b/11436795
If an attachment download fails due to a timeout, or
an exception being thrown from startDownload(), we'd call
cancelDownload() on it. But this didn't actually cancel,
it would remove it from the inProgres list, but leave it
in the list of all downloads, so we'd immediately retry it.
This is bad for two reasons:
1. It can starve out other attachment downloads that could
have been successful.
2. It will keep attempting to do network work, even if it's
hopeless, forever, draining battery.
Now, if an attachment download fails in this way, for the first
few times, we'll move it to the tail end of the list of
downloads we'd like to perform. If it fails more than 10 times,
we'll give up completely. Giving up is not permanent, if we
have a reason to attempt a download again (such as the user
tapping on it), then it will get added back to the download
service and retried.
Change-Id: I5364a7d8b4b25ce299b8dcf061db6e9ce12daf75
b/11436795
Some of the logging I enabled here actually causes an
exception to be thrown because the format didn't match
the args in the log command.
Change-Id: If86942e64927c0e8df7573ef099824899e20c289
b/11535121
Now we only delete messages with the same serverId and account
if the account is an exchange account.
Change-Id: Ic2ebb465ccdb38724b88daac8ac40771c7a24bed
b/11294681
We had some really broken logic about handling search
results.
In IMAP search, we would request, in a single pass,
FLAGS, ENVELOPE, STRUCTURE, and BODY_SANE. BODY_SANE means
the first N bytes of message content, whether it be from
the message text or attachments. This is different from how
sync works: In sync, we get FLAGS and ENVELOPE in one pass,
and in a later pass get STRUCTURE and first body part text
for each message.
If the total size of the message exceeded the maximum limit
for BODY_SANE, then we'd mark the message as partial, which
would cause us to create a dummy attachment in copyMessageToProvider().
This is a weird solution to the problem of POP messages not
being completely loaded, because in POP message body and
attachments can't be requested separately, so the dummy attachment
just signified that we needed to fetch more data.
This system fails completely on IMAP, because just fetching the
rest of the body will not get you the attachments.
But even if that code is disabled, attachments in search results
still didn't work properly. For reasons I don't yet understand,
if we requet both STRUCTURE and BODY_SANE at the same time, either
we don't received the full attachment metadata, or we ignore it, and
only use the attachments whose contents could actually fit in the
limit imposed by BODY_SANE. So attachments that didn't fit,
or didn't completely fit, would either be missing or corrupt
and unretriveable.
So, end result: It's not clear why we were trying to load
BODY_SANE all in one pass, unlike how it works for sync.
In fact, the way sync does it now makes a lot of sense: We
load FLAGS and ENVELOPE data (small) and put the in the DB
immediately so they can be displayed. In the second pass we
load the (potentially large) structure and message body. If this
is the right solution for sync, it's probably the right solution
for search. So now, that's what we do.
There is cleanup I'd like to do post MR1: Some code is duplicated
between sync and search that could be consolidated, but we're in
low risk mode now so I only changed search code.
Change-Id: I11475e290cda04b91f76d38ba952679e8e8964d5
b/11436795
Now, if we ever insert or update an attachment to have
a blank location, we'll log a warning with stack trace.
Also, logging from ADS now uses the same log tag as everything
else, so we'll be able to see it without needing to turn
on some funny log tag.
Change-Id: Ic566cd87e8893128d074b897d7594a01ae12bc8c
For now, it sends the device model name as friendly name, in lieu
of actually having a user-supplied friendly name. This is wrong
for at least two reasons:
1) We need to have an actual user-supplied friendly name, but that's
not easy to find.
2) This really shouldn't be a provider query -- it should be something
the Exchange can know locally (ideally this is a system preference
but that's not currently implemented). This workaround just lets
us have some reasonable value that we can update easily.
Bug: 11161234
Change-Id: If83ad768736de19c9d0e833d1f86a6ce9daf5039
b/11294681
The problem is that when we try to open an attachment for a
message in search results, it fails. The reason is that part of
loading the attachment, we need to open the remote folder the
message is in. For search results, the message's mailboxKey is
the special fake "search_results" folder, which doesn't actually
exist on the server.
For this change, I've added a new column called "mainMailboxKey".
For search results, this column will be populated with the real
mailbox the message is in. It will be blank for other messages.
This is a quick and low risk fix for this bug, but it's kind
of awkward. We would prefer to do one or both of the following
some time after MR1.
1. Make the "search_results" folder be a virtual folder, the same
way that unread, starred, and other virtual folders are. For these,
there is actually no mailbox row in the database, just some
queries that check various flags in the messages and behave
like folders in the UI. The messages actually still reside in the
real folders.
2. Remove the requirement to open the folder at all to load the
attachment.
Change-Id: I825ab846f78bf8b041a5d1d579260dc5d7b4c522
