CHANGE.LOG

I consider libtrash pretty much "stable", in the sense that I don't have any
immediate plans of adding new features and that I don't know of any bugs. Of
course, if any problem, bug or misbehaviour is either reported or detected
by me (during my daily usage of libtrash), I will promptly investigate it,
correct the relevant section of the code and then release a new, corrected
version.

Yet, I do hope that you will tell me about any important feature(s) you
think libtrash is still lacking! :-)

- Manuel Arriaga (2002/08/20)


--------------------------------------------------------------------------

Changes 0.1 -> 0.2: (2001-08-11)

- Fixed bug introduced into the code just before releasing libtrash-0.1. It
should now run as I originally meant it to. The fix consisted in adding the
line

*d_ptr = '\0';

at line 1639 of file main.c, inside function "cleanse_path()".

- Fixed macro bug: if you follow the instructions at the top of the source
file to (dis|e)nable debug messages, it now respects your option.

--------------------

Changes 0.2 -> 0.3 (2001-08-18)

- Fixed bug in found_under_dir() which would cause false positives in cases
such as when /dir-a is in the directory list and found_under_dir() is asked
whether /dir-abc is under that directory list. Did so by adding second test
to the if statement in that function, checking if dir_name is followed by a
slash in abs_path.

- Reorganized code, dividing it into three separate files, and performed
necessary changes to the scope of some variables.

- Now libtrash is disabled by setting an environmental variable, TRASH_OFF,
to YES. This allows us to offer the user the option to have libtrash, when
disabled, recall her, after a program exits, that it isn't running and that
it might be a good idea to re-enable it. 

- The scripts libtrash_off, libtrash_on and hardrm are no longer provided;
tips on how to implement their functionality using shell aliases are offered
in the README file.

- The code which sets the configuration variables from values in the
environment has been encapsulated into a function called get_conf_env().
This way, besides achieving a better structured _init(), we open the way for
allowing file-based configuration in the future (e.g., in the form of an
alternative get_conf_file() function).

- Now the Makefile uses 'ginstall' rather than 'cp' to put the lib in its
place.

-------------------

Changes 0.3 -> 0.4 (2001-08-20)

- Warnings about libtrash being disabled are now printed to stderr rather
than stdout, so that shell scripts which process the output of other
programs don't get mixed up with the warning. (Most notoriously, if you
disabled libtrash - by setting TRASH_OFF=YES in the environment - before
compiling the Linux kernel it would fail because the Makefile would set the
variable ARCH incorrectly due to an error processing the output of 

shell uname -m | sed -e s/i.86/i386/ -e s/sun4u/sparc64/ -e s/arm.*/arm/ -e
s/sa110/arm/

which contained (more than) one warning. When the warning is printed to
stderr instead, this problem no longer occurrs.

- The configuration variable TRASH_RECURSE has been renamed to (the
not-much-better name) TRASH_PROTECT_TRASH, in the hope of making its meaning
(vaguely) clearer. As pointed out in the file README-1ST, this means that
you must now set a different variable in your environment if you wish to
override this compile-time default.

- Rewrote the end of unlink().

- build_absolute_path now invokes cleanse_path.

- lstat() is now used instead of stat(), so that symlinks are recognized as
such, and code has been added to always ignore them.

- dir_ok() has been modified so that it only requires/sets the permissions
we really need for directories: write- and execute-permission.

- files under UNREM_DIRS are now also protected from attempts to destroy
them coming from users other than root;

- added support for ignoring files with names ending in certain extensions
(default list: ".o;.log;.aux").

- major news: now calls to unlink(), rename(), fopen(), freopen() and open()
can all be intercepted to make accidental data loss even more difficult.

-----------------------

Changes 0.4 -> 0.5 (2001-09-05)

- Corrected bug (?) in reformulate_new_path(): it now uses access() to check
the user's permissions over an existing directory.

- Corrected bug in dir_ok(): the trash can (and all folders beneath it) are
now created with read-permission.

