNNN, \UNNNNNNNN syntax, now supports all valid

unicode code points. Previously is was restricted to the C

universal character subset, which restricted most points <= 0x9F.

runcon now exits with status 125 for internal errors. Previously upon

internal errors it would exit with status 1, which was less distinguishable

from errors from the invoked command.

'split -n N' now splits more evenly when the input size is not a

multiple of N, by creating N output files whose sizes differ by at

most 1 byte. Formerly, it did this only when the input size was

less than N.

'stat -c %s' now prints sizes as unsigned, consistent with 'ls'.

** New Features

cksum now accepts the --base64 (-b) option to print base64-encoded

checksums. It also accepts/checks such checksums.

cksum now accepts the --raw option to output a raw binary checksum.

No file name or other information is output in this mode.

cp, mv, and install now accept the --debug option to

print details on how a file is being copied.

factor now accepts the --exponents (-h) option to print factors

in the form p^e, rather than repeating the prime p, e times.

ls now supports the --time=modification option, to explicitly

select the default mtime timestamp for display and sorting.

mv now supports the --no-copy option, which causes it to fail when

asked to move a file to a different file system.

split now accepts options like '-n SIZE' that exceed machine integer

range, when they can be implemented as if they were infinity.

split -n now accepts piped input even when not in round-robin mode,

by first copying input to a temporary file to determine its size.

wc now accepts the --total={auto,never,always,only} option

to give explicit control over when the total is output.

** Improvements

cp --sparse=auto (the default), mv, and install,

will use the copy_file_range syscall now also with sparse files.

This may be more efficient, by avoiding user space copies,

and possibly employing copy offloading or reflinking,

for the non sparse portion of such sparse files.

On macOS, cp creates a copy-on-write clone in more cases.

Previously cp would only do this when preserving mode and timestamps.

date --debug now diagnoses if multiple --date or --set options are

specified, as only the last specified is significant in that case.

rm outputs more accurate diagnostics in the presence of errors

when removing directories. For example EIO will be faithfully

diagnosed, rather than being conflated with ENOTEMPTY.

tail --follow=name now works with single non regular files even

when their modification time doesn't change when new data is available.

Previously tail would not show any new data in this case.

tee -p detects when all remaining outputs have become broken pipes, and

exits, rather than waiting for more input to induce an exit when written.

tee now handles non blocking outputs, which can be seen for example with

telnet or mpirun piping through tee to a terminal.

Previously tee could truncate data written to such an output and fail,

and also potentially output a "Resource temporarily unavailable" error.

* Noteworthy changes in release 9.1 (2022-04-15) [stable]

** Bug fixes

chmod -R no longer exits with error status when encountering symlinks.

All files would be processed correctly, but the exit status was incorrect.

[bug introduced in coreutils-9.0]

If 'cp -Z A B' checks B's status and some other process then removes B,

cp no longer creates B with a too-generous SELinux security context

before adjusting it to the correct value.

[bug introduced in coreutils-8.17]

'cp --preserve=ownership A B' no longer ignores the umask when creating B.

Also, 'cp --preserve-xattr A B' is less likely to temporarily chmod u+w B.

[bug introduced in coreutils-6.7]

On macOS, 'cp A B' no longer miscopies when A is in an APFS file system

and B is in some other file system.

[bug introduced in coreutils-9.0]

On macOS, fmt no longer corrupts multi-byte characters

by misdetecting their component bytes as spaces.

[This bug was present in "the beginning".]

'id xyz' now uses the name 'xyz' to determine groups, instead of xyz's uid.

[bug introduced in coreutils-8.22]

'ls -v' and 'sort -V' no longer mishandle corner cases like "a..a" vs "a.+"

or lines containing NULs. Their behavior now matches the documentation

for file names like ".m4" that consist entirely of an extension,

and the documentation has been clarified for unusual cases.

[bug introduced in coreutils-7.0]

On macOS, 'mv A B' no longer fails with "Operation not supported"

when A and B are in the same tmpfs file system.

