Somehow this was only causing issues (afaik) with
`draw-decoration-load-save` getting corrupted, perhaps because other
processes either use the shared dram stack or the gigantic spr stack.
Slight change to float divide operations (again). Now it only turns into
inverse multiplication if the float is a power of 2 (positive or
negative). Non-zero immediate divisors will be compiled as regular float
divisions but will forgo the extra branches and checks for divide by
zero.
Also fixes#2584
Allows for removing a line in the subtitle editor finally, also fixed
the following:
- The path the editor used didn't include the game name
- Loading the subtitle project is too slow to do on startup now
(4-5seconds in a debug build), do it on demand now
Updates the decompiler for the new format and there's new macros. This
new format should be easier to read/parse.
Also rewrote `sp-init-fields!` (both jak 1 and 2) from assembly to GOAL.
Hopefully I did not miss any regressions in Jak 1/2 while updating the
files, it's a lot.
- [x] compare NTSC-K
- [x] compare NTSC-J
- [x] compare PAL
- [x] figure out version order
- [x] ~~write delta patch for spanish text~~ no need for now
Fixes#2530
Fixes issue where warp effect looks wrong near the edge of the screen -
there was an unhandled `REGION_CLAMP` texture setting.
Fixes a potential bug where "warp page" things wouldn't be drawn at all
because there is no PC warp bucket. Unclear if anything actually fits
this category, but it doesn't hurt.
Turn on PC-format texture uploads for the water page so the precursor
guy uses the right texture. It has to use generic because it abuses the
generic death query thing to spawn particles.
Workaround for some issues with rebuilding level files after changing
engine files. Not a perfect solution, but probably good enough.
Doesn't actually do anything in Jak 2 because the collide mesh isn't
extracted, but the functionality is all there. Also updated the renderer
a bit to keep the colors more readable.
Fixes crashes when killing the big spider, killing predator metal heads,
and watching the end cutscene. The presursor guy is the wrong color, but
I don't have time to look into this today.
Fixes citizens that disappear (permanently) when being pushed into a
"wall" which is the border of a nav mesh that is exactly aligned with
coordinate axes.
This bug feels very similar to punch glitch, where code worked on PS2
only because some value didn't reach exactly 1.0 or 0.0 with ps2-style
floating rounding. It might be better to track down the source of the
"only 1.0 on PC" value, rather than patching downstream code to handle
it, but I can't easily find it in this case - there's a lot of code that
touches this heading value. (it's also possible this bug happens on ps2,
but the result is the guard appears to face the wrong direction, rather
than disappearing).
Detailed debug notes:
Ran a process that just calls this function:
```
(defun print-guard-info ()
(let ((proc (process-by-name "crimson-guard-level-42" *active-pool*)))
(when proc
(format *stdcon* "PROC: ~A~%" (-> proc state))
(let* ((pd (the process-drawable proc))
(css (-> pd node-list))
(cs (-> css data 15))
)
(format *stdcon* "joint: ~A~%" (-> cs joint))
(format *stdcon* "pos: ~`vector`P~%" (-> cs bone position))
)
)
)
)
```
it prints out the state, and the bone position for some bone that
happens to be on the upper body. It goes to NaN when the upper half
disappears, in the state `tazer-hostile`.
Modified the code in this state to call this function in a bunch of
places:
```
(defun guard-nan-debug ((guard process-drawable) (info string))
(when (string= (-> guard name) "crimson-guard-level-42")
(format 0 "[guard-nan] ~S : ~`vector`P~%" info (-> guard node-list data 15 bone position))
)
)
```
which prints the bone position to stdout.
This shows that the problem happens after `post`, but before `trans`:
```
[guard-nan] post-end : #<vector 2723306.0000 269921.0312 388825.2187 1.0000 @ #x1f1350>
[guard-nan] trans-start : #<vector NaN NaN NaN NaN @ #x1f1350>
```
this is probably as part of the bone math.