-----------------------

Changes 0.5 -> 0.6 (2001/09/22)

- (new!) support for file-based configuration of libtrash has finally been
added.

- corrected the way we handle dlsym() errors.

- major code rewrite (fopen(), freopen() and open() were merged into a
single "intelligent" - or "magic" :) - function).

- added new features, such as LIBTRASH_CONF_FILE_UNREMOVABLE and (crude)
support for inter-device file saving when GLOBAL_PROTECTION is set.

- no more problems with SETUID programs being unable to successfully perform
operations on files due to having insufficient permissions (the problem was
caused by the fact that libtrash performed permission checks with the
process' real UID rather than its effective UID - now the effective UID is
used). 

- libtrash is now compiled with REENTRANT defined, which is supposed to make
libtrash thread-safe. (Besides the use of errno, there was nothing
"thread-unsafe" in libtrash; now, with REENTRANT defined, no problems should
arise.)

- when a wrapper fails, or the destruction of a "protected" file is denied,
libtrash now tries to set a meaningful value in errno so that the calling
program can "interpret" the cause of the error; if that proves itself
impossible, it sets errno to zero in order to avoid having misleading error
messages displayed to the user. (This happened often in previous versions of
libtrash when an attempt to destroy a "protected" file failed: since the
wrapper for unlink() didn't set errno to a meaningful value, rm usually
printed the wrong error message. In such situations, libtrash now sets errno
to EACCES, thereby generating a "insufficient permissions" error message,
which - although not a perfect match - does hint at the reason of the
failure.)

- added Daniel Sadilek's cleanTrash script to the tarball, which helps in
keeping the disk usage by libtrash at an acceptable level.

- added IGNORE_EDITOR_TEMPORARY option.

----------------------

Changes 0.6 -> 0.7 (2002/01/05)

- Restructured libtrash, so that the initialization routine is called every
time a wrapper function is invoked. This solves the only problems which have
been reported: incompatibility with Samba and the apps in the procps package
('ps', 'top', etc...). I very much preferred the previous design, but this
really seems to be the only way to make libtrash completely transparent to
other applications. I hope this won't have a noticeable impact on
performance, although it certainly made the code a good lot uglier than I
would like it to be.

-----------------------

Changes 0.7 -> 0.8 (2002/02/03)

- Changed the code in TEST_FILE_EXISTENCE() (open-funs.c) so that special
files (such as character devices, FIFOs, etc) are ignored.


----------------------

Changes 0.8 -> 0.9 (2002/03/21)

- Now the 64-bit versions of fopen(), freopen() and open() are also wrapped
-- fopen64(), freopen64() and open64(). This closes the gap which existed in
the protection against accidental file truncation, e.g., with 'cp'. I didn't
know that those functions existed, and always wondered why libtrash didn't
work with 'cp'... Everything should work fine now.

-----------------------

Changes 0.9 -> 1.0 (2002/05/26)

- Added handy configuration variable UNCOVER_DIRS.

- Cleaned documentation.

-----------------------

Changes 1.0 -> 1.1 (2002/06/02)

- UNCOVER_DIRS can now be used to easily "uncover" your own trash can, in
case you have PROTECT_TRASH set. 

- The README file now describes two ways to activate libtrash.

------------------------

Changes 1.1 -> 1.2 (2002/06/08)

- Changed code in found_under_dir() so that directories named with a
"trailing slash" were processed correctly. This became important after I
added the UNCOVER_DIRS feature, because shell auto-completion puts a slash
after the name of directories (eg, "UNCOVER_DIRS=/u(Tab)" results in
"UNCOVER_DIRS=/usr/"). UNCOVER_DIRS should be even more more handy now.

- Added the variable EXCEPTIONS and the function is_an_exception(). The
motivation behind this is to allow mount/umount to handle their lock file
correctly while libtrash is running and UNREMOVABLE_DIRS contains /etc.

---------------------------

Changes 1.2 -> 1.3 (2002/08/20)

- Added an alternative cleanTrash script contributed by Martin Corley.

