* Major changes in release 4.4.0

The 4.4.0 release of findutils is a stable release, succeeding the
final release in the previous development series, 4.3.13.  However,
since many users will have previously been using the previous stable
release series, this section describes the changes between the 4.2.33
release (which was the final 4.2.x release) and 4.3.0.

Some items in the lists of changes are prefixed by bug numbers (though
some of them are simply enhancements, not bugs).

Apart from the changes in version number and development versus stable
status, the only differences between 4.3.13 and 4.4.0 are bug fixes
#15472 and #20873.

It's possible that some of the bug fixes mentioned as fixed are in
fact fixes for bugs both introduced and fixed in 4.3.x (and thus not
present in 4.2.x at all).  While I have tried not to list those, some
may have slipped through.


** Functional enhancements to locate

*** slocate compatibility

The slocate database format is supported, both for reading by locate
and writing by updatedb.

Preliminary changes intended to eventually allow setuid operation of
locate have also been made.  For the moment, please don't install GNU
locate as a set-user-ID program (except for testing purposes; if you
do so, please make sure that untrusted users cannot execute the
set-user-ID locate program).

Use of an slocate database which was built with a nonzero security
mode (at the moment, GNU updatedb will not do this) forces locate's
"-e" option to be turned on, and that has an effect on the "-S" option
which is probably surprising for most users.

*** Other changes

Locate can now read old-format locate databases generated on machines
with a different byte order.  It does this by guessing the byte order,
so the result is not completely reliable.  If you need to share
databases between machines of different architectures, you should use
the LOCATE02 format (which has other advantages, as explained in the
documentation).

A new option, --max-database-age, has been added to locate.   

Translation of locate --limit problems is improved.

The /proc filesystem is excluded from the locate database (by
default; change PRUNEPATHS to modify this behaviour).

** Functional enhancements to find

*** fts

By default, find now uses the fts() function to search the file
system.  The use of fts greatly increases find's ability to search
extremely deep directory hierarchies.

You can tell that the version of find you are using uses FTS, because
the output of "find --version" will include the word "FTS".

Currently two binaries for 'find' are built.  The configure option
--without-fts can be used to select whether 'find' uses fts:

				  With fts	Without fts
default configuration		  find		oldfind
configure --with-fts		  find		oldfind
configure --without-fts		  ftsfind	find

New tests, -readable, -writable, -executable.  These check that a file
can be read, written or executed respectively.

*** Changes to printf

The -printf action (and similar related actions) now support %S,
which is a measurement of the sparseness of a file.

*** Changes to -perm

The test "-perm /000" now matches all files instead of no files.  For
over a year find has been issuing warning messages indicating that
this change will happen.  We now issue a warning indicating that the
change has already happened (in 4.3.x only, there is no plan to make
this change in the 4.2.x series).

*** Time stamp resolution

The tests -newer, -anewer, -cnewer, -mtime, -atime, -ctime, -amin,
-cmin, -mmin and -used now support sub-second time stamps, including
the ability to specify times with non-integer arguments.

The -printf format specifiers also support sub-second time stamps:
  atime   ctime  mtime
  %a      %c     %t
  %AS     %CS    %TS 
  %AT     %CT    %TT 
  %A+     %C+    %T+ 
  %AX     %CX    %TX 

*** Changes to -prune

The -prune action now always evaluates as true (this is also a
bug fix).

*** New tests

The new test -newerXY supports comparison between status times for
files.  One of the status times for a file being considered (denoted
X) is checked against a reference time (denoted Y) for the file whose
name id the argument.  X and Y can be:

   a 	Access time
   B	Birth time (st_birthtime, currently unsupported)
   c	Change time
   m	Modification time
   t	Valid only for the reference time; instead of comparison
	against a file status time, the argument is a time string.
	Not yet supported.

For example, -newermm is equivalent to -newer, and -neweram is true if
the file being considered was accessed more recently than the
reference file was modified.  The -newerXY test supports subsecond
timestamps where these are available.  The X=B variant is not yet
implemented.

#11668: FreeBSD extensions for time specification are now implemented.

