Release 0.0.8C

Got a quick chunk of time to play around with this again.  The minimap is now working properly, and that memory fix I talked about is now in.  The tileset may also have been tweaked, I’m not sure on that one.  Grab it here!

Memory, Python, and You.

Rather recently I learned about a handy little program called Valgrind. I get to use it to debug some C++ code that works with dynamic memory to fix any leaks that occurred. As it turns out, this program can do a lot more.

Something I’ve noticed for a while now is that Graff takes up a pretty significant amount of memory–around 40 megabytes of RAM. I never really thought about it too hard until now, but I discovered that with the command:

valgrind --tool=massif python Start.py
...
ms_print massif.out.** > output

it takes snapshots of Graff’s memory usage. The first time running this and opening up “output” gives this result:

mem1

So currently I get ~43MB usage of the heap (or in python terms, plain memory) by just opening up the program and running around. Later in that output file, I notice a lot of calls to “SDL_CreateRGBSurface (in /usr/lib/libSDL-1.2.so.0.11.2)” seem to be eating up the most memory percentage-wise.

Doing some digging, I find out pygame converts images into a .RAW format for fast blitting–unfortunately, the file size is pretty hefty. Looking through my code, I notice that within Entity.py, when an entity is initialized, it loads up a file called “tiles.png” to pull the entity’s picture out of that and use it. After that, that loaded image is just kept there, permanently until the program quits. Tiles.png is only 42.3KB, though, which seems a little weird.

I decide to see just how many times entity loads up this image and stores it away forever. It turns out that this file is being loaded around 200 times, and is then saved as a member variable “self.sheet”. Since we don’t need it anymore once the entity’s image is ripped out of that sprite sheet, I decide to add one extra line to my code:

delete self.sheet

, after the entity’s image is set. What does that memory chart look like now?

mem2

15.64 MB is now used. By adding one extra line of code to delete an unused part of the code, I cut memory usage by 2/3!

Tinkering around some more, I find the pygame init routines take up a total of about 5MB, which means the code portion that I wrote takes up about 10MB of RAM. Perhaps now I’ll start paying attention to where memory ends up in Graff.

Map generation #4.5 – tidying up some ends

In my last post about this map-making process, I seemed to have left off at the part where the image itself is generated in its own separate program. Converting this image into a matrix for use in Graff was relatively straightforward. Pygame’s surfarray module allows one to access a surface (the screen) and treat it as a matrix of color values. From there I pulled out the color, used a dict datatype to convert that color into a tile type, and then write that into a matrix for use with the Graff; that wasn’t much of a problem.

However, one glitch I’ve noticed is that often the map would have some holes in it–it seemed like the algorithm would skip over a tile, perhaps because it couldn’t find a properly-fitting tile, so sometimes you get things that looked like this:

glitchy tiles

As it turns out, that was the case. I simply did not make enough tiles to cover each possible combination. After thinking that I’d have to make a mind-boggling number of additions to the 6×6 tile pieces I already have, I edited the code so that if the tile it needed could not be found, it should spit out that key combination and then I’d see where that would take me. Altogether, only about 30 “impossible” combinations of tiles ever seemed to come up, so I just added those extra ones onto the tileset.

After doing that, the results are much cleaner. Some glitches still remain (mainly I missed a few combinations, it seems), and I’ll probably adapt this to work with a smaller scale (say 3×3 instead of 6×6) or make changes to it to develop other areas, but I think I’ve just about exhausted the usefulness of posting about this process.

Still a little glitchy, but better!

Oh, there is also the problem of the map not always being one complete piece. I still haven’t decided in what way I’ll solve that one yet, but there are some more interesting questions about Graff that I think I need to find an answer to first.

0.0.8

Release here!

This’ll probably be the last update for a few weeks. Trying to go back to Python after spending a few weeks of working with C++ feels so..different. I might switch this project over to C++ if I get equated well enough with it, or maybe I’ll just start working on a C++ project instead and keep this around too. Busy busy busy. Either way, more activity will come in December or so.

The only glitch I’ve noticed with this release is the minimap is a little screwy. I know why it doesn’t work, but if I hold off on this until I fix it, it’ll be even longer before this post comes up.