[bug introduced in coreutils-9.0]

'mv -T --backup=numbered A B/' no longer miscalculates the backup number

for B when A is a directory, possibly inflooping.

[bug introduced in coreutils-6.3]

** Changes in behavior

cat now uses the copy_file_range syscall if available, when doing

simple copies between regular files. This may be more efficient, by avoiding

user space copies, and possibly employing copy offloading or reflinking.

chown and chroot now warn about usages like "chown root.root f",

which have the nonstandard and long-obsolete "." separator that

causes problems on platforms where user names contain ".".

Applications should use ":" instead of ".".

cksum no longer allows abbreviated algorithm names,

so that forward compatibility and robustness is improved.

date +'%-N' now suppresses excess trailing digits, instead of always

padding them with zeros to 9 digits. It uses clock_getres and

clock_gettime to infer the clock resolution.

dd conv=fsync now synchronizes output even after a write error,

and similarly for dd conv=fdatasync.

dd now counts bytes instead of blocks if a block count ends in "B".

For example, 'dd count=100KiB' now copies 100 KiB of data, not

102,400 blocks of data. The flags count_bytes, skip_bytes and

seek_bytes are therefore obsolescent and are no longer documented,

though they still work.

ls no longer colors files with capabilities by default, as file-based

capabilities are rarely used, and lookup increases processing per file by

about 30%. It's best to use getcap [-r] to identify files with capabilities.

ls no longer tries to automount files, reverting to the behavior

before the statx() call was introduced in coreutils-8.32.

stat no longer tries to automount files by default, reverting to the

behavior before the statx() call was introduced in coreutils-8.32.

Only `stat --cached=never` will continue to automount files.

timeout --foreground --kill-after=... will now exit with status 137

if the kill signal was sent, which is consistent with the behavior

when the --foreground option is not specified. This allows users to

distinguish if the command was more forcefully terminated.

** New Features

dd now supports the aliases iseek=N for skip=N, and oseek=N for seek=N,

like FreeBSD and other operating systems.

dircolors takes a new --print-ls-colors option to display LS_COLORS

entries, on separate lines, colored according to the entry color code.

dircolors will now also match COLORTERM in addition to TERM environment

variables. The default config will apply colors with any COLORTERM set.

** Improvements

cp, mv, and install now use openat-like syscalls when copying to a directory.

This avoids some race conditions and should be more efficient.

On macOS, cp creates a copy-on-write clone if source and destination

are regular files on the same APFS file system, the destination does

not already exist, and cp is preserving mode and timestamps (e.g.,

'cp -p', 'cp -a').

The new 'date' option --resolution outputs the timestamp resolution.

With conv=fdatasync or conv=fsync, dd status=progress now reports

any extra final progress just before synchronizing output data,

since synchronizing can take a long time.

printf now supports printing the numeric value of multi-byte characters.

sort --debug now diagnoses issues with --field-separator characters

that conflict with characters possibly used in numbers.

'tail -f file | filter' now exits on Solaris when filter exits.

root invoked coreutils, that are built and run in single binary mode,

now adjust /proc/$pid/cmdline to be more specific to the utility

being run, rather than using the general "coreutils" binary name.

** Build-related

AIX builds no longer fail because some library functions are not found.

[bug introduced in coreutils-8.32]

* Noteworthy changes in release 9.0 (2021-09-24) [stable]

** Bug fixes

chmod -v no longer misreports modes of dangling symlinks.

[bug introduced in coreutils-5.3.0]

cp -a --attributes-only now never removes destination files,

even if the destination files are hardlinked, or the source

is a non regular file.

[bug introduced in coreutils-8.6]

csplit --suppress-matched now elides the last matched line

when a specific number of pattern matches are performed.

[bug introduced with the --suppress-matched feature in coreutils-8.22]

df no longer outputs duplicate remote mounts in the presence of bind mounts.

[bug introduced in coreutils-8.26]

df no longer mishandles command-line args that it pre-mounts

[bug introduced in coreutils-8.29]

du no longer crashes on XFS file systems when the directory hierarchy is

heavily changed during the run.