*** Other changes to find

#20688: The warning printed by -name or -iname when the pattern to
match contains a slash can now be silenced by -nowarn.  This warning
was originally introduced unconditionally in 4.2.21.

For find, debug output can now be enabled at runtime with the -D
option.  This causes the printing of various sorts of information
about find's internal state and progress.

The find option -nowarn cannot itself produce a warning (this used to
happen with commands like "find . -name quux -nowarn -print").

You now get a more helpful error message when you use command lines
which have missing expressions, such as 
      find . ( )
      find . !
      find . -a
      find . \( -not \)
      find . \( -true -a


*** Standards conformance

POSIX will standardise -path, so the documentation no longer claims
that -wholename is the 'canonical' test, and -ipath no longer
generates a warning.

When the POSIXLY_CORRECT environment variable is set, "find -perm
+a+w" is rejected as invalid.  Some other similar mode strings
starting with '+' which are not valid in POSIX are also rejected.

Find now follows POSIX rules for determining where directories end and
expressions start.  This means that "find \(1 \!2 \, \)" now searches
in the four named directories, rather than trying to parse an
expression. (Savannah bug #15235).

#21039: Setting the POSIXLY_CORRECT environment variable now turns off
warnings by default, because POSIX requires that only diagnostic
messages (and -ok prompts) are printed on STDERR, and diagnostic
messages must also result in a nonzero exit status.

#20803: POSIX requires that -prune always returns true.  Previously it
returned false when -depth was in effect and true otherwise.

** Functional ehnahcements to xargs

While there are a number of bug fixes in xargs in this release (as
compared to the previous stable release), there are no functional
enhancements as such.

** Performance Enhancements

*** Cost-based optimiser

Find now has a rudimentary cost-based optimiser.  It has an idea of
the basic cost of each test (i.e. that -name is very cheap while -size
is more expensive).  It re-orders tests bearing in mind the cost of
each test and its likely success.  Predicates with side effects (for
example -delete or -exec) are not reordered.  The optimiser is not
yet enabled by default, but the new option -O controls the query 
optimisation level.  To see this in action, try
  find -D opt -O3 . -type f -o -type c -o -size 555 -name Z
and compare the optimised query with:
  find -D opt -O3 . -size 555 -o -type c -o -type f -name Z
and
  find -D opt     . -size 555 -o -type c -o -type f -name Z

Over time, as optimisations are proven to be robust and correct, they
will be moved to lower optimisation levels.  Some optimisations have
always been performed by find (for example -name is always done early
if possible).

** Security Fixes

#20014: Findutils-4.3.7 includes a patch for a potential security
problem in locate.  When locate read an old-format database, it read
file names into a fixed-length buffer allocated on the heap without
checking for overflow.  Although overflowing a heap buffer is often
somewhat safer than overflowing a buffer on the stack, this bug still
has potential security implications.

This bug also affected the following previous findutils releases:

 - All releases prior to 4.2.31
 - Findutils 4.3.0 to 4.3.6.

This bug has been assigned CVE number CVE-2007-2452.

** Bug Fixes

#22057: Actually rename the old locate database to the new one
 atomically, instead of just claiming the rename is atomic in a
 comment.

#22056: -Xtime tests are off by one second (e.g. rm -f x; touch x;
 find x -mtime 0 should print x).

#21960: xargs should collect the exit status of child processes even
 if the total count of unreaped children has not yet reached the
 maximum allowed.

#21568: Switch to checking the gnulib code out with native git, not
 CVS.  This affects mainly those who check findutils code out of CVS.

#20970: Trailing slash on directory arguments breaks -name.  "find
 foo/ -name foo" now correctly matches foo and printf foo/.  See POSIX
 interp
 http://www.opengroup.org/austin/interps/uploads/40/14959/AI-186.txt

#20865: Using both -delete and -prune without explicitly using -depth
 is now an error.  Traditionally, -delete has always turned -depth on
 anyway, so this is not a functional change.  However, using -depth
 (implicitly or explicitly) makes -prune a no-op.  This change is
 intended to avoid nasty surprises for people who test with "-print"
 and then change it to "-delete" when they are happy.