To check, I added some prints to `execute-math-engine`, before and after
the call to `do-joint-math`:
```
[guard-nan-math] pre-math : #<vector 2722236.5000 268609.5312 385339.9062 1.0000 @ #x1f1350>
[guard-nan-math] post-math : #<vector NaN NaN NaN NaN @ #x1f1350>
```
The first part of `do-joint-math` is calling the
`generate-frame-function`, which does animation blending to compute a
bunch of joint transforms. I dumped these:
```
(let ((jaf (the-as joint-anim-frame (+ 2400 (scratchpad-object int)))))
(format 0 "generate frame:~%")
(format 0 "~`matrix`I~%" (-> jaf matrices 0))
(format 0 "~`matrix`I~%" (-> jaf matrices 1))
(dotimes (i (-> obj mgeo num-joints))
(format 0 "~`transformq`P~%" (-> jaf data i))
)
)
```
and confirmed that they always look good.
The next part is "prebind", which allows something to modify the
`joint-anim-frame`. Nothing does this on the guard.
The next part is what I've named the "root bind", which computes the
transform of the root bone based on the process-drawable root's
position.
```
(when (the-as (function cspace transformq none) t9-3)
(when *djm-debug* (format 0 "djm: first bind func ptrs #x~X #x~X~%" cspace<-transformq! t9-3))
(when *djm-debug* (format 0 "djm: first bind func input: ~`transformq`P~%" (-> v1-20 param1)))
((the-as (function object object object none) t9-3) v1-20 (-> v1-20 param1) (-> v1-20 param2))
(when *djm-debug* (format 0 "djm: first bind func result:~%~`matrix`I~%" (-> v1-20 bone transform)))
)
```
in this case, the first print confirms that we're using
`cspace<-transformq!` as the root binding function.
When the guard entirely disappears, it is caused by the root of the
process drawable having NaNs as its `quat`:
```
djm: first bind func input: #<transformq @ #x1f04f0
trans:2724817.7500 262144.0000 388891.1875 1.0000
quat: NaN NaN NaN NaN
scale: 1.0000 1.0000 1.0000 1.0000>
djm: first bind func result:
[001f0e70] matrix
[ NaN] [ NaN] [ NaN] [ NaN]
[ NaN] [ NaN] [ NaN] [ NaN]
[ NaN] [ NaN] [ NaN] [ NaN]
[2724817.7500] [ 262144.0000] [ 388891.1875] [ 1.0000]
```
After this, it's possible to get the lower half of the guard to return,
but I believe the real problem is this first `quat` being NaN. Then the
nans probably spread everywhere.
So now we now to look for the `(-> guard root quat)` becoming NaN:
```
(defun guard-nan-debug ((guard process-drawable) (info string))
(when (string= (-> guard name) "crimson-guard-level-42")
(format 0 "[guard-nan] ~S : ~`vector`P~%" info (-> guard root quat))
)
)
```
and we see this happens in post somewhere
```
[guard-nan] post-start : #<vector 0.0000 -1.0000 0.0000 0.0000 @ #x1f0500>
[guard-nan] post-end : #<vector NaN NaN NaN NaN @ #x1f0500>
```
Looking at individual methods
```
[guard-nan] before-142 : #<vector 0.0000 -1.0000 0.0000 0.0000 @ #x1f0500>
heading: #<vector 0.0000 0.0000 -1.0000 0.0000 @ #x3432670>
[guard-nan] before-143 : #<vector NaN NaN NaN NaN @ #x1f0500>
heading: #<vector 0.0000 0.0000 -1.0000 0.0000 @ #x3432670>
```
It looks like the problem is converting this heading to a quaternion.
Adds a decent way to customize the folders the project file expects the
iso data and decompiler data to be in. When you run any version other
than the default, for example Jak 1 PAL, it uses the `gameName`
decompiler config to consume and output it's results.
However the project file will assume `jak1` unless you hard-code it
differently -- basically, it needs to be explicitly told just the
decompiler is told what version to use.
We now have a per-user REPL Config json file, so that can be used to
override the default `jak1` behaviour.
Fixes#1993