CHANGES:
   Added a (barebones) title screen.
   Custom keybindings via title screen.
   Added “Key” py file and an ini file for manual editing if need be.
   Digging function cleaned up somewhat.
   Integrated new map generator.
      A fully glitch-free version is finished, I just need to update it.
   Changed scrolling/grass behaviors for more natural movement.
   Automatic screenshots placed in the Screenshots directory. Use the SCROLL LOCK key.

Additionally on that page under Misc, I’ve added in a dialog.py / dialog.txt file to show how my dialogue system will hopefully work. Run it through the terminal

Brief update.

Got some feedback on how the player movement worked and so I decided to switch it to the original idea I had–when the player gets too close to the bounds of the screen, then the screen recenters itself on the player, as opposed to re-centering every single time the player moves. Likewise the grass is now drawn so that it looks “attached” to the tile, instead of before where the grass appeared to just float under the screen like an odd parallax effect (see image; night sky is what the grass looked like if anyone hasnt tried out this thing).

Also, I’ve worked in the map generator into the game itself. It looks a bit weird, to be sure:
ingame map generator
But I feel like currently that’s due to tileset issues and not necessarily the code. Don’t mind the odd looking minimap, still have a few kinks to work out.

Also, now I’ve got a way to output screenshots (properly labelled to boot) into a ‘screenshots’ directory when you press the printscreen key. Saves time when I’m on Windows, thats for sure.

Map generation #4

#4? Where’d #3 go? Post #3 would be useless because, from where I finished at #2, I’ve completely rebuilt the code in a separate file. Refactoring, joy.

Last time I posted, I had an algorithm that was reasonably good at generating a network of hallways. I don’t think I have any screenshots, but with some edits to the tileset, you can get some really wacky maps generated out of it, one of which could work as a ruined set of roads that I might save later, though I don’t have a picture of it. I’ll probably stash away that code file for later.

Put simply, most of the previous code followed this basic pattern:
make a matrix, and randomly assign an index to it, with the index corresponding to a certain image of a predef.
afterwards, go through and check each matrix cell for connections between predefs that should be there, and connect them.

The important part about it was that each index was a binary number from 0001 -> 1111. When you introduce in rooms, however, that scheme needs to be so modified that my “randomly do it and fix it later” approach fell to the ground. Not only do you have connections for 0 and 1 (nothing and hallway), you’ve got 2, 3, and 4 (room lower, room upper, room filled), and so the numbers just get rather large when trying to do it the old way. What I’ve got now, looks like this:

make a matrix of zeros
set the corner tiles to something that does not leave the grid and has 2 connections going inside
for each tile in the top row:
  pick a tile whose connection type (0,1,2,3,4) on the left of the tile matches that of the left tile's right side
for each tile in the left column:
  [repeat, checking upper tile instead of left tile]
for each tile inside the bounds of the matrix:
  [again repeat, checking upper and left tile's connections so they match up]
for each tile on the bottom row:
  [repeat, checking left tile, upper tile]

and so on, until the matrix has been filled, then draw the appropriate tiles to the screen in the right positions. So, instead of doing it all and checking it later, we’re immediately putting in a tile that is known to fit the previously known connections. What really delayed me was that the corners, top, left, right, bottom, and inside all check for different things–the top tiles have to set themselves to be a good tile who’s top is a 0 and left matches the adjacent tile; the inside tiles have to check the left tile and the top tile; the bottom row has to have a bottom of 0, match the top, and match the left tile, and so forth. As a result there wasn’t a very easy way for me to take all four of those action and combine them into one thing.

What’s the result now? Here are some shots of the same code, but with edits to the tileset.png part:

Option 1: Plain tiles
plain map
Option 2: Walls
wall map generation
Option 3: Tile edit 1
curvy map generation
Option 4: Tile edit 2
crumbling map generation

There’s a few glitches in part because sometimes, it’s impossible to connect the right tile–that combination just can’t exist, so you get a few gaps there when defining say, the walls in that second picture in the bottom right–optimally all the white should be outlined by red. Eventually I might fix that, either by adding in those “impossible tiles” into the set or something else, though that’ll be saved for later. I’ll also have to re-check the code once more to see if there isnt just a simple glitch that’s getting in the way, too.

