Graphical text, BASIC tokeniser and flood-filling

Tuesday, 29th July 2008

I've got a fairly hackish "graphical text" mode set up (enabled with VDU 5, disabled with VDU 4) that causes all text that is sent to the console to be drawn using the current graphics mode (at the graphics cursor position, using the graphics colour and logical plotting mode and graphics viewport). This allows text to be drawn at any position on-screen, but is (understandably) a bit slower and doesn't let you do some of the things you may be used to (such as scrolling text, copy-key editing and the like).

I've also done some work on a tool to convert files from the PC to use in BBC BASIC. It takes the form of a Notepad-like text editor:

BBC BASIC programs are stored in a tokenised format (usually .bbc files on a PC) and need to be wrapped into a .8xp for transferring to the calculator. The editor above can open .8xp, .bbc and .txt directly, and will save to .8xp.

The detokeniser can be passed a number of settings, which can be used to (for example) generate HTML output, like this. The indentation is generated by the detokeniser (leading/trailing whitespace is stripped by the tokeniser). The tool can also be used to directly convert binaries into .8xp files if need be.

I've been doing a little work on a flood-filling algorithm. (PLOT 128-135, 136-143). The above images show its progress; on the left is the first version (which can only fill in black). There is a hole in the bottom-left of the shape, so the leaking is intentional. It also stops one pixel away from the screen boundary -- this too is intentional (it clips against the viewport). The second version, in the middle, plugs the leak and applies a pattern (which will be a dither pattern in BBC BASIC) to the filled area. On the right is the third version, which will fill over black or white pixels with a pattern.

The main filling algorithm needs a 764 byte buffer for the node queue and three 16-bit pointer variables to manage the queue. I've rounded the queue size up to 768 bytes, so it fits neatly on one of the RAM areas designed to store a bitmap of the display.

The problem is filling with a pattern. The way I currently do this is to back up the current screen image to a second 768-byte buffer, fill in black as normal, then compare the two buffers to work out which bits have been filled and use those as a mask to overlay the dither pattern. This is quite a lot of RAM, just to flood-fill an image!

For those who are interested, I'm using the "practical" implementation of a flood fill algorithm from Wikipedia.