(on Linux) where the "total messages" value (stored in the SMB status header)
is stale when the header is updated in another open instance of the message
base.  Sorry, I forgot to complete the commit earlier than I originally said I
did.

of the first message in the thread, or 0 if unknown (e.g. legacy message). When
a new message is created, the thread_id value of the message being replied-to
will be used, else if the new msg is not a reply, then the new message's number
will be the value used for the thread_id.
This "thread-id" feature (Deuce's idea) allows threads with a common parent to
remain logically grouped after the original messages in the thread have been
purged (though the original hiearchy will start to be lost).
smb_addmsg() was retrieving the *header* of the replied-to message (via FTN
REPLYID): optimized by just retrieving the index.
Moved unused SMB fixed header fields: times_downloaded and last_downloaded
to provide more logitech binary alignment of the fixed header fields. Since
these fields are not yet used, this should have no affect on existing bases.

field value to be replaced.

Overhauled the hfield creation functions to allow insertion (prepending) of
header fields instead of always appending to the existing list fo fields (this
is necessary for the mail server to *insert* Received header fields as required
by RFC2821).
Upped the SMBLIB version 2.51.

an empty string, use the second SENDER for the "from" convenience pointer.

Bumped smblib version to 2.50.

The CRC headers now include gen_defs.h

effect.

of a message.
smb_putmsg() now calls smb_init_idx() automatically to initialize or
re-synchronize the index fields with the header fields - no longer need to call
smb_init_idx() from smb_addmsg() and other places (e.g. fixsmb).
Moved smb_init_idx() from smbhash.c to smblib.c.

correct offset (from the beginning of the index) in the msg struct.
smb_putmsgidx() will now return an error if asked to write to an index record
offset past the end of the file (typically leaving a "hole" of zeroes in the
index file).
smb_putmsgidx() and smb_putmsghdr() now returns the result of the final
fflush() call (normally 0, same as SMB_SUCCESS).

like that.

if required to fill "holes" in the passed smbmsg_t.

NET_INTERNET. reverse_path must point to an ASCII string and if from_net.type==
NET_FIDO, then reverse_path would point to garbage (binary FTN address).

smb_lockmsghdr() was sleeping because the lock attempt failed (the header was
already locked in smb_addmsg()). So smb_lockmsghdr() will now only sleep if
the unlock() call was successful.
Bugfix in smb_getmsghdr(), msg->total_hfields was incremented before the read
and malloc of hfield_dat, if the read failed, the subsequent call to
smb_freemsgmem() would segfault.

message from the end of the index (-1 = last message, -2 = second to last, etc)

been previously locked. We call smb_unlocksmbhdr() in a lot of places where it
may not actually be locked, in which case it over-writes the smb->last_error
string with an unlock failure (needlessly).

Created convenience macro: smb_hfield_bin().

unlock() returns failure (non-zero).
smb_open() now checks return value of smb_unlockmsghdr().

(or any useful error number). It just returns a "non-zero" value, the same
"non-zero" value regardless of the cause of failure. Still not positive if
fread() sets errno in any/all CRTLs or not. :-(

MSGID's and FTN MSGID's) so that a bit-mask may be passed to functions like
smb_findhash(), allowing selective duplicate searches without pre-marking
hashes (to be skipped) in a message's hash list. This limits the number of
possible hash sources to 32 total (down from 256) - not a real concern.

This change will render existing msg-id's in hash files obsolete.

The usage of the following functions were effected:
smb_addmsg() - from BOOL (dupechk) to bit-field (dupechk_hashes)
smb_msghashes() - removed dupechk kludge (to pre-mark text hash)
smb_findhash() - added source_mask argument to select which source types

This allows a single version of smb.lib to be linked with either single or
multi-threaded applications.

and allow for selective linking of objects from library archives.

smb_get_hfield() when passing a pointer to an hfield.
smb_hashes() now longer takes an smb_t* argument.