Regardless, the results themselves look nice to me. I have coded in place a way to (somewhat painstakingly) add in additional tiles. Compared to the generator that gave me the idea, if you select only Dungeon 1 tiles, the results are pretty close. If I want to change the distribution and say, add more hallways than rooms to a map, I could always add in additional hallway predefs, so that the random.choice will be more inclined to pick hallway connections than room connections. Likewise, the pictures above show how altering a few of the ~40 or so tiles can make the map look significantly different.

Where will this little thing go next? I probably won’t bother spending a post writing about it, but pygame has a surfarray subsection to the module that lets me convert pixels on an image into a numpy matrix. From there it’s rather trivial to integrate what I’ve got into the standard Map.py set of code I’ve got and get it playable.

Code of this and the dialogue system will be up on my other site..eventually.

Map generation #2

Alright, so where were we? Right, we had that ugly looking “almost looks like a somewhat broken maze” kind of thing, that looked like this:
map attempt 1

And there were a few issues with it. The first is that we had a lot of non-connected tiles. My first gut instinct was to just remove the dead ends at the tilesheet image–which, I may not have mentioned, is just a 90×6 image, and every 6×6 ‘box’ of pixels represents a particular connection–ends, T junctions, etc. The first four of these are the dead-end connections. To remove them, I first tried doing just that: taking the T junctions and just copy/pasting over the dead ends with them, I get this:
covering up connections

Better, but it still looks a little off. The problem now is that we’ve got ‘islands’ of connected segments, and it looks a little weird. After some staring blankly at the screen, the I decided to fix this I’d have to actually work with code, specifically, this code:


def checkAdjacent(array,row,col):
  New = 0
  current = array[row,col] #current type is a numpy.int32, not plain old python int
  current = '%04d' % int(bin(current)[2:]) #net result is a 4 digit binary string
  #pull out the 4 digit binary string from each adjacent tile
  try: top = '%04d' % int(bin(array[row-6,col])[2:])
  except IndexError: top = None

  #check if adjacent tile 'wants' to connect
  if str(left)[2]=='1' or current[0] == '1':New+=8#;print "catch right"
  if str(top)[3]=='1' or current[0] == '1':New+=4#;print "catch bottom"

return New

Sooo, what does this all do? Put simply, it scans through the matrix and changes predefs so that they appear joined by changing one of the 0's in the four bit long number to a 1 to match the adjacent predef. To explain further:

First, I'll note that I added some redundant code that could (should?) have gone into a function and made this thing about 4 times smaller; I just left out the redundant parts where it checks the other 3 directions for both cases.

The first bit of code is where the function takes the current cell in an array (defined as array[row,col]), and using the line '%04d' % int(bin(number)[2:]) converts a "numpy int" (NOT a regular int, oddly) into a binary number that is then sliced and padded and converted into a string all at once and formatted. Long story short, it takes a number like 6, and turns it into a string of "0110". Then, we do that for the cell found to the left, right, bottom, and top of the current cell (top is the only operation shown). Try/Except syntax lets me fudge it so that if I try to get a bad index it just doesn't go through, instead of quitting with an IndexError.

Once we do that, we enter the second part of the function, where what happens is that we are effectively OR'ing the value of "current" in binary with the corresponding digit of each adjacent tile, and that result is fed into the int New by adding binary's significant figures accordingly (8=1000, 4=0100, etc). This is where the cells are actually coordinated into being joined. Outside of the function, the value of 'current' is then replaced with that of New when the function is called.

Now we simply call that function on every cell in the matrix, and what happens is this:
fixed connections

...well that's a little different. Something I didn't mention was that along the way is that I've been fudging the dimensions a little, now I'm up to 6x6 pixel sets of predefs and everything is bigger. Regardless, I'm closer to making a reasonable'ish network of hallways, but now the opposite problem is happening than before: the map is now just a plain old grid and is too connected, and that's not interesting. So how about we add a condition that predefs are joined only, say, 50% of the time? Well we do that, and end up with:

6

It looks a little better. There seem to be definite sections of hallways, and I can almost see where rooms could be set via those grids. Looking back at that online dungeon map generator, I've noticed 2 distinct types of connections: hallway connections, and room connections. I'm not sure where I'm going with this at this point, but if I make more headway with this and find a way to reasonably explain it, I'll post next about getting rooms in here.