On some systems, the DB upgrade was failing because there were
too many rows for the cursor window. This moves some looping
operations into single update statements by using the substr()
command, and chunks the rest using a series of LIMITs.
1) Display the individual sender name in a group conversation.
2) Add an "address" column to MmsDatabase and keep FROM there.
3) Remove all blocking operations from MmsDatabase.Reader path.
4) Strip SMIL and other undisplayable parts from part count.
5) Fix places where messages weren't being correctly decrypted.
1) We now delay MMS notifications until a payload is received,
or there's an error downloading the payload. This makes
group messages more consistent.
2) All "text" parts of an MMS are combined into a second text
record, which is stored in the MMS row directly rather than
as a distinct part. This allows for immediate text loading,
which means there's no chance a ConversationItem will resize.
To do this, we need to include MMS in the big DB migration
that's already staged for this application update. It's also
an "application-level" migration, because we need the MasterSecret
to do it.
3) On conversation display, all image-based parts now have their
thumbnails loaded asynchronously. This allows for smooth-scrolling.
The thumbnails are also scaled more accurately.
1) We now try to hand out cursors at a minimum. There has always been
a fairly clean insertion layer that handles encrypting message bodies,
but the process of decrypting message bodies has always been less than
ideal. Here we introduce a "Reader" interface that will decrypt message
bodies when appropriate and return objects that encapsulate record state.
No more MessageDisplayHelper. The MmsSmsDatabase interface is also more
sane.
2) We finally rid ourselves of the technical debt associated with TextSecure's
initial usage of the default SMS DB. In that world, we weren't able to use
anything other than the default "Inbox, Outbox, Sent" types to describe a
message, and had to overload the message content itself with a set of
local "prefixes" to describe what it was (encrypted, asymetric encrypted,
remote encrypted, a key exchange, procssed key exchange), and so on.
This includes a major schema update that transforms the "type" field into
a bitmask that describes everything that used to be encoded in a prefix,
and prefixes have been completely eliminated from the system.
No more Prefix.java
3) Refactoring of the MultipartMessageHandler code. It's less of a mess, and
hopefully more clear as to what's going on.
The next step is to remove what we can from SmsTransportDetails and genericize
that interface for a GCM equivalent.
Yet another setting that most users will never touch. Workaround for
those who would is to use a different identity key per device.
Let this be a sacrifice to the android settings design pattern gods.
The vast majority of users will never uncheck this option. Those who
would can send an unencrypted untagged message through the system sms
app. It would then be stored locally in the clear, but it was already
transmitted in the clear and likely stored on the recipient's side in
the clear, so the security gains of locally encrypting are low, and
again, this seems an extremely rare edge case.
By android design pattern specs for the settings menu, we should kill
this preference.
Android design pattern best practice recommends placing the settings
menu item below all other items because it's not frequently needed.
In that spirit, I've also moved the clear passphrase option to be first
since it is likely to be used more often than database import/export.
