Large size of Mimefactor.render() futures
increases the size of all callers
down to set_config() and background_fetch().
I also had to Box::pin one call to fetch_new_msg_batch
and one call to fetch_single_msg.
Before, if the user fixed their clock incorrectly set to the future, they needed to delete
previously sent messages or wait until this future comes again so that new sent messages are added
to the bottom. Strictly speaking, the problem isn't fixable because we don't know if messages were
incorrectly timestamped into the future or they are timestamped correctly and the clock is now
incorrectly set to the past. Anyway, adding messages to the middle of the chat isn't a good way to
inform the user about the problem.
I used some AI to draft a first version of this, and then reworked it.
This is one of multiple possibilities to fix
https://github.com/chatmail/core/issues/8041: For bots, the IncomingMsg
event is not emitted when the pre-message arrives, only when the
post-message arrives. Also, post-messages are downloaded immediately,
not after all the small messages are downloaded.
The `get_next_msgs()` API is deprecated. Instead, bots need to listen to
the IncomingMsg event in order to be notified about new events. Is this
acceptable for bots?
THE PROBLEM THAT WAS SOLVED BY THIS:
With pre-messages, it's hard for bots to wait for the message to be fully downloaded and then process it.
Up until now, bots used get_next_msgs() to query the unprocessed messages, then set last_msg_id after processing a message to that they won't process it again.
But this will now also return messages that were not fully downloaded.
ALTERNATIVES:
In the following, I will explain
the alternatives, and for why it's not so easy to just make the
`get_next_msgs()` API work. If it's not understandable, I'm happy to
elaborate more.
Core can't just completely ignore the pre-message for two reasons:
- If a post-message containing a Webxdc arrives later, and some webxdc updates arrive in the meantime, then these updates will be lost.
- The post-message doesn't contain the text (reasoning was to avoid duplicate text for people who didn't upgrade yet during the 2.43.0 rollout)
There are multiple solutions:
- Add the message as hidden in the database when the pre-message arrives.
- When the post-message arrives and we want to make it available for bots, we need to update the msg_id because of how the `get_next_msgs()` API works. This means that we need to update all webxdc updates that reference this msg_id.
- Alternatively, we could make webxdc's reference the rfc724_mid instead of the msg_id, so that we don't need stable msg_ids anymore
- Alternatively, we could deprecate `get_next_msgs()`, and ask bots to use plain events for message processing again. It's not that bad; worst case, the bot crashes and then forgets to react to some messages, but the user will just try again. And if some message makes the bot crash, then it might actually be good not to try and process it again.
- Store the pre-message text and `PostMsgMetadata` (or alternatively, the whole mime) in a new database table. Wait with processing it until the post-message arrives.
Additionally, the logic that small messages are downloaded before post-messages should be disabled for bots, in order to prevent reordering.
We already set sort_timestamp to 0 for "Messages are end-to-end encrypted."
since 8f1bf963b4.
Do this for "Others will only see this group after you sent a first message."
and "Messages in this chat use classic email and are not encrypted." as well
so no messages can be added on top.
This avoids creating 1:1 chat on a second device when joining a channel.
Now when joining a channel there may be no 1:1 chat with the inviter
when the channel is created. In this case we still create the channel
as unblocked even if 1:1 chat would be a contact request
because joining the channel is an explicit action
and it is not possible to add someone who did not scan a QR
to the channel manually.
The part of logic there adjusting the sort timestamp forward if the parent message has a greater
sort timestamp wasn't tested explicitly by any test. I only saw one unrelated "golden test" failure
when commented it out.
(Related to #8027)
Fix https://github.com/chatmail/core/issues/8042
The problem was that after receiving the bcc_self'ed pre-message in
`receive_imf`, the logic there only looked for a pending
`smtp`-table-entry that matches the rfc724_mid, and if there was none
then it thought "Great, apparently the message is fully sent out, we can
mark it as delivered!".
But with pre-messages, the same message can have two `smtp` entries (one
for the pre-message and one for the post-message), and the message
should only be marked as delivered once both of them are sent out.
Now, I changed the logic to look for all entries with the same msg_id.
This is actually the same SQL query used in smtp.rs, so, I extracted it
into a new function; feel free to suggest a better name for it.
I tested on Android that it now works fine.
I'll add a test in a follow-up PR.