---------------------------

Changes 1.3 -> 1.4 (2002/10/13)

- Fixed bug in DO_OPEN_SPECIFIC_PREPARATION() macro inside open-funs.c. 
This bug, reported by Frederic Connes, was preventing files from retaining
there current permission set when they were opened with open()/open64() in
(O_CREAT | O_TRUNC) mode.

- Changed name of configuration variables UNLINK_OFF, RENAME_OFF, FOPEN_OFF,
FREOPEN_OFF and OPEN_OFF to the more intelligible INTERCEPT_UNLINK,
INTERCEPT_RENAME, INTERCEPT_FOPEN, INTERCEPT_FREOPEN and INTERCEPT_OPEN,
following Frederic Connes' suggestion.
  

----------------------------

Changes 1.4 -> 1.5 (2002/11/06)

- Fixed bug in readline() which was making libtrash stop reading the
configuration file when it found a blank line. This bug was reported by Dan
Stutzbach, who also provided me with a patch.

- Fixed memory leak (cfg->uncovered_dirs wasn't being free()d) reported by
Ryan Brown.


-----------------------------

Changes 1.5 -> 1.6 (2002/12/06)

- Added USER_TEMPORARY_DIRS feature.

-----------------------------

Changes 1.6 -> 1.7 (2003/01/07)

- Applied two patches contributed by Frederic Connes which (i) create a
symlink from libtrash.so to libtrash.so.{VERSION} when 'make install' is run
and (ii) fix a bug in the cleanTrash script.

-----------------------------

Changes 1.7 -> 1.8 (2003/02/04)

- Following a suggestion from Paul Adams, libtrash now ignores files with a
null byte count. This should prevent trash cans from filling up with
worthless lock files.

-----------------------------

Changes 1.8 -> 1.9 (2003/02/26)

- Fixed bug in the hidden_file() function, which was causing libtrash to
enter an infinite loop when asked to remove a file with a name starting with
multiple dots.

-----------------------------

Changes 1.9 -> 2.0 (2003/03/05)

- Added the IGNORE_RE ("ignore regular expression") feature to libtrash. The
code which implements it was contributed by Jacek Sliwerski (rzyjontko).

-----------------------------

Changes 2.0 -> 2.1 (2003/06/25)

- Fixed a bug in the build_absolute_path() function, which made that
function return an error code when passed a path to a file directly under
the root of the file system. 

-----------------------------

Changes 2.1 -> 2.2 (2003/12/07)

- Fixed a bug in the convert_relative_into_absolute_paths() function, which
was calling strcat() instead of strcpy().
 
- Included the strash script, kindly provided by Frederic Connes, in the
cleanTrash directory.


-----------------------------

Changes 2.2 -> 2.3 (2004/11/07)

- Fixed a bug which led libtrash to crash when functions were called with a
NULL pointer as the pathname.


-----------------------------

Changes 2.3 -> 2.4 (2005/Sep/24)

- Implemented a work-around necessary due to a undocumented "feature" of GNU
libc, by which, if there are multiple symbols with the same name in a
library, dlsym() returns a pointer to the *oldest* version of that function
call. This was causing libtrash to invoke the old version of fopen() found
in glibc; if the calling code then attempted to use a wide-char supporting
function on the returned file pointer, GNU libc would crash. We now make
sure that any file pointers we return are obtained from GNU libc's 2.1
version of fopen(). (As of this time, 2.1 seems to be the highest version
number listed in the table of symbols in glibc.)

- Took this chance to "sprinkle" five additional calls to dlerror() *before*
the calls to dlsym()/dlvsym(), as should be done to correctly detect
dlsym()/dlvsym() errors (which is achieved by then calling dlerror() *after*
calling those functions).

-----------------------------

Changes 2.4 -> 2.5 (2006/Oct/26)

- Modified build_absolute_path() so that it becomes a useful wrapper around
canonicalize_file_name(). I recently learned that this function does not
handle paths to inexistent files at all; having a wrapper which can cope
with that (although not useful in the context of libtrash) is something
nice.

