This allows multiple client/connector pairs to run at the same time. It also includes a few other miscellaneous small changes that accumulated as I went. They can be split if desired
- Whatever the `client_socket:send` line (~440) was doing with that missing operator, it's no longer doing. Don't ask me how it was working before. Lua is witchcraft.
- Removed the `settimeout(2)` which causes the infamous emulator freeze (and replaced it with a `settimeout(0)` when the server socket is created). It appears to be unnecessary to set a timeout for discovering a client. Maybe at some point in time it was useful to keep the success rate for connecting high, but it seems to not be a problem if the timeout is 0 instead.
- Also updated the Emerald setup to remove mention of the freezing.
- Connector script now picks the first port that's not in use in a range of 5 ports.
- To summarize why I was previously under the impression that multiple running scripts would not detect when a port was in use:
1. Calling `socket.bind` in the existing script will first create an ipv6 socket.
2. A second concurrent script trying to bind to the same port would I think fail to create an ipv6 socket but then succeed in creating an ipv4 socket on the same port.
3. That second socket could never communicate with a client; extra clients would just bounce off the first script.
4. The third concurrent script will then fail on both and actually give an `address already in use` error.
- I'm not _really_ sure what's going on there. But forcing one or the other by calling `socket.tcp4()` or `socket.tcp6()` means that only one script will believe it has the port while any others will give `address already in use` as you'd expect.
- As a side note, our `socket.lua` is much wonkier than I had previously thought. I understand some parts were added for LADX and when BizHawk 2.9 came out, but as far back as the file's history in this repo, it has provided a strange, modified interface as compared to the file it was originally derived from, to no benefit as far as I can tell.
- The connector script closes `server` once it finds a client and opens a new one if the connection drops. I'm not sure this ultimately has an effect, but it seems more proper.
- If the connector script's main function returns because of some error or refusal to proceed, the script no longer tries to resume the coroutine it was part of, which would flood the log with irrelevant errors.
- Creating `SyncError`s in `guarded_read` and `guarded_write` would raise its own error because the wrong variable was being used in its message.
- A call to `_bizhawk.connect` can take a while as the client tries the possible ports. There's a modification that will wait on either the `connect` or the exit event. And if the exit event fires while still looking for a connector script, this cancels the `connect` so the window can close.
- Related: It takes 2-3 seconds for a call to `asyncio.open_connection` to come back with any sort of response on my machine, which can be significant now that we're trying multiple ports in sequence. I guess it could fire off 5 tasks at once. Might cause some weirdness if there exist multiple scripts and multiple clients looking for each other at the same time.
- Also related: The first time a client attempts to connect to a script, they accept each other and start communicating as expected. The second client to try that port seems to believe it connects and will then time out on the first message. And then all subsequent attempts to connect to that port by any client will be refused (as expected) until the script shuts down or restarts. I haven't been able to explain this behavior. It adds more time to a client's search for a script, but doesn't ultimately cause problems.
Changelog:
Features:
- New goal
- Chaos Chao
- Raise a Chaos Chao to win!
- New optional Location Checks
- Chao Animal Parts
- Each body part from each type of animal is a location
- Chao Stats
- 0-99 levels of each of the 7 Chao stats can be locations
- The frequency of Chao Stat locations can be set (every level, every 2nd level, etc)
- Kindergartensanity
- Classroom lessons are locations
- Either all lessons or any one of each category can be set as locations
- Shopsanity
- A specified number of locations can be placed in the Chao Black Market
- These locations are unlocked by acquiring `Chao Coin`s
- Ring costs for these items can be adjusted
- Chao Karate can now be set to one location per fight, instead of one per tournament
- Items
- If any Chao locations are active, the following will be in the item pool:
- Chao Eggs
- Garden Seeds
- Garden Fruit
- Chao Hats
- Chaos Drives
- The starting eggs in the garden can be a random color
- Chao World entrances can be shuffled
- Chao are given default names
- New Traps
- Reverse Trap
Quality of Life:
- Chao Save Data is now separate per-slot in addition to per-seed
- This allows a single player to have multiple slots in the same seed, each having separate Chao progress
- Chao Race/Karate progress is now displayed on Stage Select (when hovering over Chao World)
- All Chao can now enter the Hero and Dark races
- Chao Karate difficulty can be set separately from Chao Race difficulty
- Chao Aging can be sped up at will, up to 15×
- New mod `config` option to fine-tune Chao Stat multiplication
- Note: This does not mix well with the Mod Manager "`Chao Stat Multiplier`" code
- Pong Traps can now activate in Chao World
- Maximum range for possible number of Emblems is now 1000
- General APWorld cleanup and optimization
- Option access has moved to the new options system
- An item group now exists for trap items
Bug Fixes:
- Dry Lagoon now has all 11 Animals
- Eternal Engine - 2 (Standard and Hard Logic) now requires only `Tails - Booster`
- Lost Colony - 2 (Hard Logic) now requires no upgrades
- Lost Colony - Animal 9 (Hard Logic) now requires either `Eggman - Jet Engine` or `Eggman - Large Cannon`
Modifies various access rules in the lufia2ac world with the aim of making them evaluate quicker.
Instead of having to determine the reachability of another location, they now only have to count items in state, which is faster.
(Also made it reuse the identical lambda for multiple locations, which might save a smidgen of memory.)
In a lot of cases, Factorio would write data to file first, then attach that file into zip. It now directly attaches the data to the zip and encapsulation was used to allow earlier GC in places (rendered templates especially).
* Lingo: Fix painting shuffle logic issue in The Wise
* Lingo: More generic painting cycle prevention
* Lingo: okay how about now
* Lingo: Consider Owl Hallway blocked painting areas in vanilla doors
* Lingo: so honestly I should've seen this one coming
* Lingo: Refined req_blocked for vanilla doors
* Lingo: Orange Tower Basement is also owl-blocked
* Lingo: Rewrite randomize_paintings to eliminate rerolls
Now, mapping is done in two phases, rather than assigning everything at once and then rerolling if the mapping is non-viable.