[bug introduced in coreutils-8.25]

env -S no longer crashes when given unusual whitespace characters

[bug introduced in coreutils-8.30]

expr no longer mishandles unmatched \(...\) in regular expressions.

[bug introduced in coreutils-6.0]

ls no longer crashes when printing the SELinux context for unstattable files.

[bug introduced in coreutils-6.9.91]

mkdir -m no longer mishandles modes more generous than the umask.

[bug introduced in coreutils-8.22]

nl now handles single character --section-delimiter arguments,

by assuming a second ':' character has been specified, as specified by POSIX.

[This bug was present in "the beginning".]

pr again adjusts tabs in input, to maintain alignment in multi column output.

[bug introduced in coreutils-6.9]

rm no longer skips an extra file when the removal of an empty directory fails.

[bug introduced by the rewrite to use fts in coreutils-8.0]

split --number=K/N will again correctly split chunk K of N to stdout.

Previously a chunk starting after 128KiB, output the wrong part of the file.

[bug introduced in coreutils-8.26]

tail -f no longer overruns a stack buffer when given too many files

to follow and ulimit -n exceeds 1024.

[bug introduced in coreutils-7.5]

tr no longer crashes when using --complement with certain

invalid combinations of case character classes.

[bug introduced in coreutils-8.6]

basenc --base64 --decode no longer silently discards decoded characters

on (1024*5) buffer boundaries

[bug introduced in coreutils-8.31]

** Changes in behavior

cp and install now default to copy-on-write (COW) if available.

I.e., cp now uses --reflink=auto mode by default.

cp, install and mv now use the copy_file_range syscall if available.

Also, they use lseek+SEEK_HOLE rather than ioctl+FS_IOC_FIEMAP on sparse

files, as lseek is simpler and more portable.

On GNU/Linux systems, ls no longer issues an error message on a

directory merely because it was removed. This reverts a change

that was made in release 8.32.

ptx -T no longer attempts to substitute old-fashioned TeX escapes

for 8-bit non-ASCII alphabetic characters. TeX indexes should

instead use '\usepackage[latin1]{inputenc}' or equivalent.

stat will use decomposed (major,minor) device numbers in its default format.

This is less ambiguous, and more consistent with ls.

sum [-r] will output a file name, even if only a single name is passed.

This is consistent with sum -s, cksum, and other sum(1) implementations.

** New Features

cksum now supports the -a (--algorithm) option to select any

of the existing sum, md5sum, b2sum, sha*sum implementations etc.

cksum now subsumes all of these programs, and coreutils

will introduce no future standalone checksum utility.

cksum -a now supports the 'sm3' argument, to use the SM3 digest algorithm.

cksum --check now supports auto detecting the digest type to use,

when verifying tagged format checksums.

expr and factor now support bignums on all platforms.

ls --classify now supports the "always", "auto", or "never" flags,

to support only outputting classifier characters if connected to a tty.

ls now accepts the --sort=width option, to sort by file name width.

This is useful to more compactly organize the default vertical column output.

ls now accepts the --zero option, to terminate each output line with

NUL instead of newline.

nl --line-increment can now take a negative number to decrement the count.

stat supports more formats for representing decomposed device numbers.

%Hd,%Ld and %Hr,%Lr will output major,minor device numbers and device types

respectively. %d corresponds to st_dev and %r to std_rdev.

** Improvements

cat --show-ends will now show \r\n as ^M$. Previously the \r was taken

literally, thus overwriting the first character in the line with '$'.

cksum [-a crc] is now up to 4 times faster by using a slice by 8 algorithm,

and at least 8 times faster where pclmul instructions are supported.

A new --debug option will indicate if pclmul is being used.

md5sum --check now supports checksum files with CRLF line endings.

This also applies to cksum, sha*sum, and b2sum.

df now recognizes these file systems as remote:

acfs, coda, fhgfs, gpfs, ibrix, ocfs2, and vxfs.

rmdir now clarifies the error if a symlink_to_dir/ has not been traversed.

