William Morgan [Mon, 4 May 2009 12:29:54 +0000 (05:29 -0700)]
make sup-sync's --start-at actually useful
Using --start-at with sup-sync will now seek forward to the next valid message
if given an offset that's not on a message boundary. This makes it possible to
use that option without going through the laborious process of finding the
exact message boundary.
Also die unless --start-at is used with only one source, since using it across
multiple sources is almost definitely an error. (And if that's really what you
want, you can just call sup-sync multiple times.)
William Morgan [Tue, 28 Apr 2009 13:34:27 +0000 (09:34 -0400)]
fix mbox splitting regexp
I dunno. This helps with the "From problem", but at the expense of being
too specific than the mbox spec really demands. I don't think there's a
really right solution, in general (due to the mbox format being a
fundamentally broken one), but I'm hoping this will work with all modern
mbox files.
William Morgan [Sun, 26 Apr 2009 22:18:26 +0000 (18:18 -0400)]
don't automatically parse header on Message#new
You must now explicitly call #parse_header (e.g. when loading
from the index) or #load_from_source! (e.g. when loading from
disk).
This prevents redundant loading from the source when adding to the
index, as #snippet must be called before adding, and that requires
loading the entire message from the source.
William Morgan [Sun, 26 Apr 2009 22:16:48 +0000 (18:16 -0400)]
make Index#sync_message take an extra speedy arg
Previously it would check the index if docid and entry were nil,
even though nil is a valid result of a previous check. Add an
extra argument controlling whether this check is performed again.
William Morgan [Sun, 26 Apr 2009 15:44:37 +0000 (11:44 -0400)]
change email header parsing. MASSIVE SPEEDUP!
restructure the email header parsing code to improve speed and to
improve API. MBox::read_header now returns a hash containing all
headers, with keys downcased but nothing else.
update Message#parse_header to accept this new format.
update unit tests to reflect new format.
The exciting news is that changes in how the parsing works have resulted
in a 480% speedup in message parsing speed (634 message/s vs 132
messages/s on my machine), across ALL source types. hell yes.
Resync listable_labels and applyable_labels with reality
The listable_labels method listed all labels except unread
which seems an unneeded complication/restriction, so it's
renamed to all_labels. The applyable_labels was in fact
user-defined labels so it's renamed to user_defined_labels.
When responding to an email with a recipient_email field (i.e. an Envelope-To
header), look up the name in the account list, but keep the original email
address.
William Morgan [Wed, 25 Mar 2009 15:49:56 +0000 (08:49 -0700)]
overwrite from and to fields in index when saving a message
See comments in the code. This allows you to forcibly update the from/to
address stored in the index for messages, which means you can correct
the bad stuff stored in there due to previous versions of Sup that
canonicalized email addresses.
William Morgan [Wed, 25 Mar 2009 13:25:48 +0000 (06:25 -0700)]
rework Rakefile for "rake gem" and "rake tarball"
Move version into sup-version.rb and filelist into sup-files.rb. Rake gem
now builds the gem; rake tarball builds the tarball. No Hoe involved.
Update release-script accordingly.
William Morgan [Wed, 25 Mar 2009 12:33:45 +0000 (05:33 -0700)]
make ThreadIndexMode#save optionally threaded
The call to #save during #cleanup needs to block because this is where
state gets saved immediately before exit. Other calls to #save, e.g.
those triggered by "$", can be backgrounded.
William Morgan [Mon, 23 Mar 2009 15:55:09 +0000 (08:55 -0700)]
don't add :unread and :inbox for dupes
Allow merging in of labels from duplicate messages, EXCEPT for
:unread and :inbox labels. (These two have special semantics
and, in contrast to the other special-semantics labels, might
appear on a new message.)
William Morgan [Mon, 23 Mar 2009 14:11:16 +0000 (10:11 -0400)]
update unit tests to reflect header parsing
Blank headers now have a value of "", not nil. Nothing wrong with that.
The test captures behavior that doesn't really matter, but I've updated
it to at least reflect reality.
projects / sup / log