Previously, anyone (even across federation) could invite you to a
room and the bot would join. It may not have provided a useful purpose,
but it still joined all rooms it was invited to.
We now only join rooms when we're invited by a person who is actually
allowed to use the bot.
Fixes https://gitlab.com/etke.cc/postmoogle/-/issues/17
Fixes:
> `mailbox` of this room set to `test@domain@domain`
Previously fixed in 97a4d6c7bc, but it seems like we unintentionally
reintroduced this bug again at some point after that.
When someone first joins a room, they see some commands (`mailbox`,
`owner`, ..) and they know they are getters and setters, but they have
no good example as to how to use them.
Is it `!pm mailbox SOMETHING` or `!pm mailbox=SOMETHING` or something
else?
It's better if the introduction text gives you the full command you need
to get started (e.g. `!pm mailbox SOME_MAILBOX`), instead of a partial
command that you don't know how to use (e.g. `!pm mailbox` - this is
merely a getter and will not set your mailbox to `SOME_MAILBOX`).
Starting from this, I thought it would be a good idea to make all
option getters tell you how the commands are to be used. If you send
`!pm mailbox` and it tells you "not yet set", it should also tell you
how to actually set it (e.g. `!pm mailbox VALUE`).
If're an admin and mess up the `users` list, you won't see "owner"
commands.
If you're just a regular room user (not an admin, not an owner), you'll
only see the `help` command in the `help` message.
Both of these situations may make you wonder:
- is that all there is?
- earlier I saw more commands, so what's going on?
Adding "and accessible to you" hopefully clears things up, or at least
it tries to make the help message more correct.
Now that we use Match() in allowAdmin() as well, it's awkward to
have it return `true` when called with an empty admin list.
No admins defined was taken to mean "everyone is an admin".
We can either have a `len(users) == 0` check in `allowAdmin` which
rejects the request, or we can change `Match()` so that it doesn't
return positive responses when called with an empty list. Doing the
latter sounds better. It's more natural that matching against an empty list
will yield "no match".
This can be improved in the future, to show some additional information
about each mailbox like:
- "how many users are in that room"
- "which users are in that room"
- "who is the owner of the mailbox"
This can all be done later though.
Checking using `settings.Allowed` is odd. Not all commands are related
to setting configuration settings. Admin commands are coming in the
future, for which this is certainly not the case.
We now do access checks early on (during command processing), so command
handlers can be clean of access checks. If we're inside of a command
handler, the user is privileged to run it.
Retrieving the settings object for each and every option was wasteful
I don't like how we're doing custom formatting for optionMailbox in many
different places. This should be redone.
Before:
> * **`!pm help`** - Show this help message
> * **`!pm stop`** - Disable bridge for the room and clear all configuration
> * **`!pm mailbox`** - Get or set mailbox of the room
> * **`!pm owner`** - Get or set owner of the room
> * **`!pm nosender`** - Get or set `nosender` of the room (`true` - hide email sender; `false` - show email sender)
> * **`!pm nosubject`** - Get or set `nosubject` of the room (`true` - hide email subject; `false` - show email subject)
> * **`!pm nohtml`** - Get or set `nohtml` of the room (`true` - ignore HTML in email; `false` - parse HTML in emails)
> * **`!pm nothreads`** - Get or set `nothreads` of the room (`true` - ignore email threads; `false` - convert email threads into matrix threads)
> * **`!pm nofiles`** - Get or set `nofiles` of the room (`true` - ignore email attachments; `false` - upload email attachments
After:
> * **`!pm help`** - Show this help message
> * **`!pm stop`** - Disable bridge for the room and clear all configuration
> * **`!pm mailbox`** (currently `something@example.com`) - Get or set mailbox of the room
> * **`!pm owner`** (currently `@someone:example.com`) - Get or set owner of the room
> * **`!pm nosender`** (currently `false`) - Get or set `nosender` of the room (`true` - hide email sender; `false` - show email sender)
> * **`!pm nosubject`** (currently `true`) - Get or set `nosubject` of the room (`true` - hide email subject; `false` - show email subject)
> * **`!pm nohtml`** (currently `false`) - Get or set `nohtml` of the room (`true` - ignore HTML in email; `false` - parse HTML in emails)
> * **`!pm nothreads`** (currently `false`) - Get or set `nothreads` of the room (`true` - ignore email threads; `false` - convert email threads into matrix threads)
> * **`!pm nofiles`** (currently `false`) - Get or set `nofiles` of the room (`true` - ignore email attachments; `false` - upload email attachments)
Works around:
> utils/user_test.go:5:1: cognitive complexity 18 of func `TestRuleToRegex` is high (> 15) (gocognit)
Alternatively, we may look for a way to skip this test only.
