The spurious exit removed in 9365ae59 meant that 7 more tests are now
running.

A section of the lifecycle documentation describes the statuses
available during ticket creation, mainly to document what the ''
(empty string) transition key signifies. However, the reference
to the default configuration was out of sync with the actual
default configuration.

Mobile searches are just as bookmarkable as full searches.

it would be very handy for any extension that wants
to implement different algorithms for attaching
messages to tickets, for example by reference
headers or by ids from external systems.

First idea was to pass MIME::Entity right into
ParseTicketId function as second argument, but
this variant is more flexible. With this change
ExtractTicketId can be overriden to change where
id searched, but still have access to RT's
ParseTicketId that just finds id in any string.

It's convenient to save or pass around links to search results rather
than needing to save and share and then load the saved search.
While there are potential issues here, it's much more user friendly
to allow it.

Wipeout works to remove items from the shredder cache, so it
can delete entries from the hash before the iterator used by
each gets there. From perldoc -f each:

"If you add or delete a hash’s elements while iterating over it,
entries may be skipped or duplicated‐‐so don’t do that."

Converted to values, which does not use the iterator.

See also:

http://issues.bestpractical.com/Ticket/Display.html?id=16070

Original branch provided by Bradley Bell:
https://github.com/btb/rt/tree/4.0/stop-shredder-dataloss

The table_info fix in 361934d3 revealed that transactions and links
remain after a child ticket has been shredded. Marking test as
a todo.

The dump_current_and_savepoint function in shredder/utils.pl relies
on the table_info call from DBD::SQLite to return all of the tables in
the test RT database. The call was using empty strings as the first
two parameters, but instead of returning all tables, it returned none.

The first parameter is catalog which isn't supported in SQLite,
and the second, schema, seemed to be performing a LIKE and matching
nothing.

Based on the examples in the DBD::SQLite tests, passing undef
for the first two parameters and '%' for the third parameter, table,
returns all tables.

This change returns data for the test comparison and in
doing so reveals a failing test in t/shredder/01ticket.t

The primary improvement is noticeable when searching for tickets with
links to articles:

    RefersTo = 'a:42'

now works.  Previously the full URI form was required:

    RefersTo = 'fsck.com-article://example.com/article/42'

which is not terribly friendly and not (intentionally) exposed anywhere
else.

Due to the shredder tests not using RT::Test, no warning was actually
output (and no test was failed) by code in the previous commit.
Regardless, update the code to cope with the failure case as suggested.

Since 1e280ff2, we switch the current user by creating a new RT::Ticket
object and load the current ticket into it.  However, this also has a
subtle behavior change -- previously, if a ticket had ending
transactions and was removed from the underlying database, appropriate
scrips would still be triggered for it, despite the ticket no longer
existing.  Depending on the content of the scrips, this may have led to
unexpected results.  In the new behavior, the ticket object passed to
the scrips unexpectedly has no id.

Before running _ApplyTransactionBatch, detect if the ticket record is no
longer valid, and return, dropping the transactions; in doing so, warn
of the confluence of events that led to this scenario, rare as it is.

While this is technically a behavior change (from both before and after
1e280ff2), it almost certainly only arises during testing, and previously
generated subtly incorrect or inconsistent results.  The new behavior
reports the failure, maintains consistency, and suggests a solution.

Matching behaviour is modeled on the matching of SSL certificate CNs to
domain names.

Put it there for faster testing and forgot to delete

An email may say that it's UTF-8, but it
can contain anything else. We were trusting mail,
but there are too many MUAs, MDAs MTAs that do
things wrong.

If we don't do this conversion then at least on Pg RT fails
to create tickets with error 'invalid byte sequence for encoding
"UTF8"'.

Test file has help functions to test Perl templates
as well as Simple templates. Actually it does test Perl
templates.

two very simple and similar test files, template-insert.t
is just better and covers more. There is nothing to keep
in template.t

IsWatcher doesn't suffer from wrong email address.

AddWatcher checks validity, whether it's RT's address
and other things.