You must use **spaces** and not tabs to indent your configuration file! (`\t` is the escaped code for a tab, which is not allowed in YAML)
### I get errors like "ParserError: while parsing a block mapping ... expected <block end>, but found '<block sequence start>'
This likely indicates an indentation issue. When you create a list in YAML (PyLink's config format), all entries must be indented consistently. For example, this is **bad**:
Take a few minutes to familiarize yourself with YAML, the markup language we use for the config file.
[CraftIRC](https://github.com/Animosity/CraftIRC/wiki/Complete-idiot%27s-introduction-to-yaml), [Ansible](https://docs.ansible.com/ansible/latest/reference_appendices/YAMLSyntax.html), and [Wikipedia](https://en.wikipedia.org/wiki/YAML) all provide excellent guides (with examples) on its basic structure.
A common misconception is that the YAML format is something specific to Python or PyLink, but this is not the case! YAML is a programming language-independent standard which *happens* to use indents for structures like Python does, but [parsers for it exist just about everywhere](http://yaml.org/).
The reason I (James) chose it for this project is essentially a restatement of its common benefits:
2) Is the target network's IRCd actually binding to the port you're trying to use? If there is a port conflict with another program, the IRCd may fail to bind to specific ports but *still start* on others which are free.
4) Is there a working connection between the source and target servers? Use ping to test this, as routing issues between providers can cause servers to become unreachable.
1) Check for recvpass/sendpass/server hostname/IP mismatches - usually the IRCd will tell you if you're running into one of these, provided you have the right server notices enabled (consult your IRCd documentation for how to enable these).
See https://github.com/jlu5/PyLink/issues/463 - the problem appears to be caused somewhere in Python's SSL stack and/or OpenSSL, and not directly by our code.
Unfortunately, the only workarounds so far are to either disable SSL/TLS, or wrap a plain IRC connection in an external service (stunnel, OpenVPN, etc.)
PyLink does not support inbound connections - much like regular services such as Atheme or Anope, it only connects outwards *to* IRCds. (If you don't understand what this means, it means you should turn autoconnect **off** for PyLink)
**No!** Only the PyLink administrator needs to host a PyLink instance with the `relay` plugin loaded, as each instance can connect to multiple networks. Everyone else only needs to add a link block on their IRCd.
First, check whether the SQUIT message includes the nick that triggered the netsplit. If this nick includes any characters not allowed in regular IRC, such as the slash ("/"), or is otherwise an invalid nick (e.g. beginning with a hyphen or number), this likely indicates a bug in PyLink Relay. These problems should be reported on the issue tracker.
However, if the nick mentioned is legal on IRC, this issue is likely caused by a max nick length misconfiguration: i.e. the Relay server is introducing nicks too long for the target network. This can be fixed by setting the `maxnicklen` option in the affected network's PyLink `server:` block to the same value as that network's `005``NICKLEN` (that is, the `NICKLEN=<num>` value in `/raw version`).
See https://github.com/jlu5/PyLink/blob/master/docs/advanced-relay-config.md#custom-clientbot-styles, especially the section "Disabling Colors/Control Codes".
### Relay is occasionally dropping users from channels!
This usually indicates a serious bug in either Relay or PyLink's protocol modules, and should be reported as an issue. When asking for help, please state which IRCds your PyLink instance is linking to: specifically, which IRCd the missing users are *from* and which IRCd the users are missing *on*. Also, be prepared to send debug logs as you reproduce the issue!
- Another tip in debugging this is to run `showchan` on the affected channels. If PyLink shows users in `showchan` that aren't in the actual user list, this is most likely a protocol module issue. If `showchan`'s output is correct, it is instead probably a Relay issue where users aren't spawning correctly.
Essentially, PyLink maps IRCd modes together by name, so modes that use different characters on different IRCds can be recognized as the same "mode". Tables of supported channel modes, user modes, and extbans (in 2.0+) can be found at https://github.com/jlu5/PyLink/tree/devel/docs/modelists. Note that third party/contrib modules implementing modes are generally *not* tested / supported.
Relay in particular uses whitelists to determine which modes are safe to relay: for 2.0.0, this is https://github.com/jlu5/PyLink/blob/71a24b8/plugins/relay.py#L903-L969. **Most *channel* modes recognized by PyLink are whitelisted and usable with Relay**, with the following exceptions:
- "registered" channel / user modes (InspIRCd, UnrealIRCd **+r**) - this is to prevent conflicts with local networks's services.
- "permanent" channel modes (commonly **+P**) - it's not necessary for remote networks' channels to also be permanent locally.
- Flood protection modes are only relayed between networks running the same IRCd (UnrealIRCd <-> UnrealIRCd or InspIRCd <-> InspIRCd).
- Modes and extbans that specify a forwarding channel - mangling channel names between networks is far too complicated and desync prone.
- InspIRCd's m_ojoin **+Y** and m_operprefix **+y** are ignored by Relay.
- auditorium (InspIRCd **+u**), delayjoin (UnrealIRCd, P10, InspIRCd **+D**), and any other modes affecting join visibilites are not supported.
Support for user modes is not as complete:
- Filter type modes such as callerid (**+g**), regonly (**+R**), noctcp (UnrealIRCd **+T**) are *not yet* supported by Relay.
- Service protection modes (UnrealIRCD **+S**, InspIRCd **+k**, etc.) are not forwarded by Relay to prevent abuse.
### How does Relay handle kills?
See https://github.com/jlu5/PyLink/blob/devel/docs/relay-quickstart.md#kill-handling
### How does Relay handle KLINE/GLINE/ZLINE?
It doesn't. https://github.com/jlu5/PyLink/issues/521#issuecomment-352316396 explains my reasons for skipping over this:
* The weakest link, whether this be a malicious/compromised/mistaken oper or a misconfigured services instance, can easily wreak havoc by banning something they shouldn't.
* KLINE relaying goes against the concept of partial network links and creates serious animosity when opers disagree on policy. If KLINEs are shared, opers are essentially shared as well, and this is not the goal of Relay.
This indicates either a bug in PyLink's protocol module or (less commonly) a bug in your IRCd. Hint: ENDBURST is likely not being sent or received properly, which causes service bot spawning to never trigger.
Make sure you're using an [officially supported IRCd](https://github.com/jlu5/PyLink#supported-ircds) before requesting help, as custom IRCd code can potentially trigger S2S bugs and is not something we can support.