It doesn't seem complex at all, so it looks like some false-positive.
This used to return an error back when it was dealing with wildcards
(which may or may not have compiled to a valid regex).
But it now deals with pre-compiled regexes and has no chance of failing,
so we need no `error` returns.
This does not do anything useful just yet.
It will be hooked to access control checks later on.
Wildcards are converted to regular expressions, because it was simpler
to do that than to write (or include) some ugly wildcard matcher library.
It also provides more flexibility, should we wish to use it later.
Regular expressions should also work well performance-wise.
We compile wildcards to regexes early on (during configuration
processing) and fail if we detect a bad pattern. This is meant to
catch various problems (typos or other mistakes) that could happen.
For this to work, configuration building had to be redone, since it can
now return an error. This may be useful in the future for validating
other configuration settings.
Related to https://gitlab.com/etke.cc/postmoogle/-/issues/1
If tests (`make test`) fail, the file won't be removed and the git
working tree will appear dirty. Someone might accidentally commit this
`cover.out` file.
In various places, we build messages using `Sprintf` before passing them
to `Notice()`.
If we let `Notice()` do string formatting, we run the chance of having
it try to format our already-preformatted text. If our text includes
format references (e.g. `%s`), it would cause a problem (`%s(MISSING)`).
One way to trigger it is to change the bot prefix from `!pm` to `%pm`.
Doing so, `sendHelp()` would create some help message which contains
`%pm` references. `Notice()` would then try to process them as well,
leading to a bunch of `%!p(MISSING)m` in the final text.
It previously said "the following commands" were supported
and it was only listing subcommands (help, stop, ..)
without any indication of how the user should construct the full command
(`PREFIX SUB_COMMAND`).
For perfect clarity, we now list full commands in the help message. Example:
> The following commands are supported:
>
> - !pm help - Show this help message
> - !pm stop - Disable bridge for the room and clear all configuration
> - !pm mailbox - Get or set mailbox of the room
> - !pm owner - Get or set owner of the room
> - !pm nosender - Get or set nosender of the room (true - hide email sender; false - show email sender)
> - !pm nosubject - Get or set nosubject of the room (true - hide email subject; false - show email subject)
The new help message is prefix-aware, instead of hardcodign `!pm`.
If the bot is running with a custom prefix (not `!pm`), this is even
more helpful, as it lets people know what the prefix is. Reading
documentation elsewhere and seeing `!pm STUFF` will no longer confuse
anyone.
With this change, we also make use of the existing `Notice()` function,
so we don't need to duplicate some code related to sending notices.
Without this, if settings are not found in account data (`M_NOT_FOUND`),
we won't initialize the map and we'll panic later:
> assignment to entry in nil map
Breaking settings for all new rooms is definitely not what we inteded.
Likely a regression since 726bc95c26, but related to fcac0a202d as
well.
This also adds `Mailbox()` and `Owner()` getters for completeness.
Wording has been changed a bit to avoid saying "that room". It sounds
better if it's "this room" or just "the room".
Sanitizing options on Get() ensures that when someone asks
for a given option which may not be defined (`nosubject`, `nosender`),
we'll return a valid value (`'true'` or `'false'`) and not `''` (empty
string, undefined).
This way, users do not need to wonder if "nosender is not set" is
handled like "true" or "false" or in some 3rd way. They also don't need
to think about "how to unset this setting, now that I've set it to something".
Options will appear to have a default sanitized value no matter if
they've explicitly been set or not.
The `NoSender()` and `NoSubject()` getters are just there for
convenience, so that we won't need to do casting in other places.
The configuration setting is called `Hide*` instead of `Show*`, because
it's backward compatible with existing configuration settings.
This is useful for when you setup an email forwarding inbox and you're
always sending to it through the same email address. In that case, you
don't need to see the email address in each Matrix message.
In the future, another similar `bool` setting (`hide-subject`) will land,
which controls whether the email's subject is shown in the final message
or not. That setting can make use of most of the same setup (all of
`handleBooleanConfigurationKey`).
Slavi PantaleevandAine