This is the case on GNU/Linux systems, where the trailing slash is ignored.

stat and tail now know about the "devmem", "exfat", "secretmem", "vboxsf",

and "zonefs" file system types. stat -f -c%T now reports the file system

type, and tail -f uses polling for "vboxsf" and inotify for the others.

timeout now supports sub-second timeouts on macOS.

wc is up to 5 times faster when counting only new line characters,

where avx2 instructions are supported.

A new --debug option will indicate if avx2 is being used.

* Noteworthy changes in release 8.32 (2020-03-05) [stable]

** Bug fixes

cp now copies /dev/fd/N correctly on platforms like Solaris where

it is a character-special file whose minor device number is N.

[bug introduced in fileutils-4.1.6]

dd conv=fdatasync no longer reports a "Bad file descriptor" error

when fdatasync is interrupted, and dd now retries interrupted calls

to close, fdatasync, fstat and fsync instead of incorrectly

reporting an "Interrupted system call" error.

[bugs introduced in coreutils-6.0]

df now correctly parses the /proc/self/mountinfo file for unusual entries

like ones with '\r' in a field value ("mount -t tmpfs tmpfs /foo$'\r'bar"),

when the source field is empty ('mount -t tmpfs "" /mnt'), and when the

filesystem type contains characters like a blank which need escaping.

[bugs introduced in coreutils-8.24 with the introduction of reading

the /proc/self/mountinfo file]

factor again outputs immediately when stdout is a tty but stdin is not.

[bug introduced in coreutils-8.24]

ln works again on old systems without O_DIRECTORY support (like Solaris 10),

and on systems where symlink ("x", ".") fails with errno == EINVAL

(like Solaris 10 and Solaris 11).

[bug introduced in coreutils-8.31]

rmdir --ignore-fail-on-non-empty now works correctly for directories

that fail to be removed due to permission issues. Previously the exit status

was reversed, failing for non empty and succeeding for empty directories.

[bug introduced in coreutils-6.11]

'shuf -r -n 0 file' no longer mistakenly reads from standard input.

[bug introduced with the --repeat feature in coreutils-8.22]

split no longer reports a "output file suffixes exhausted" error

when the specified number of files is evenly divisible by 10, 16, 26,

for --numeric, --hex, or default alphabetic suffixes respectively.

[bug introduced in coreutils-8.24]

seq no longer prints an extra line under certain circumstances (such as

'seq -f "%g " 1000000 1000000').

[bug introduced in coreutils-6.10]

** Changes in behavior

Several programs now check that numbers end properly. For example,

'du -d 1x' now reports an error instead of silently ignoring the 'x'.

Affected programs and options include du -d, expr's numeric operands

on non-GMP builds, install -g and -o, ls's TABSIZE environment

variable, mknod b and c, ptx -g and -w, shuf -n, and sort --batch-size

and --parallel.

date now parses military time zones in accordance with common usage:

"A" to "M" are equivalent to UTC+1 to UTC+12

"N" to "Y" are equivalent to UTC-1 to UTC-12

"Z" is "zulu" time (UTC).

For example, 'date -d "09:00B" is now equivalent to 9am in UTC+2 time zone.

Previously, military time zones were parsed according to the obsolete

rfc822, with their value negated (e.g., "B" was equivalent to UTC-2).

[The old behavior was introduced in sh-utils 2.0.15 ca. 1999, predating

coreutils package.]

date now pads nanoseconds on the right, not the left. For example,

if the time is currently 1590020079.003388470 seconds after the

Epoch, then "date '+%s.%-N'" formerly output "1590020079.3388470",

and it now outputs "1590020079.00338847".

ls issues an error message on a removed directory, on GNU/Linux systems.

Previously no error and no entries were output, and so indistinguishable

from an empty directory, with default ls options.

uniq no longer uses strcoll() to determine string equivalence,

and so will operate more efficiently and consistently.

