<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
<title>hwtr</title>
<subtitle>Reverse engineering Hot Wheels Turbo Racing for the PlayStation (Stormfront Studios / Electronic Arts, 1999, SLUS-00964) and porting it natively to the PC, using tools written for the job in Rust. No game data included.</subtitle>
<id>/</id>
<link rel="self" href="/feed.xml"/>
<link rel="alternate" href="/"/>
<link rel="alternate" type="application/json" href="/log.json"/>
<updated>2026-10-05T00:00:00Z</updated>
<generator>cairns 0.9.0</generator>
<entry>
<title>Pickups taken and back, checked; Random&#39;s draw order fixed</title>
<id>/90-pickups-taken-and-back-checked-random-s-draw-order-fixed</id>
<link rel="alternate" href="/90-pickups-taken-and-back-checked-random-s-draw-order-fixed"/>
<updated>2026-10-05T00:00:00Z</updated>
<category term="race"/>
<summary>Taking a pickup and the pickups coming back had no check against the original.</summary>
<content type="html">&lt;p&gt;Taking a pickup and the pickups coming back had no check against the original. racecheck skips any step where a power-up is taken. Building the check turned up one real bug.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The bug: &quot;Random&quot; drew from the wrong order.&lt;/strong&gt; 0x800671a8 loads each power-up the track&#39;s pickups name into a list at 0x800d2698. The list&#39;s add (0x80061c00) links each new one at the head, so the original holds them newest first. A &quot;Random&quot; pickup draws the k-th of that list (0x80067ac4). The app pushed them oldest first, so with the same seed a Random pickup gave a different power-up from the original&#39;s on any track with two or more power-ups. The load is now &lt;code&gt;hwtr_game::powerup::load_defs&lt;/code&gt;, newest first, and the app uses it.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The codec.&lt;/strong&gt; hwtr-hle&#39;s &lt;code&gt;original::powerup&lt;/code&gt; reads the original&#39;s state into a &lt;code&gt;PowerUps&lt;/code&gt;:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;the power-up list&lt;/li&gt;
&lt;li&gt;the pickups (24 bytes each: name, number, power-up number, when taken, out, collision object)&lt;/li&gt;
&lt;li&gt;each car&#39;s held list from 0x8012ff04 (nodes of item, previous, next and count; entries of 12 bytes)&lt;/li&gt;
&lt;li&gt;the clock 0x800d0e34&lt;/li&gt;
&lt;li&gt;the unlock record&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Tests (hle tests/powerup.rs):&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;power_ups_load_in_the_originals_order&lt;/code&gt; runs on all 11 tracks. It fails on the old order.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;pickups_are_taken_and_come_back_as_in_the_original&lt;/code&gt; runs 150 rounds per track. In each, a random car takes a random pickup, put out or not, through 0x80067f98. Then the clock moves up to 6 s on and 0x80068130 steps. It compares every car, the pickups, the held lists, the unlock record and the seed. Over the run that is 1443 taken, 897 run out, 965 back and 140 Random draws. The port&#39;s own &lt;code&gt;turbo_given&lt;/code&gt; event has nothing in memory, so it is cleared before comparing, as the apply test already did.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The take&#39;s other helpers, 0x80067c3c and 0x80067c98, never matter after the drop-all (0x80067d50) has emptied the list. The test confirms it.&lt;/p&gt;
&lt;p&gt;power-ups.md now covers the load order, the records and the tests.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The unlock pickups checked, and a page for the power-ups</title>
<id>/89-the-unlock-pickups-checked-and-a-page-for-the-power-ups</id>
<link rel="alternate" href="/89-the-unlock-pickups-checked-and-a-page-for-the-power-ups"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="race"/>
<summary>The power-ups have been ported since the front-end batch (entries 28 to 44), but the docs never had a page for them.</summary>
<content type="html">&lt;p&gt;The power-ups have been ported since the front-end batch (entries 28 to 44), but the docs never had a page for them. They also had no check on the cars the unlock pickups note.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The unlock record.&lt;/strong&gt; UNCAR1 and UNCAR2 carry flags 1 &amp;lt;&amp;lt; 24 and 1 &amp;lt;&amp;lt; 25. In 0x80065520, a player&#39;s car (flag 1) in slot 0 or 1 copies the track&#39;s first or second car (0x800d0ea6, 0x800d0ea8) into the record. The record is laid out by car, then by player: 0x800d0eaa and 0x800d0eac hold the first car for players 1 and 2, 0x800d0eae and 0x800d0eb0 the second. The port keeps it by player (&lt;code&gt;PowerUps::unlocked[player][car]&lt;/code&gt;), and the app&#39;s &lt;code&gt;Race::result&lt;/code&gt; packs it per player.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt; &lt;code&gt;the_unlock_pickups_note_cars_as_the_original_does&lt;/code&gt; (hle tests/powerup.rs) takes each pickup with the car in slot 0, 1 and 2. Each car is made a player&#39;s under full physics, since only such a car takes a power-up. The test compares the record. My first version left slots 1 and 2 as computer cars, so it noted only slot 0. Both sides agreed there, but player two&#39;s slot was never exercised. Forcing full physics fixed that, and the test now requires four notes. To call it, &lt;code&gt;PowerUps::apply&lt;/code&gt; is now public.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;docs/engine/power-ups.md&lt;/strong&gt; is new. It covers the &lt;code&gt;.PUP&lt;/code&gt; layout (factors, flags, kick), taking a pickup, &quot;Random&quot; (never the unlocks), the frame&#39;s expiry, the unlock record and the tests.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The controls into the car, checked for any stick</title>
<id>/88-the-controls-into-the-car-checked-for-any-stick</id>