#20834: Avoid segmentation violation for -execdir when $PATH is unset.
 Assume that the PATH is safe in this situation.

#20802: If -delete fails, find's exit status will now be non-zero.
 However, find still skips trying to delete ".".

#20547: The version information printed by find, xargs, locate,
 updatedb, frcode and code now complies with the GNU Project's coding
 standards.

#20310: configure uses hosts's support status for "sort -z" when
 generating the updatedb script for use on the target.  This is
 inappropriate when cross-compiling, so avoid doing that.

#20273: When xargs is successful without consuming all of stdin (for
 example, with the -E option), and stdin is seekable, xargs now
 correctly restores the file position, even on platforms where exit()
 does not follow the POSIX rules of doing likewise.  Likewise for find
 (for example, with the -ok action).

#20157: Avoid segfault in locate when run as root.  This is caused by
 a buffer overrun, but at this time no exploit mechanism is known.

#20139: find -[acm]time -N (wrongly) includes files from N days ago,
 as well as (correctly) from less than N days ago.

#20005: Tests -mtime -n and -mtime +n incorrectly treated like -mtime
 n.

#19948: Fixed an assertion failure on IRIX 6.5 (O_NOFOLLOW is defined
 to 0 there).

#19923: Fixed an array overrun in groups[] array of 'locate' when run
 by or as root.  This bug appears not to be exploitable.  If locate is
 not installed setuid, the bug is not exploitable.  For setuid
 installations, it is conceivable that there could be an information
 leak if the user uses the -d option or the -e option, though the
 maintainer has been unable to provoke this on an x86 system.

#19871: Typos in find.1

#19871: Spurious .R directives in man page produced error messages from
 GNU troff.  This is now fixed (they are corrected to .B).

#19806: The -samefile predicate might get fooled by inode reuse.  We
 now hold open a file descriptor on the reference file to prevent
 this.

#19768: Better detection of corrupted old-style locate databases
 (e.g. if the database is too short to include a complete bigram
 table).

#19766: The frcode and code programs now detect write errors more
 reliably.

#19658: When cross-compiling, "make clean" no longer deletes the
 generated file doc/regexprops.texi, because there is no way to
 regenerate it.

#19634: Test suite now passes (again) if "." is on your $PATH.

#19619: Findutils builds once again on Cygwin.

#19605: Issue an error message (and later return nonzero exit status)
 if a symbolic link loop was encountered during directory traversal.

#19596: Correct the comparison in the find man page and Texinfo manual
 between %b and %s (the divisor is 512 not 1024).

#19484: bigram.c and code.c fail if the first pathname recorded begins
 with a space

#19483: Inconsistent option highlighting in updatedb man page

#19416: The result of I/O operations in print-related actions is now
 checked, and failures are reported.  Any failure will cause find's
 exit status to be nonzero.  The predicate itself will continue to
 return true.

#19391: When xargs knows that the system's actual exec limit is larger
 than the compiled-in ARG_MAX, use the system's limit without
 generating an assertion failure.

#19371: Fix compilation failure on systems which #define open to
 open64 (and similarly with the close system call).  This fixes
 Savannah bug #19371, affecting AIX 5.3.

#18714: In the POSIX locale, vertical tabs and form feeds are not
 field separators.

#18713: Quoted but empty arguments which occur last on an xargs input
 line are no longer ignored, but instead produce an empty argument.

#18466: we now avoid this bug by limiting "-execdir ...+" to just one
 argument for the time being.  There is a performance penalty for
 doing this.  We hope to make a better fix in a later release.

#18414: Tests for "find -readable" are skipped for the superuser, as
 on some systems (e.g. Cygwin with an Administrative user) users can
 read mode-000 files.

#18384: excess bracket in xargs --help

#18320: Zero bytes in input should give warning

#18222: find -printf '%H %P' once again prints the right result if
 more than one start point was given on the command line.

#18203: A duplicate report of bug #17478.

#17782: find -execdir now correctly puts the prefix "./" before the
 expansion of "{}" rather than at the start of the argument it appears
 in.  Please note that if you use the -exec or -execdir actions with a
 shell, then you may be vulnerable to shell code injection attacks, so
 don't do that.  It's not a security defect in find - you should not
 be passing untrusted data (such as file names chosen by other people)
 to the shell.

