Uses fmt in recollide_reanchor. This has a performance impact in code that is already slow so this might need to be reverted; I think that should be part of a comprehensive optimization pass, however, and making it worse for now as part of a general code cleanup is probably better.
this could be golfed further by rewriting the str..= line in terms of fmt, but the performance impact of inserting the extra `fmt` calls and parsing into the tostring recursion seems likely to be a problem.
oops, need to special-case the first part of the split list.
amusingly, I don't need to special-case zero-length format strings because that will skip the entire loop and output "", which seems correct. (I am also not special-casing zero-length segments because `s[1] == nil` will go into the error handler and that seems fine.
I discovered your tostring function for debugging, might as well make it reachable from fmt. I also realized I forgot to convert to using "p" when I introduced it so I fixed that
the "split on %" strategy makes parsing "%%" complicated, so the "actually literally %" placeholder is now "%~" rather than "%%". since thsi resembles a real sprintf even less now I renamed it to "fmt".
also actually closes the `if m == "~"` block. (which was `if m == "%"` before this patch)
this spends extra tokens to handle "invalid format character" to catch using a % where %% is intended. this is designed to grow with further format chars if needed; I have ideas for not-exactly-POSIX %d, %x, %h, %!, %#, %c, and possibly others but there is absolutely no reason to spend tokens on these things until we need them. (that said, existing debugging output might benefit from some of these other formats, but that debug code is commented out, so maybe nevert.)
Debug mouse is now its own module, so it can be separated from the hint system, since it is useful for more than just positioning hints. It now has the following enhancements:
1. Clock-based table cyclng now has a helper function (cycle)
2. Debug mouse color cycling is distinct from hint color cycling, so debug position readout remains legible
3. Debug position readout now stays on screen even when the cursor is near or past the edges
4. Debug cursor cycles between a mouse sprite specifically marking the exact pixel that is being sampled, an "X" for text character sizing, and a "□" for positioning the centered 3x3 characters often used as hint target markers
5. Map cell coordinates (in square brackets) are displayed in addition to pixel coordnates (in parentheses)
Sprite 50 is now the mouse cursor. Color 15 is color cycling for debug readouts.
Debug mouse features can be disabled by commenting out `add(real_modules, debugmouse)`.
I've done a little bit of golfing but this is stiill a token expense. I'm going to write a crappy sprintf function to save tokens everywhere we're assembling strings from their component parts.
Test hints now both display on level 0
Allows testing progressive display of more hints.
Code review feedback, debug mode
Reviewed-on: pyrex/chameleonic#18
Co-authored-by: Kistaro Windrider <kistaro@gmail.com>
Co-committed-by: Kistaro Windrider <kistaro@gmail.com>
Tileset fuckery
First servicable
Better, black BG
Redo crate graphics
Make debug wall a less confusing color
Redo first few levels
Do another level!
Tile the whole intro
Merge branch 'main' into tiles (bad)
Fix bad merge
Co-authored-by: Nyeogmi <economicsbat@gmail.com>
Reviewed-on: pyrex/chameleonic#16
I feel like there's a way to build on this theme. I think this is the minimal version of the puzzle; I'd be glad to remove elements if it turns out I can.
previous attempt tried to use shifts to get 0.4 to 0.2 or 0.8 and multiplication to get -0.8, forgetting that the multiplication would also turn 0.4 into 0 along the way. oops. I got ratholed on the mathematical approach, but a very straightforward lookup table can get more done.
Reviewed-on: pyrex/chameleonic#14
Co-authored-by: Kistaro Windrider <kistaro@gmail.com>
Co-committed-by: Kistaro Windrider <kistaro@gmail.com>
replace comparisons with bit math bullshit
integers in the range [0, 15] fit entirely in the bit mask 0x000F. integers out of that range will have at least one bit 0x0010 or higher, or will have the sign bit 0x8000 set. so to find out if one of two numbers is out of range [0, 15], we can check the bit mask of their bitwise or.
this saves tokens and cycles. it is also completely illegible. very in the spirit of Pico-8, I love it.
comment the bullshit
it needs it
packed crate representation
don't bother exploding crates into four bools, and then comparing them all individually to a bunch of conditions. absurd bit manipulation bullshit saves cycles and tokens. leaving a crate's movement rule represented as four bits means we can exploit our previous calculation of dx1 and dy1, which must each either be 0x0001 or 0x8FFF, and violently hammer them down to align with this bit-packed representation, giving this glorious little atrocity.
Fix crate math.
I forgot that -1 & 1 = 1 rather than 0 so all the bit math didn't work. But I can fix it with polynomial algebra! this is much better.
Save tokens on movemebnt checks
I promise this is mathematically equivalent-ish to the original. (0.2 and its multiples are nonterminating decimals in base 2, so there's a little jank when the negative shift right is a shift left.)
Trimming
Trim up redundant nil checks, sequential assignments that could be on a shared line, and repeated references to a deeply nested variable.
Reviewed-on: pyrex/chameleonic#13
Co-authored-by: Kistaro Windrider <kistaro@gmail.com>
Co-committed-by: Kistaro Windrider <kistaro@gmail.com>
kistaro
pyrex