blob: ecaf4170ef889c5cf59b41c59f9264d691b9061d [file] [log] [blame]
0.9.0
~~~~~
First version.
0.9.0a
~~~~~~
Removed 'ranlib' from Makefile, since most modern Unix-es
don't need it, or even know about it.
0.9.0b
~~~~~~
Fixed a problem with error reporting in bzip2.c. This does not effect
the library in any way. Problem is: versions 0.9.0 and 0.9.0a (of the
program proper) compress and decompress correctly, but give misleading
error messages (internal panics) when an I/O error occurs, instead of
reporting the problem correctly. This shouldn't give any data loss
(as far as I can see), but is confusing.
Made the inline declarations disappear for non-GCC compilers.
0.9.0c
~~~~~~
Fixed some problems in the library pertaining to some boundary cases.
This makes the library behave more correctly in those situations. The
fixes apply only to features (calls and parameters) not used by
bzip2.c, so the non-fixedness of them in previous versions has no
effect on reliability of bzip2.c.
In bzlib.c:
* made zero-length BZ_FLUSH work correctly in bzCompress().
* fixed bzWrite/bzRead to ignore zero-length requests.
* fixed bzread to correctly handle read requests after EOF.
* wrong parameter order in call to bzDecompressInit in
bzBuffToBuffDecompress. Fixed.
In compress.c:
* changed setting of nGroups in sendMTFValues() so as to
do a bit better on small files. This _does_ effect
bzip2.c.
0.9.5a
~~~~~~
Major change: add a fallback sorting algorithm (blocksort.c)
to give reasonable behaviour even for very repetitive inputs.
Nuked --repetitive-best and --repetitive-fast since they are
no longer useful.
Minor changes: mostly a whole bunch of small changes/
bugfixes in the driver (bzip2.c). Changes pertaining to the
user interface are:
allow decompression of symlink'd files to stdout
decompress/test files even without .bz2 extension
give more accurate error messages for I/O errors
when compressing/decompressing to stdout, don't catch control-C
read flags from BZIP2 and BZIP environment variables
decline to break hard links to a file unless forced with -f
allow -c flag even with no filenames
preserve file ownerships as far as possible
make -s -1 give the expected block size (100k)
add a flag -q --quiet to suppress nonessential warnings
stop decoding flags after --, so files beginning in - can be handled
resolved inconsistent naming: bzcat or bz2cat ?
bzip2 --help now returns 0
Programming-level changes are:
fixed syntax error in GET_LL4 for Borland C++ 5.02
let bzBuffToBuffDecompress return BZ_DATA_ERROR{_MAGIC}
fix overshoot of mode-string end in bzopen_or_bzdopen
wrapped bzlib.h in #ifdef __cplusplus ... extern "C" { ... }
close file handles under all error conditions
added minor mods so it compiles with DJGPP out of the box
fixed Makefile so it doesn't give problems with BSD make
fix uninitialised memory reads in dlltest.c
0.9.5b
~~~~~~
Open stdin/stdout in binary mode for DJGPP.
0.9.5c
~~~~~~
Changed BZ_N_OVERSHOOT to be ... + 2 instead of ... + 1. The + 1
version could cause the sorted order to be wrong in some extremely
obscure cases. Also changed setting of quadrant in blocksort.c.
0.9.5d
~~~~~~
The only functional change is to make bzlibVersion() in the library
return the correct string. This has no effect whatsoever on the
functioning of the bzip2 program or library. Added a couple of casts
so the library compiles without warnings at level 3 in MS Visual
Studio 6.0. Included a Y2K statement in the file Y2K_INFO. All other
changes are minor documentation changes.
1.0
~~~
Several minor bugfixes and enhancements:
* Large file support. The library uses 64-bit counters to
count the volume of data passing through it. bzip2.c
is now compiled with -D_FILE_OFFSET_BITS=64 to get large
file support from the C library. -v correctly prints out
file sizes greater than 4 gigabytes. All these changes have
been made without assuming a 64-bit platform or a C compiler
which supports 64-bit ints, so, except for the C library
aspect, they are fully portable.
* Decompression robustness. The library/program should be
robust to any corruption of compressed data, detecting and
handling _all_ corruption, instead of merely relying on
the CRCs. What this means is that the program should
never crash, given corrupted data, and the library should
always return BZ_DATA_ERROR.
* Fixed an obscure race-condition bug only ever observed on
Solaris, in which, if you were very unlucky and issued
control-C at exactly the wrong time, both input and output
files would be deleted.
* Don't run out of file handles on test/decompression when
large numbers of files have invalid magic numbers.
* Avoid library namespace pollution. Prefix all exported
symbols with BZ2_.
* Minor sorting enhancements from my DCC2000 paper.
* Advance the version number to 1.0, so as to counteract the
(false-in-this-case) impression some people have that programs
with version numbers less than 1.0 are in someway, experimental,
pre-release versions.
* Create an initial Makefile-libbz2_so to build a shared library.
Yes, I know I should really use libtool et al ...
* Make the program exit with 2 instead of 0 when decompression
fails due to a bad magic number (ie, an invalid bzip2 header).
Also exit with 1 (as the manual claims :-) whenever a diagnostic
message would have been printed AND the corresponding operation
is aborted, for example
bzip2: Output file xx already exists.
When a diagnostic message is printed but the operation is not
aborted, for example
bzip2: Can't guess original name for wurble -- using wurble.out
then the exit value 0 is returned, unless some other problem is
also detected.
I think it corresponds more closely to what the manual claims now.
1.0.1
~~~~~
* Modified dlltest.c so it uses the new BZ2_ naming scheme.
* Modified makefile-msc to fix minor build probs on Win2k.
* Updated README.COMPILATION.PROBLEMS.
There are no functionality changes or bug fixes relative to version
1.0.0. This is just a documentation update + a fix for minor Win32
build problems. For almost everyone, upgrading from 1.0.0 to 1.0.1 is
utterly pointless. Don't bother.