Wayne Davison [Mon, 10 Jan 2005 10:03:10 +0000 (10:03 +0000)]
- Call finish_transfer() with its new arg. This ensures that we only
create a backup file if we're about to overwrite the basis file.
- Only clear make_backups for the redo phase if partial_dir isn't set.
Wayne Davison [Mon, 10 Jan 2005 00:31:55 +0000 (00:31 +0000)]
Moved the reading of the final MSG_DONE message here from main.c
(this message comes from the receiver). This ensures that any
redo-pass files and delete-after processing are known to be done
prior to the start of our hard-link and dir-time-munging loops.
Wayne Davison [Mon, 10 Jan 2005 00:31:55 +0000 (00:31 +0000)]
Moved the reading of the final MSG_DONE message from here to
generator.c (the message comes from the receiver). This ensures
that any redo-pass files and delete-after processing are known to be
done prior to the start of the hard-link and dir-time-munging loops.
Wayne Davison [Sun, 2 Jan 2005 09:08:59 +0000 (09:08 +0000)]
Backed out changes to send_deflated_token() that surrounded the
call to deflate(..., Z_INSERT_ONLY) -- the underlying bug was
caused by the zlib code not handling Z_INSERT_ONLY in the case
where the server has disabled compression.
Wayne Davison [Sun, 2 Jan 2005 00:55:55 +0000 (00:55 +0000)]
In send_deflated_token(), the section that handles "token != -2"
now breaks up the calls to deflate() into CHUNK_SIZE chunks, just
like the other sections of the code.
Wayne Davison [Sat, 1 Jan 2005 21:08:20 +0000 (21:08 +0000)]
- Use an int32 for the each block-size variable.
- Fixed a problem in send_deflated_token() where the data we are
compressing might not finish processing in one call.
Wayne Davison [Sat, 1 Jan 2005 21:08:11 +0000 (21:08 +0000)]
- Use an int32 for the each block-size variable.
- Improved the layout and error-checking of the code that
calculates the block-size and the number of checksum bits.
Wayne Davison [Sat, 1 Jan 2005 21:08:07 +0000 (21:08 +0000)]
- Use an int32 for the each block-size variable.
- Renamed the local block_size arg to blk_size (to avoid
confusion with the global block_size variable).
Wayne Davison [Fri, 31 Dec 2004 00:39:59 +0000 (00:39 +0000)]
The code that tries to read an error from the socket in an abnormal-
exit situation was properly forcing the io_timeout value down to 30
seconds, but failing to set the select_timeout value.
Wayne Davison [Sat, 27 Nov 2004 17:52:47 +0000 (17:52 +0000)]
- Added support for --copy-dest, which behaves like --link-dest,
but it copies the identical files instead of hard-linking them.
- Added support for multiple --*-dest options.
Wayne Davison [Sat, 20 Nov 2004 07:08:44 +0000 (07:08 +0000)]
Call set_msg_fd_in() during the early phase of being a client sender
so that we monitor the socket for any messages that it might send to
us (and thus avoid a potential hang when verbosity is high).
Wayne Davison [Sat, 20 Nov 2004 07:07:37 +0000 (07:07 +0000)]
Some minor improvements to read_msg_fd() made it safe to use both
set_msg_fd_in() and read_msg_fd() during the early phase of being
a client sender (up through the sending of the file list). This
makes sure that the sender monitors the socket from the receiver
for any messages that it might send to us, and thus avoids a hang
when verbosity is high.
Wayne Davison [Wed, 17 Nov 2004 19:29:20 +0000 (19:29 +0000)]
- Added the --bwlimit option to the options that are handled in
combination with --daemon. This value is both a default bwlimit
value and a maximum limit if the client asks for something larger.
- Enabled popt support for option aliases.
Wayne Davison [Wed, 3 Nov 2004 20:30:31 +0000 (20:30 +0000)]
Checking in the g2r-basis-filename patch that ensures that the receiver
uses the same basis file that the generator used (avoiding a duplicate
check that could cause a hang if a compare-dest file was a named pipe).
Wayne Davison [Mon, 11 Oct 2004 10:43:25 +0000 (10:43 +0000)]
- Got rid of some awkward spacing in the long_options table.
- Use the POPT_TABLEEND macro to end the long_options table.
- Improved the count_args() function to accept a NULL pointer.
- Simplified the code that sets *argc using count_args().
Wayne Davison [Fri, 1 Oct 2004 06:58:23 +0000 (06:58 +0000)]
The device-handling code is no longer omitted based on HAVE_MKNOD
(this is because do_mknod() might be using a different function
to handle fifos and sockets without the help of mknod().