#17478: Error messages from find can garble the console.

#17477: find -printf '%' (that is, where the format has a trailing %)
 now generates an error message.

#17437: Corrected the handling of X in symbolic permissions (such as
 -u+w,a+X).  

#17396: find -mtime -atime -ctime does not support fractional part
 (see "Functional changes" below)

#17372: The fts-based find executable (the default configuration uses
 fts) is now much faster when -maxdepth is used on filesystems with
 high fanouts.

#16738: "find .... -exec ... {} +" now works if you have a large
 environment and many files must be passed to the -exec action.  The
 same problem affected the -execdir action, though since the number of
 files in a given directory will normally be smaller, the problem was
 worse for -exec.

#16579: Updatedb now works if it is running as a user whose login
 shell is not actually a shell.

#16378: Assertion failure if stat() returns 00000 as the mode of a
 file.  This apparently can happen occasionally with broken NFS
 servers.

#15800: If find finds more subdirectories within a parent directory
 than it previously expected to based on the link count of the parent,
 the resulting error message now gives the correct directory name
 (previously an error message was issued but it specified the wrong
 directory).

#15531: The -prune action now behaves correctly when applied to a
 file.

#15472: Error messages that print ino_t values are no longer truncated
 on platforms with 64-bit ino_t.

#15384: Find misbehaves when parent directory is not readable.

#14748: find -perm /zzz gives wrong result when zzz evaluates to an
 all-zero mask

#14535: correctly support case-folding in locate (that is, "locate
 -i") for multi-byte character environments such as UTF-8.  Previously,
 if your search string contained a character which was outside the
 single-byte-encoding range for UTF-8 for example, then the
 case-folding behaviour failed to work and only exact matches would be
 returned.

** Documentation Fixes

#20873: Indicate that * matches / and leading dot in filenames for
"find -path".

#18554: Documented the construct -exec sh -c 'foo "$@" bar' {} +

#15360: The global effect of options (other than -daystart and
 -follow) is now explained more clearly in the manual page.

The locatedb.5 man page now documents the (default) LOCATE02 format
more clearly, and also documents the slocate database format.

The maximum and default values applying to the -s option of xargs are
now documented more clearly in the manual page.

** Compilation Fixes

If you configure the source code and then run the tests with "make
check", the test suite fails rather than defaulting to testing the
system binaries.

#19416: _FORTIFY_SOURCE warn_unused_result warnings

#19948: Assertion failure O_NOFOLLOW != 0 on IRIX 6.5

#19965: Compilation failure on OSF/1 4.0; non-declaration of uintmax_t

#19965: Fixed a compilation failure on OSF/1 4.0 (no definition of the
 type uintmax_t).

#19966: Findutils should now build on systems which have the modf()
 and fabs() functions in the maths library, -lm.  This includes some
 versions of HP-UX and Solaris.

#19966: find should link against -lm for modf() and fabs()

#19967: Build successfully with C compilers that don't support the GCC
 construct __attribute__((__noreturn__)).

#19967: Use of __attribute((__noreturn__)) makes compilation fail with
 some non-GCC compilers

#19970: Cannot cast from pointer to bool using gnulib's <stdbool.h>

#19970: Compile correctly on C89 systems where the "_Bool" type is not
 provided, taking into account the limitations of the gnulib
 replacement for stdbool.h.

#19979: Compilation errors on BeOS

#19980: Don't use the functions putw() or getw() since these are not
 in current POSIX.  Use the gnulib version of wcwidth() where the
 system does not provide it.

#19981: Don't call setgroups if the function isn't available.

#19983: Now compiles on DEC C V5.9-005 on Digital UNIX V4.0 (or at
 least, should).

#20128: Fix compilation error of find/tree.c on AIX with GCC.

#20263: Compilation fix for DEC Alpha OSF/1 cc, which forbids the
 ordering comparison of function pointers.

#20594: Allow fine-tuning of the default argument size used by xargs
 and find at ./configure time.