** New Features

ls now supports the --time=birth option to display and sort by

file creation time, where available.

od --skip-bytes now can use lseek even if the input is not a regular

file, greatly improving performance in some cases.

stat(1) supports a new --cached= option, used on systems with statx(2)

to control cache coherency of file system attributes,

useful on network file systems.

** Improvements

stat and ls now use the statx() system call where available, which can

operate more efficiently by only retrieving requested attributes.

stat and tail now know about the "binderfs", "dma-buf-fs", "erofs",

"ppc-cmm-fs", and "z3fold" file systems.

stat -f -c%T now reports the file system type, and tail -f uses inotify.

** Build-related

gzip-compressed tarballs are distributed once again

* Noteworthy changes in release 8.31 (2019-03-10) [stable]

** Bug fixes

'base64 a b' now correctly diagnoses 'b' as the extra operand, not 'a'.

[bug introduced in coreutils-5.3.0]

When B already exists, 'cp -il A B' no longer immediately fails

after asking the user whether to proceed.

[This bug was present in "the beginning".]

df no longer corrupts displayed multibyte characters on macOS.

[bug introduced with coreutils-8.18]

seq no longer outputs inconsistent decimal point characters

for the last number, when locales are misconfigured.

[bug introduced in coreutils-7.0]

shred, sort, and split no longer falsely report ftruncate errors

when outputting to less-common file types. For example, the shell

command 'sort /dev/null -o /dev/stdout | cat' no longer fails with

an "error truncating" diagnostic.

[bug was introduced with coreutils-8.18 for sort and split, and

(for shared memory objects only) with fileutils-4.1 for shred]

sync no longer fails for write-only file arguments.

[bug introduced with argument support to sync in coreutils-8.24]

'tail -f file | filter' no longer exits immediately on AIX.

[bug introduced in coreutils-8.28]

'tail -f file | filter' no longer goes into an infinite loop

if filter exits and SIGPIPE is ignored.

[bug introduced in coreutils-8.28]

** Changes in behavior

cksum, dd, hostid, hostname, link, logname, sleep, tsort, unlink,

uptime, users, whoami, yes: now always process --help and --version options,

regardless of any other arguments present before any optional '--'

end-of-options marker.

nohup now processes --help and --version as first options even if other

parameters follow.

'yes a -- b' now outputs 'a b' instead of including the end-of-options

marker as before: 'a -- b'.

echo now always processes backslash escapes when the POSIXLY_CORRECT

environment variable is set.

When possible 'ln A B' now merely links A to B and reports an error

if this fails, instead of statting A and B before linking. This

uses fewer system calls and avoids some races. The old statting

approach is still used in situations where hard links to directories

are allowed (e.g., NetBSD when superuser).

ls --group-directories-first will also group symlinks to directories.

'test -a FILE' is not supported anymore. Long ago, there were concerns about

the high probability of humans confusing the -a primary with the -a binary

operator, so POSIX changed this to 'test -e FILE'. Scripts using it were

already broken and non-portable; the -a unary operator was never documented.

wc now treats non breaking space characters as word delimiters

unless the POSIXLY_CORRECT environment variable is set.

** New features

id now supports specifying multiple users.

'date' now supports the '+' conversion specification flag,

introduced in POSIX.1-2017.

printf, seq, sleep, tail, and timeout now accept floating point

numbers in either the current or the C locale. For example, if the

current locale's decimal point is ',', 'sleep 0,1' and 'sleep 0.1'

now mean the same thing. Previously, these commands accepted only

C-locale syntax with '.' as the decimal point. The new behavior is

more compatible with other implementations in non-C locales.

test now supports the '-N FILE' unary operator (like e.g. bash) to check

whether FILE exists and has been modified since it was last read.

env now supports '--default-signal[=SIG]', '--ignore-signal[=SIG]', and

'--block-signal[=SIG], to setup signal handling before executing a program.

env now supports '--list-signal-handling' to indicate non-default

signal handling before executing a program.

** New commands

basenc is added to complement existing base64,base32 commands,

and encodes and decodes printable text using various common encodings:

base64,base64url,base32,base32hex,base16,base2,z85.

** Improvements