There are a lot of other problems with sending large files, though:
- The pre-message is sent before the post-message, so that for the
receiver it looks as if the message arrived, but stays in
"downloading..." forever
- There is quite a time delay between clicking on "Send" and the
outgoing message appearing in the chat
- The message shortly gets a letter icon right after it is sent
- I'm wondering if there is a way to give feedback to the user
immediately if the message is too big
- It's unclear when exactly we want to send read receipts
I'll open a follow-up issue for these.
This is to avoid sorting incoming messages that
are slightly in the past above system messages
about SecureJoin. SecureJoin messages are
timed according to smeared timestamp,
so even in the local tests they are in the future
by a few seconds.
We set timestamp of this info message to 0
to make it always appear in the beginning of the chat.
To avoid new chats being sorted to the end of the chatlist,
we ignore such 0 and use chat creation timestamp
when sorting the chatlist.
We know which message was added from the return value
of receive_imf(). It may be that the first chat
in the chatlist is not the one where the message was received
if there is a pinned chat or if
just received message is old.
It is not clear now what this is testing.
Golden test shows messages ordered
incorrectly according to the timestamps,
they should be ordered the other way round.
Comment talks about fetching from mvbox and inbox
in paralell which is a rare case that
could have happened if one message is left in the inbox
and the other message is a chat message moved to mvbox.
We never download anything that is not moved to the target folder.
The test also resides in "verified chats" tests
which are all legacy tests we kept after
replacing the concept of verified/protected chats
with key contacts in 2.x.
If our key is gossiped, the message is intended for us.
The check for address is redundant for incoming messages as
if we received the message then it was addressed to us.
This whole protection code can eventually be removed
as we have intended recipient fingerprints already,
it only protects against forwarding of messages
sent by old clients.
With `ORDER BY` statement SQLite searches
the `imap` table by `transport_id` and for each found row
scans the whole `imap_markseen` table.
Number of `imap` entries for each `transport_id`
is usually large as we need to know
which UIDs to delete on IMAP server
when deleting a message.
```
sqlite> EXPLAIN QUERY PLAN
SELECT imap.id, uid, folder FROM imap, imap_markseen
WHERE imap.id = imap_markseen.id
AND imap.transport_id=?
AND target = folder
ORDER BY folder, uid;
QUERY PLAN
|--SEARCH imap USING INDEX sqlite_autoindex_imap_1 (transport_id=?)
`--SCAN imap_markseen
```
Without `ORDER BY` statement SQLite scans `imap_markseen`
table which is expected to be small,
and then searches `imap` table by `rowid` for each found result.
```
sqlite> EXPLAIN QUERY PLAN
SELECT imap.id, uid, folder FROM imap, imap_markseen
WHERE imap.id = imap_markseen.id
AND imap.transport_id=?
AND target = folder;
QUERY PLAN
|--SCAN imap_markseen
`--SEARCH imap USING INTEGER PRIMARY KEY (rowid=?)
```
Query planning was tested with SQLite 3.52.0.
It is possible to explictly make
query planner move sorting to the last step
with `ORDER +folder, +uid`, but this is not recommended
in SQLite documentation
(see <https://www.sqlite.org/optoverview.html#uplus>).
It is also possible to add indexes,
but indexes use space,
adding them requires an SQL migration,
and each index needs to be updated so it will slow down writes.
Only inbox loop is changed because non-inbox loop is going to be removed
together with `mvbox_move`.
Added transport IDs to the log and logging around quota updates.
Removed some logs that add noise,
like logging that IDLE is supported each time right before using it.
The original idea was to always lock `write_mutex` before acquiring an `InnerPool.semaphore` permit
to avoid ABBA deadlocks, but when refactoring a PR for b696a242fc,
that was forgotten.
This doesn't really change the program flow as we have `Context::housekeeping_mutex` anyway,
just simplifies the code.
If we learn about this message being available on IMAP later,
we will add another available_post_msgs row.
If we don't delete the row, we will keep failing each time
until IMAP entry becomes available and it may not happen.
Previously, if the mvbox_move folder is missing, then core will loop
infinitely, because `new_mail` is never set to false.
The fix is to first set `new_mail` to false, then return if the folder
is missing.
This is the bug @hpk42 experienced when commenting in
https://github.com/chatmail/core/issues/7989
---------
Co-authored-by: holger krekel <holger@merlinux.eu>