-----------------------------

Changes 2.5 -> 2.6 (2007/Feb/18)

- Added support for unlinkat(), renameat(), openat() and openat64() function
calls (featured in recent versions of GNU libc). Currently these functions
are supported without ever actually invoking their "real" versions found in
GNU libc. This defeats their very purpose (which, according to the man page,
is to avoid race conditions by having parallel changes to the dir structure
"above" the path these functions operate on sabotaging their functioning).
This will be changed in a future version.

- Added support for creat() and creat64().

-----------------------------

Changes 2.6 -> 2.7 (2007/May/4)

- Jacek Sliwerski rewrote parts of the reformulate_new_path() function so
that (i) it uses a faster logarithmic search and (ii) does not use the
log10() function.


-----------------------------

Changes 2.7 -> 2.8 (2007/May/6)

- Removed a call to log10() which became a unresolved symbol since we
stopped linking against -lm.

-----------------------------

Changes 2.8 -> 2.9 (2007/Dec/2)

- FOR PREVIOUS USERS OF LIBTRASH: since some recent distributions have
started using /media as the top dir for removable media mount points, I 
have updated the default value of REMOVABLE_MEDIA_MOUNT_POINTS to include
it; if that doesn't suit you, edit your libtrash.conf file before
compiling. 

- Fixed premature free()ing of conf_file_path in read_config_from_file (only
a problem when compiled in DEBUG mode).

- Removed the fix introduced earlier which had us use dlvsym() when trying to get a
pointer to the real fopen(), since its hard-coded request for version
"GLIBC_2.1" of fopen no longer works on recent systems. (In *TWO* locations:
init() and the macro GET_REAL_FUNCTION() inside open-funs.c.)

- Added a simple test, fopen-wc-test, to the Makefile which asks users to
notify me if it looks like the removal of that fix will break things for
them (hopefully won't happen).

- Moved the patch and warning concerning glibc 2.2.5 into a separate dir to
avoid unnecessarily scaring/confusing users.

- Added a check for the GNU libc fopen() not being available before using it
inside read_config_from_file() and move().

-----------------------------

Changes 2.9 -> 3.0 (2007/Dec/3)

- Fixed mistaken prototypes for the real_fopen, real_freopen and their
64-bit counterparts function pointers (their return type was specified as
int rather than FILE*)

- Fixed wrong error return value of move() in a new test I added in the last
version (was returning NULL rather than -1).

- Changed the way we obtain the pointers to the "real" functions. Since
removing the old dlvsym() fix (in v2.9) broke things for some users, we now
compile a small test binary (linking-helper.c), run ldd on it and then
inspect the output of running 'readelf -s' on the different libs listed by
ldd. The result is that we now can specify a symbol version which we know is
available in the user's system and therefore use dlvsym() in place of dlsym()
(which, as was the original problem, for some functions -- notably, fopen() --
returns an older version of the function rather than the version returned by
the run-time linker).

- Support for the "at functions" (unlinkat/renameat/openat/openat64) is now
decided automatically by the get_symbol_versions.py script before compilation,
rather than having the user specify AT_FUNCTIONS=YES/NO in libtrash.conf.

- Added a compile time check for a functioning proc file system.

-----------------------------

Changes 3.0 -> 3.1 (2007/Dec/25)

- Updated the get_symbol_versions.py so that the symbol version used
for each wrapped GNU libc function is that identified with "@@" in the
output of 'readelf -s [path to lib]'. (Following advice received on the
binutils mailing list.)


-----------------------------

Changes 3.1 -> 3.2 (2008/Jan/23)

- Added a check for a sufficiently recent version of Python to the
get_symbol_versions.py configuration script. 

-----------------------------

Changes 3.2 -> 3.3 (2014/Jun/08)

- Renamed _init/_fini functions to libtrash_init/libtrash_fini to
avoid symbol clashes when loading audacious plugins. (Thanks to Kamil
Dudka for sending me a patch.)