Multi-Threaded Multi-Node Utils (Mutil) / News: Recent posts

CHANGES
- Fixed preservation of execute permission on recent lustre versions
- Fixed compiler warnings for direct include of liblustreapi.h
- Fixed handling of long input lines when using --read-stdin

CHANGES
- Fixed final file size in some cases when src ends with hole
- Fixed --sparse=never corruption in some cases when --double-buffer enabled

CHANGES
- Fixed write error handling when double buffering enabled
- Added additional asynchronous i/o error handling

CHANGES
- Fixed ignored #mutil# prefix in msum during -c
- Fixed compiler warnings for gnutls_transport_set_ptr/LOV_MAX_STRIPE_COUNT
- Removed references to gnutls-extra since no longer exists in recent gnutls
- Removed note about hash leaf size adjustment during some cases
- Removed --as-needed from ld args since not always supported (e.g. OSX)
- Added workarounds when clock_gettime does not exist (e.g. OSX)
- Changed to System V semaphores since POSIX semaphores broken on OSX
- Now supports OSX (note that default gcc 4.2.1 does not support OpenMP)

CHANGES
- Updated to 8.x coreutils with improved sparse file handling/performance
- Added ability to enable/disable static linking of gcrypt/gnutls
- Added --preallocate option to minimize extents of non-sparse files
- Fixed gnutls tests to check for srp functions missing on centos/rhel
- Fixed missing ENONMEM during gcrypt thread enable on centos/rhel
- Fixed no longer used parameters to hash functions
- Fixed final file size during partial copy with hole at end

CHANGES
- Fixed intermittent deadlock due to misbehaving OpenMP locks (bug report and debugging support by G. Butler)

CHANGES
- Fixed segfault when hashing multiple portions of the same file that are larger than the split size during --read-stdin
- Fixed segfault when hashing unreadable file
- Added --dst-offset option to copy to specific position in target
- Added --stripe-count option to set absolute/size-relative striping
- Added --print-src option to show src file instead of dst during --print-stats and --print-hash
- Added ability to copy/hash zero length subsets of files
- Removed --store-hash option since not viable given xattr performance
- Added some notes about CentOS/RHEL in INSTALL

CHANGES
- Fixed various compiler warnings
- Fixed last zero-length read when --double-buffer enabled
- Fixed improper off_t parsing of various options
- Fixed offset of last posix_fadvise on write buffer
- Fixed unaligned file offsets during --direct-read
- Fixed alignment multiple, which must be higher than open man page states
- Added ability to set posix_fadvise as default in reads and/or writes

The Mutil 1.76.5 release had the wrong coreutils-7.6 patch (was still the patch for Mutil 1.76.4). The 1.76.5 patch was improperly named coreutil-7.6 (without an "s"). The release is now fixed to contain the correctly named files.

CHANGES
- Fixed references to fadvise/--fadvise instead of fadvise_read/--fadvise-read (bug report by W. Kettler)
- Fixed error messages referencing --read_stdin instead of --read-stdin

CHANGES
- Fixed race condition between last unaligned read and disabling direct I/O when --direct-read and --double-buffer enabled

CHANGES
- Added note about need for static version of libgpg-error
- Hash of stdin (i.e. piped input and file name "-") now works in msum
- Hash output format of mcp now exactly matches msum
- Added --check-tree option to isolate corruption
- Added --offset and --length options to mcp to copy specific portion
- Added file restriping to mcp based on file size for Lustre targets
- Added --print-stripe to mcp to print striping changes to stderr
- Fixed divide by 0 when hash leaf size is 0 due to default split size
- Changed hash output so non-standard output is reported in comments
- Now adjust split size to buffer size when 0 < split size < buffer size
- Fixed making of holes in mcp in double buffer case
- Split --fadvise into --fadvise-read and --fadvise-write
- Added --skip-chmod to mcp to retain temporary permissions used during copy
- Fixed --direct-read errors on some file systems due to last unaligned read

CHANGES
- Fixed high memory utilization during hashing of large files.

First public release of Mutil.