Changelog
bar (1.11.0+debian-2) unstable; urgency=low
* changed the target (stable => unstable). The previous settings were
inappropriate. Tanks to Adam D. Barratt who reminded it. Closes: #593260
bar (1.11.0+debian-1) stable; urgency=low
* upgraded to the new upstream version. Copied part of the debian changelog
from upstream to the file Changelog. The current changelog is now
reserved for maintainer's messages. Added the '+debian' suffix in
the package name
bar (1.10.9-1) stable; urgency=low
* first publication in Debian. Closes: #593260
* taken the maintenance of the package with Michael Peek's agreement
* upgraded Standards-Version to 3.9.1 and compat to 7
* added dependencies on misc:Depends and shlibs:Depends
* protected "-$(MAKE) distclean" in debian/rules
* replaced "dh_clean -k" by "dh_prep" in debian/rules
* commented out "$(MAKE) check" in debian/rules as this target is rather
done for the mainstream developer
* replaced the full text citation of the GPL-2 by a pointer, in
debian/copyright
bar (1.10.9) stable; urgency=low
* Bugfix: Grant McDorman found a bug in args.c in which the code was not
ANSI-C compliant, and submitted a patch to fix. This patch has been
incorporated into the source. Thanks Grant!
bar (1.10.8) stable; urgency=low
* Feature added: -c / --continue <size>
Allow the user to instruct bar that this is a continuation of a previous
transfer, and that we are continuing from size <size>. Use of this option
will throw off throughput and ETA calculations at the first, with the
numbers settling down a bit as the transfer continues.
* Feature added: "cp mode"
If no -if, --in-file, -of, --out-file command line options are used, and
unknown command line options are encountered, then bar will assume that
the first unknown command line option encountered is a path to an input
file, and the second (if found) is the path to an output file.
When this feature is implemented, if the destination is a directory, then
bar will create a file of the same name as the input file in the
destination directory, and will attemt to preserve file modes.
bar (1.10.7) stable; urgency=low
* This is a bugfix release:
- Do not exit with an I/O error if the error is EAGAIN.
bar (1.10.6) stable; urgency=low
* This is a bugfix release:
- A bug in args.c caused incorrect command line parsing. This is why I
shouldn't use one-letter variables, especially variables that look
similar, like 'c' and 'o'. Thanks to Carl Lowenstein for catching this
bug and helping me to trace it down it so quickly.
bar (1.10.5) stable; urgency=low
* Modified configure scripts:
- Now tests to find the page size of the compiling system and uses this
value by default. This can be overridden w/ --with-page-size.
- Compiles and runs tests to see if sysconf() returns a usable value. On
some systems -1 was being returned, but since the configure script found
a sysconf() function it was configuring bar to use it, but bar would
bomb with an error when it tried to use posix_memalign() w/ -1 as the
system's memory page size. So now, even if you have sysconf()
available, if sysconf() returns -1 for the page size, bar will be
configured not to use aligned memory.
- Compiles and runs tests to see if posix_memalign() is usable. In the
event of failure, separate assertions in the test code will reveal
whether posix_memalign() failed because if EINVAL, ENOMEM, or if there
was some other error. (This test could be made smarter. At the moment
the only action taken is that the configure script will log the
assertion to config.log. But it wouldn't be too hard to notify the user
of the error returned and/or take some specific action.)
- In the event that posix_memalign() checks fail, retry the test two more
times. Once defining _GNU_SOURCE and, if that fails, a second time
defining _XOPEN_SOURCE=600. I've not run accross any systems that need
these defined, but according to some man pages they may need be.
- Compiles and runs tests on aligned memory usability -- merely a
formailty at this point though. The above checks should be sufficient.
- Prints out configuration settings for the user in an easily-readable
format, so that it is obvious from the start how bar is configured to
operate. Also, some helpful text is printed on how to troubleshoot bar
and point the user to the new TROUBLESHOOTING text file.
* Fixed a reported bug in io.c's use of posix_memalign() reported by Doncho
N. Gunchev.
* New file, TROUBLESHOOTING, gives the user some things to try and
instructions on how to report bugs.
* ChangeLog is now a link to debian/changelog. All future changelog entries
will follow a debian format.
* Original changelog contents:
2005-04-01: Release 1.10.4
- Minor code change to allow bar to compile on systems without
snprintf().
2005-01-19: Release 1.10.3
- Bug sighted in ioWrite() by Chris Gentle, sank same:
avail = io.buffer_size = io.buffer_used;
Not only should this have been:
avail = io.buffer_size - io.buffer_used;
But the variable avail is not even used in ioWrite(). Shame on me for
not noticing.
