Sorry, I was a bit bummed out when I last posted.
It's not unpacking per se, in that there's additional information extracted. The AI script is read into memory in a specific way, and I suspect that changing this would be quite a bit of effort.
In general (as I'm sure people know), the AI data is organized into two blocks - the nonreact block and the reaction block. Within each block is a set of zero to ten condition/action (cnd-act) sets. The game loads the AI script into memory in the following way: First, the non-react block is loaded. The four-byte conditions are stored at 4759,X (17 bytes/cnd). The corresponding actions are stored at 4803,X (64 bytes/act). X in this case is the index for each cnd-act set, starting at zero and incremented for each set, up to ten. Next, the react block is loaded in a similar way to 4A83/4B2D.
Without going through each AI script and looking it over, I couldn't say if these offsets can be compressed at all. I am particularly suspicious of the condition block saving - why does the game need 17 bytes to hold 4 bytes? I'm a little skeptical that the action block really needs 64 bytes as well. On first glance, I can't see any code that saves additional data to those structures. Perhaps this is simply a relic of changing code, where there was more to the condition block and it got pared down to the four byte structure without getting the memory allocation changed. It would be a big undertaking to check all of this, and while I might work on it as a side project, for now I'm willing to simply let it be and find other, smaller regions of free RAM to use.
As an aside, I believe I found a rare bug in the "If Target Status Set" condition. At C2/287A, as part of checking the Dead status (201A bit 7) it also checks the target's HP. However, instead of an ORA for both bytes of the word as is normally done, it ORAs the low bytes together. Oops. This results in a case where targets that have a HP multiple of #100 will be counted as dead for this check, even if they are not. It's a one byte fix.
Topic: Quick Confirmation Question Re: AI Scripts (Read 1541 times)