doc update

master
Perttu Ahola 2010-11-29 23:07:03 +02:00
parent 05507afcba
commit 4a8973aeac
2 changed files with 18 additions and 48 deletions

View File

@ -56,6 +56,10 @@ Compiling on GNU/Linux:
- Neither of which can probably be found from your distro's repository.
- Compiling each of them should be fairly unproblematic, though.
Compiling on Windows:
- Be sure to
#define JMUTEX_CRITICALSECTION
in jmutex.h before compiling it. Otherwise mutexes will be very slow.
License of Minetest-c55
-----------------------

View File

@ -19,6 +19,7 @@ with this program; if not, write to the Free Software Foundation, Inc.,
/*
=============================== NOTES ==============================
NOTE: Things starting with TODO are sometimes only suggestions.
NOTE: VBO cannot be turned on for fast-changing stuff because there
is an apparanet memory leak in irrlicht when using it (not sure)
@ -64,14 +65,13 @@ TODO: - Scripting
SUGGESTION: Modify client to calculate single changes asynchronously
TODO: Moving players more smoothly. Calculate moving animation from
data sent by server.
TODO: Moving players more smoothly. Calculate moving animation
in a way that doesn't make the player jump to the right place
immediately when the server sends a new position
TODO: There are some lighting-related todos and fixmes in
ServerMap::emergeBlock
TODO: Make a dirt node and use it under water
FIXME: When a new sector is generated, it may change the ground level
of it's and it's neighbors border that two blocks that are
above and below each other and that are generated before and
@ -81,55 +81,32 @@ SUGGESTION: Use same technique for sector heightmaps as what we're
using for UnlimitedHeightmap? (getting all neighbors
when generating)
TODO: Set server to automatically find a good spawning place in some
SUGG: Set server to automatically find a good spawning place in some
place where there is water and land.
- Map to have a getWalkableNear(p)
- Is this a good idea? It's part of the game to find a good place.
TODO: Transfer more blocks in a single packet
SUGG: A blockdata combiner class, to which blocks are added and at
destruction it sends all the stuff in as few packets as possible.
TODO: If player is on ground, mainly fetch ground-level blocks
TODO: Fetch stuff mainly from the viewing direction
SUGG: If player is on ground, mainly fetch ground-level blocks
SUGG: Fetch stuff mainly from the viewing direction
TODO: Expose Connection's seqnums and ACKs to server and client.
SUGG: Expose Connection's seqnums and ACKs to server and client.
- This enables saving many packets and making a faster connection
- This also enables server to check if client has received the
most recent block sent, for example.
TODO: Add a sane bandwidth throttling system to Connection
FIXME: There still are *some* tiny glitches in lighting as seen from
the client side. The server calculates them right but sometimes
they don't get transferred properly.
- Server probably checks that a block is not sent, then continues
to sending it, then the emerge thread marks it as unsent and then
the sender sends the block as it was before emerging?
TODO: How about adding a "revision" field to MapBlocks?
SUGG: More fine-grained control of client's dumping of blocks from
memory
- ...What does this mean in the first place?
TODO: Somehow prioritize the sending of blocks and combine the block
send queue lengths
- Take two blocks to be sent next from each client and assign
a priority value to them
- Priority is the same as distance from player
- Take the highest priority ones and send them. Send as many as
fits in the global send queue maximum length (sum of lengths
of client queues)
TODO: Make the amount of blocks sending to client and the total
amount of blocks dynamically limited. Transferring blocks is the
main network eater of this system, so it is the one that has
to be throttled so that RTTs stay low.
FIXME: There is a bug that sometimes the EmergeThread bumps to
the client's emerge counter being already 0, and also the
sending queue size of the client can float to 1 or 2, which
stops the map from loading at all.
- A quick hack could be applied to ignore the error of
being at 0 and timing out old entries
SUGG: Make client send GOTBLOCKS before updating meshes
TODO: Server to load starting inventory from disk
@ -137,17 +114,7 @@ TODO: PLayers to only be hidden when the client quits.
TODO: - Players to be saved on disk, with inventory
TODO: Players to be saved as text in map/players/<name>
SUGGESTION: A map editing mode (similar to dedicated server mode)
TODO: Maybe: Create a face calculation queue on the client that is
processed in a separate thread
TODO: Make client's mesh updates to happen in a thread similar to
server's EmergeThread.
- This is not really needed, mesh update is really fast
- Instead, the lighting update can be slow
- So, this todo is not really a todo. It is a not-todo.
SUGG: Make server to send all modified blocks after a node change
after all the stuff including lighting have been updated
SUGG: A map editing mode (similar to dedicated server mode)
TODO: Make fetching sector's blocks more efficient when rendering
sectors that have very large amounts of blocks (on client)
@ -169,16 +136,16 @@ SUGG: Add a time value to the param of footstepped grass and check it
against a global timer when a block is accessed, to make old
steps fade away.
TODO: Add config parameters for server's sending and generating distance
TODO: Copy the text of the last picked sign to inventory in creative
mode
TODO: Untie client network operations from framerate
- Needs some input queues or something
SUGG: Make a copy of close-range environment on client for showing
on screen, with minimal mutexes to slow the main loop down
on screen, with minimal mutexes to slow down the main loop
SUGG: Make a PACKET_COMBINED which contains many subpackets. Utilize
it by sending more stuff in a single packet.
@ -193,8 +160,8 @@ SUGG: Split MapBlockObject serialization to to-client and to-disk
TODO: Get rid of GotSplitPacketException
Before release:
======================================================================
SUGG: Implement lighting using VoxelManipulator
- Would it be significantly faster?
TODO: Check what goes wrong with caching map to disk (Kray)
@ -203,7 +170,6 @@ TODO: Remove LazyMeshUpdater. It is not used as supposed.
Doing now:
======================================================================
TODO: Implement lighting using VoxelManipulator
======================================================================