The offending lines have been removed.
2004-11-23: Release 1.10.2
- Discovered the problem with memalign. Apparently memalign has two
definitions depending upon your version and flavor of OS:
memalign(size, boundary)
and
memalign(boundary, size)
New configure tests were designed, but bar still segfaults on some
systems, even after identifying which flavor of memalign to use.
Tests seem to show that using aligned memory gives less than a 1%
speed increase, so I have decided to not worry about it. The use of
memalign has been discontinued, although the use of posix_memalign is
still used if found. If posix_memalign is available, and
--disable-use-memalign has not been set, then use posix_memalign,
otherwise use malloc.
- Fixed a bug in the use of posix_memalign.
- Bar now has some default colors compiled in, in case ansi is turned on
and no colors are defined via a .barrc file.
- Removed test-args from the list of tests. The command line arguments
haven't changed in many versions now, and I have no intention of
changing them in the future.
2004-11-10: Release 1.10.1
- The strange I/O bug bit again. Tim Bynum reported an esoteric read
error under cygwin that sounded suspiciously similar to the EFAULT bug
seen earlier. With his help, I was able to track down the problem to
my use of aligned memory as an I/O buffer.
Created some elaborate tests within configure to try to reproduce the
problem so that the use of aligned memory allocation can be turned off
automatically on systems where this bug occurs:
Changed the way aligned memory was tested in the configure script:
a) Test to see if aligned memory is supported by the OS
b) If so, test to see if aligned memory works with read()/write()
c) If so, was --enable-use-memalign passed to configure
d) If so, attempt to use aligned memory allocation.
Changed the way vectored I/O was tested in the configure script:
a) Test to see if vectored I/O is supported by OS
b) If so, test to see if vectored I/O works with our chosen method of
memory allocation
c) If so, was --disable-use-iovec passed to configure
d) If not, attempt to use vectored I/O
Unfortunately, the tests pass with flying colors, while bar still fails.
Aligned memory allocation is now disabled by default. It can be enabled
by passing --enable-use-memalign to configure.
- Added a test to the configure script such that if HAVE_SYSCONF is
defined, check to see if _SC_PAGE_SIZE and _SC_PAGESIZE are defined.
Use whichever one is found.
- Spellcheck! It's a wonderful thing.
2004-11-09: Release 1.10
- The new changes to the I/O have been thoroughly tested on a number of
architectures and are believed to be safe.
2004-09-22: (development continues)
- The ring buffer seems to have problems with readv() on some
architectures. Investigating...
On the RedHat-9 system of the sourceforge compilefarm, readv() returns
an error and sets errno to EFAULT (segmentation fault) when
vec[0].iov_base > io.buffer. The only reference to EFAULT I could
find was in _UNIX System V Network Programming_, where EFAULT is said
to be set when there is a pointer problem. But the values of vec[]
have been verified as valid. On a Fedora Core 2 system this does not
happen. In fact, on no other system tested has this problem surfaced.
I'm clueless.
Therefore I am exploring the possibility of using a workaround. I
have written a configure test to check the operation of readv(). If
the problem is found then iovec I/O will be disabled and bar will use
regular read()/write() instead. The configure option
--disable-use-iovec has also been added, so that the user may override
the results of the configure test and disable the use of iovec even if
configure thinks it's safe.
2004-09-22: (development continues)
- Incorporated changes submitted by Christopher Chan-Nui to increase
throughput and decrease CPU usage. His changes include implementing a
page-aligned circular buffer in memory, and a larger default buffer
size.
The default buffer size can now be changed by passing
--with-buffer-size=SIZE when running configure. The new default buffer
size is 512K.
Changing your buffer size may or may not create a difference,
depending on your application of bar. For something fast, like I/O to
memory or to a fast hard drive, a larger buffer of say 1M might be
worth your while. But for I/O to something slow, like a tape drive,
the cost of using so much memory might not outweigh the increase in
performance.
- Chan-Nui's patch also changed the way the twiddle is updated.
Specifically it only updated the twiddle (a) if there was data written,
and (b) only when the rest of the line was updated. I decided to use a
twiddle as feedback that something was happening even for long update
delays. The problem is that on some machines, notably Linux on PC's,
the twiddle goes so fast that all it does is slam a slow terminal
connection, creating a bottleneck. Therefore I have made two changes
to the way the twiddle is updated:
1) The twiddle is turned off by default. This means that in order to
see the twiddle, you have to either specify -dtw or -da on the command
line, or you have to put "display-twiddle: yes" in the resource file.
