If an original references header field was supplied with the message, include
*that* in the header (since it should contain all the IDs of the replied-to
message thread) rather than just the message's reply_id (a single message
ID) when sending via NNTP. This maintains more NNTP post integrity when
sending back out via NNTP.

Not sure why we were storing received "references" in the hdr.references
returned by parse_news_header(), we don't seem to be using it anywere. Maybe
copied from 822header.js or mailproc_util.js? Seems like cruft, so remove it.

Also, don't ever send an "In-Reply-To" heder field via NNTP (it's not valid
/used for NNTP articles, per RFC 5536).

... so this was just redundant noise. Altough "References" is a valid header
field *both* for email and nntp articles, "In-Reply-To" is for email only.
per RFC 5536

We only want one Message-ID (we'll use the last, since that appears to be the
message being replied to based on SLRN behavior). This fixes reply linkage
when posing from newsreaders that put multiple message-IDs in the
references header field. This is required for SBBSecho to do its magic
creation of the FTN REPLY kludge during export.

This is the fix that Accession was looking for, I think.

This reverts commit d4692e0a.

Not the fix we're looking for.

This reverts commit a7f1a24a.

Not the fix we're looking for.

Use the fastest message base access methods documented.
Cache read/write configs and avoid parsing on read if unchanged.
Make common saves dependent on there being changes.

This is likely as good as it gets (assuming it works)

... for Acccession

Clean up as much as possible.

Almost no clients support a / in names.  Maybe I'll define an x-
extension or something when I write my own mail client.

This appears to be the only way to 100% provide mutual exclusivity between
Windows and Linux client sharing a Samba file system.

Also, make fmutex_close() take a pointer to the file descriptor so that we can
better manage the resource (i.e. don't close a descriptor that's already been
closed and possibly valid in another thread).

was printing the filename in uppercase if configured as such

Rather than a single file with config for all subs, use a separate
file for each sub.  This makes clients that do parallel connections
to different mailboxes work much better, and make loading and saving
the config for a mailbox much faster.

The caching stuff is removed to simplify things, and hopefully won't
need to come back... we should just not save config if it doesn't
change.

Only issue I'm still aware of is that 99% of IMAP clients just can't
deal with slashes in mailbox paths.  Like... at all.  The protocol
supports them, but every client seems to have a different way of
breaking with them.

No sense downloading articles we're just going to toss out the next time
we performance msgbase maintenance.

... which simplifies the fmutex_close() function and signature.

By default, the duration is infinite, but some sysops may not want this
file to be ever-grower but rather periodically pruned by the new trashman
utility.

Sysop that don't use spambait.cfg won't benefit from this change.

Also, don't get hostname unless we're surely going to write to the file.

The 'atomic_remove' argument isn't used in non-Windows builds since we can
remove() before close() on non-Windows OSes already.

I managed to hit a bug where bseen and seen were both completely
filled.  No idea how I hit it, so not positive it's fixed.

This is cheap insurance against the 43k file groung to 1.6MB for
someone else.

1) Cache the binified seen data rather than regenerate every time
2) Only call js.gc() after a complete fetch or store loop

O_EXCL doesn't appear to be working very reliably over Samba, so let's add
file locking to the equation.

glob() results might be out of date by the time we look at each file, just
skip missing QWK pack semaphore files silently. This should resolve the
".now already gone" messages too. Hopefully.

Removed some extraneous quotes around logged event command-lines.

Also when detecting a blocked event_thread, log the event_code that is/was
running last.

Should be no functional change.

Also, make save_cfg() require that the lock be held.

This should now actually work properly.
Also, add a debug_exceptions flag to rethrow exceptions so I can
get line numbers.

It shouldn't though.

Also, always use full_send(), not the socket send.