After finishing reading the Sieve script, the command parsing sometimes didn't
continue with the search arguments. This is a time-critical bug that likely
only occurs when the Sieve script is sent in the next TCP frame.

Occurs when several mails from different backend mailboxes are involved in the
COPY event. Fixed by using mail_save_context->>copy_src_mail in mailbox_copy()
instead of the mail argument. The latter can point to the backend mailbox for
virtual mailboxes, which makes no sense to the imapsieve plugin as it expects
only one source mailbox.

Panic was:

Panic: file imap-sieve-storage.c: line 337 (imap_sieve_add_mailbox_copy_event): assertion failed: (ismt->src_box == NULL || ismt->src_box == src_mail->box)

Change was:

lib-mail: message_address_parse() - Change fill_missing parameter to flags

test suite: enotify extension: Add tests for interaction between ":encodeurl" and variable size limits.

lib-sieve: enotify extension: Improve handling of variable size limit for ":encodeurl" variable modifier.

lib-sieve: variables extension: Improve handling of variable size limit for ":quotewildcard" modifier.

lib-sieve: variables extension: Respect UTF-8 character sequence boundaries when truncating variables.

Makes it easier to run with valgrind.

plugins: imap_filter_sieve: Put more effort in reconstructing a valid rcpt address for the envelope.

The sieve_user_email setting provides a fallback for when there is no Delivered-To header.

Although it was documented and implemented in essence, it was not actually
available.

Fixes e.g. FILTER+LOGOUT pipelining crash:
Panic: file imap-sync.c: line 832 (cmd_sync_delayed_real): assertion failed: (client->mailbox != NULL)

Before, the whole address list (header value) was returned as an invalid
address, ignoring any valid addresses it may still have contained.

It adds the ability to manually invoke Sieve filtering in IMAP.

Use the new SIEVE_EXECUTE_FLAG_SKIP_RESPONSES flag to prevent any unforseen
mishaps.

This causes the reject and vacation response actions to be skipped without
error.

Signed-off-by: Phil Carmody <phil@dovecot.fi>

The tests blow through these limits all of the time, which
makes the logs very noisy. These three cases are about 4/5ths
of the test logs. Presumably they're the ones that are also
most often needing expansion in real world use too.

Signed-off-by: Phil Carmody <phil@dovecot.fi>

Signed-off-by: Phil Carmody <phil@dovecot.fi>

It is the initial working directory and that may change during testing, so it is
not a good idea to determine this per test.

The problem mainly arose because the stream buffer position was used to
determine how much was written, while the stream buffer position is potentially
decreased by i_stream_try_alloc().

Panic was:

Panic: file istream.c: line 197 (i_stream_read): assertion failed: ((size_t)ret+old_size == _stream->pos - _stream->skip)