This new default behavior can be changed by specifying
--enable-default-twiddle when running configure.
2) The twiddle will only update once a second, and then only if data
was actually copied.
- These changes have drastically increased the throughput of bar on both
SPARC and Intel architectures. As a comparison between version 1.09
and 1.10, the following throughput was seen using various buffer sizes
with the following command line:
bar -if /dev/zero -of /dev/null
Where:
bar-1.09 bar-1.10
----------------------------------- -----------------------------------
buffer | throughput buffer | throughput
size | i686-Linux E4500-Solaris size | i686-Linux E4500-Solaris
-------|------------|-------------- -------|------------|--------------
1K | 83.2MB/s | 8.4MB/s 1K | 135.2MB/s | 16.0MB/s
2K | 116.2MB/s | 11.8MB/s 2K | 269.7MB/s | 32.1MB/s
4K | 144.0MB/s | 14.1MB/s 4K | 520.7MB/s | 56.9MB/s
8K | 162.9MB/s | 15.5MB/s 8K | 973.9MB/s | 92.3MB/s
16K | 137.5MB/s | 16.4MB/s 16K | 1.4GB/s | 136.3MB/s
32K | 138.8MB/s | 16.9MB/s 32K | 2.7GB/s | 178.7MB/s
64K | 127.6MB/s | 17.2MB/s 64K | 5.0GB/s | 211.6MB/s
128K | 135.1MB/s | 17.2MB/s 128K | 9.0GB/s | 233.0MB/s
256K | 133.7MB/s | 17.3MB/s 256K | 14.8GB/s | 242.2MB/s
512K | 129.8MB/s | 17.3MB/s 512K | 22.1GB/s | 248.4MB/s
1M | 128.2MB/s | 17.3MB/s 1M | 29.2GB/s | 251.1MB/s
2M | 126.9MB/s | 17.3MB/s 2M | 35.0GB/s | 255.5MB/s
4M | 126.5MB/s | 17.1MB/s 4M | 38.0GB/s | 255.3MB/s
8M | 125.5MB/s | 17.1MB/s 8M | 39.5GB/s | 246.9MB/s
16M | 124.7MB/s | 16.2MB/s 16M | 40.6GB/s | 254.2MB/s
32M | 126.1MB/s | 16.0MB/s 32M | 40.8GB/s | 244.6MB/s
64M | 128.0MB/s | 15.5MB/s 64M | 41.2GB/s | 248.9MB/s
128M | 128.0MB/s | 15.6MB/s 128M | 41.4GB/s | 253.3MB/s
256M | 129.0MB/s | 15.6MB/s 256M | 41.9GB/s | 243.3MB/s
512M | 128.0MB/s | 14.4MB/s 512M | 40.9GB/s | 227.6MB/s
1G | 62.1MB/s | 13.5MB/s 1G | 40.9GB/s | 202.1MB/s
2004-05-27: Release 1.09
IDEAS:
- Add an option to handle multi-volume output, and display progress of
current volume. Problems to overcome:
- Automatically adjusting expected volume size:
- If no volume size is given, count the first volume and expect that
much for subsequent volume?
- If for instance you're writing to a tape drive with hardware
compression, and the volume size changes from one volume to the next
dependent on the data written, how should bar handle the expected
size of the next volume?
- Retrying a failed volume:
- Where do you store the data to be re-written?
- Knowing the difference between failed media and end-of-media?
- Should multi-volume capability be a separate program that bar pipes
to, or should I overhaul bar itself and make it vastly more
complicated?
- How to handle closing and re-opening a device between media changes:
- At the end of a volume, should bar:
- Prompt and wait for input?
- Run a program or script?
- Both (configurable)?
- When opening the next volume, should bar:
- Open the same device?
- Prompt for a new device?
- Read output from a program or script?
- Some configurable option to select one of these three?
Incorporated 'b'lock-size changes submitted by Michael T Bendzick.
Added the -bl/--block-size command line option and the "block-size"
resource file keyword to allow the user to specify the default block
size used. This option is only used in specifying the size of the
expected data stream on the command line and does not affect the I/O
routines. When specifying data size in terms of blocks the -k (size of
a kilobyte) is ignored.
Incorporated a spelling correction by Michael T Bendzick.
Added -dtp/--display-total-percent/-ntp/--no-total-percent command line
options, and the "display-total-percent" resource file keyword to allow
the user to specify that bar is not to display the total percent of
expected input in summary reports.
Fixed minor bug where the background color for a space was not
initialized. Added -ti/--title command line option and "title" resource
file keyword to specify a title to display on the status line during
data I/O. This option takes a string up to 80 characters in length
(excess characters will be truncated).
Added -dti/-nti/--display-title/--no-title command line options and the
"display-title" resource file keyword to toggle the displaying of a
title. Even if activated, nothing will be printed if no title string is
set.
Added -tifg/-tibg/-tib/-tin/--title-foreground/--title-background/
--title-bold command line options and "title-foreground",
"title-background", "title-bold" resource file options to specify the
colors used in displaying the title string.
2004-01-06: Release 1.08
Added an option to display the throughput as bits/sec instead of as
bytes/sec.
Added an option to display the the data count as bits instead of as
bytes.
2003-12-15: Release 1.07
Added an option to throttle the throughput to a given number of bytes per
second.
Modified autoconf yet again for hopefully greater portability.
Configure now assumes -sw-1 by default on *-pc-cygwin systems.
2003-12-13: Release 1.06.3
Fixed another signal handler error that would cause bar to exit with
errno == EINTR if the window size changed during a call to select().
2003-12-10: Release 1.06.2
Fixed a bug that could potentially hose the terminal: installed a signal
handler that removed O_NONBLOCK from I/O descriptors when Control-C is
pressed.
Minor cleanup of some code.
2003-12-09: Release 1.06.1
Added an option to force bar to display the elapsed time only, not the
eta.
2003-11-13: Release 1.06
Chris Gentle pointed out that bar is unable to open files greater than
2gb under Linux. Modified args.c to include O_LARGEFILE (if defined) to
open() when opening files for the --in-file and --out-file arguments.
Modified autoconf script to test for native Large File support. If that
test fails, autoconf will test to see if defining _FILE_OFFSET_BITS or
_LARGE_FILES allows bar to open and stat files greater than 2gb.
Added error.h/error.c to give more descriptive error messages when stuff
fails.
Added O_CREAT to open() for -of/--out-file option to allow bar to create
a file, and modified the creation mode for rw-rw-rw-. (Duh!)
Added #include <sys/time.h> to io.c for i686-pc-cygwin.
Added test-types.c to facilitate porting tests.
Fixed a bug in args.c number parser where the command line option --size
18446744073709551615 would produce a parse error resulting from the
inability of the parser to find the nonexistent units multiplier
character.
Tested bar on the various platforms available on the SourceForge compile
farm. See PORTING for a list. Cleaned up code and many compile
warnings on several platforms.
Modified number parser in args.c to be smart enough to recognize that "."
followed by a units multiplier is not a valid number. (Ex: -s .m)
Added resource file parsing for ~/.barrc and ./.barrc. Required
revamping argument parsing in args.c.
Fixed a bug in test-args.c whereby test110() would fail on line 15321 on
systems where size_t is 8 bytes instead of 4.
2003-11-04: Release 1.05
Chris Gentle supplied a patch for the --no-summary command line option.
Added code to query the terminal to find the terminal width and to update
itself when the terminal size changes.
Rewrote command line parsing algorithms for cleaner code. Fixed an error
when parsing fractional size arguments. (i.e. "<x>.<y>" numbers, such as
"1.38m".)
Ran ispell on the stupid manual page.
2003-10-31: Release 1.04
Chris Gentle supplied a patch for the --no-time command line option.
Updated display code to include petabyte (2^50) and exabyte (2^60) in
addition to kilo, mega, giga, and tera.
Updated command line argument parser to include 'P'etabyte and 'E'xabyte
modifiers to size arguments.
Added -n* and --no-* command line options to turn off pieces of the
display.
Added -d* and --display-* command line options to turn on pieces of the
display.
Added -sw and --screen-width command line options to allow the user to
set the width of the display.
Added types.h and supporting autoconf tests for type uint64.
Updated man page.
2003-09-09: Release 1.03
Anand Mitra supplied modifications that drastically reduced CPU usage.
Minor code restructuring for optimization.
Minor changes to autoconf scripts and code, (hopefully) to make bar more
portable.
Introduced a manual page.
Introduced command line option -t / --timeout as a way to specify the
number of microseconds select() is to wait for a change in I/O.
Reversed the order of the ChangeLog file so most recent changes appear
first. (Duh.)
One more minor bug fix to the display code.
2003-08-20: Release 1.02
Minor bug fix to autoconf.
2003-08-18: Release 1.01
Minor bug fixes in the display algorithms.
2003-08-14: Release 1.0
Initial release.
# vim:et:ts=2:shiftwidth=2
bar (1.10.4) stable; urgency=low
* Initial Release.
-- Ubuntu Archive Auto-Sync <email address hidden> Wed, 10 Nov 2010 10:36:05 +0000