<link rel="alternate" href="/88-the-controls-into-the-car-checked-for-any-stick"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="input"/>
<summary>Since entry 17 one question had stayed open: does an analog stick reach the car the way a digital pad does? Half of it was already checked: reads_match_the_original (tests/pad.rs) covers the pad read for both kinds.</summary>
<content type="html">&lt;p&gt;Since entry 17 one question had stayed open: does an analog stick reach the car the way a digital pad does? Half of it was already checked: &lt;code&gt;reads_match_the_original&lt;/code&gt; (tests/pad.rs) covers the pad read for both kinds. The other half, the controls into the car (0x80034940), had no check. racecheck doesn&#39;t cover it because it copies the original&#39;s controls straight into the port&#39;s step.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The new test.&lt;/strong&gt; &lt;code&gt;controls_into_the_car_match_the_original&lt;/code&gt; (hle tests/car.rs) runs 400 rounds on each of three race states. Each round:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;writes random action levels into player one&#39;s reader: nothing, 255 as a button gives, or anything between as a stick gives&lt;/li&gt;
&lt;li&gt;sets random held buttons&lt;/li&gt;
&lt;li&gt;sets a random speed, and in one round in ten marks the race over&lt;/li&gt;
&lt;li&gt;calls 0x80034940(0)&lt;/li&gt;
&lt;li&gt;compares the whole car with &lt;code&gt;PadReader::controls&lt;/code&gt; followed by &lt;code&gt;Car::apply_controls&lt;/code&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;It passed on the first run. The test requires over 500 rounds that actually steer. A stick and a pad both reach the car as levels, so the stick works like the pad. controls.md records this.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Entry 38&#39;s unlock note is stale.&lt;/strong&gt; It said the race pickups that unlock cars weren&#39;t in the port, so a race&#39;s unlock words were always zero. They are in: &lt;code&gt;PowerUps::unlocked&lt;/code&gt; (0x80067708) records the cars, and the app&#39;s &lt;code&gt;Race::result&lt;/code&gt; packs them into the result&#39;s words for the front end. What stays open is that the unlock pickup has no hle check. That is carried here, along with 38&#39;s other questions.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Cheat 8 paints each car one colour</title>
<id>/87-cheat-8-paints-each-car-one-colour</id>
<link rel="alternate" href="/87-cheat-8-paints-each-car-one-colour"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="render"/>
<summary>Cheat 8 (code 34563456) makes the cars flat colours, not the track.</summary>
<content type="html">&lt;p&gt;Cheat 8 (code 34563456) makes the cars flat colours, not the track. I had it down in cheats.md as &quot;flat textures&quot; after one glance at the check in &lt;code&gt;tim_upload&lt;/code&gt; (0x80012a10). Following it through corrected that.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;When the fill happens.&lt;/strong&gt; &lt;code&gt;tim_upload&lt;/code&gt; fills the image with its first byte before LoadImage when all three hold:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;its fill is on (0x800d244c)&lt;/li&gt;
&lt;li&gt;cheat 8 is set&lt;/li&gt;
&lt;li&gt;the caller&#39;s last argument (keep) is 0&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Which uploads it hits.&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;The fill flag.&lt;/strong&gt; 0x800127b4 turns it on at boot. It drops only around SFX.GLM (0x80012864) and one sprite upload (0x8001dbd8).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The keep argument.&lt;/strong&gt; Of the six callers, tim_load_upload, 0x80014420 and screens_load always pass 1.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The two that fill.&lt;/strong&gt; car_load_race&#39;s direct upload (0x80021c74, CLUT at (384, 464 + slot)) and 0x80023aa8 (CLUT at (640, 448 + slot)) pass 0. Both are a car&#39;s skin.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The track.&lt;/strong&gt; Its textures go up through glm_load&#39;s own LoadImage and never meet the check.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;The fill&#39;s sizes.&lt;/strong&gt; It counts bytes from the image&#39;s halfwords. 4-bit fills the whole image with the first texel&#39;s nibble twice, and 8-bit fills it all with the first byte. 16-bit gets only its first half, but no skin is 16-bit.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;In the port:&lt;/strong&gt; &lt;code&gt;car::draw::flat_skin&lt;/code&gt; is the fill. Under cheat 8 the app re-places each car&#39;s flattened skin after the scene loads. &lt;code&gt;hwtr --track desert1 --cheats 8&lt;/code&gt; shows each car in one solid colour.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt; &lt;code&gt;the_flat_cars_fill_as_the_original&lt;/code&gt; (hle tests/tim_upload.rs) runs &lt;code&gt;tim_upload&lt;/code&gt; on the Deora&#39;s skin through the whole hle machine (the upload polls the GPU). It covers depths 0, 1 and 2, with the cheat, the fill flag and keep each switched off once, and compares the image bytes left behind. The plain CPU test machine stalls in two waits: the draw-table count at 0x800117e0, whose clock doesn&#39;t run inside a call, and libgpu&#39;s status poll.&lt;/p&gt;
&lt;p&gt;This closes entry 86&#39;s flat-texture item. The rest is carried here.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The scale cheats in the car&#39;s draws</title>
<id>/86-the-scale-cheats-in-the-car-s-draws</id>
<link rel="alternate" href="/86-the-scale-cheats-in-the-car-s-draws"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="render"/>
<summary>Cheats 2, 4 and 32 change how the car is drawn, not only its body.</summary>
<content type="html">&lt;p&gt;Cheats 2, 4 and 32 change how the car is drawn, not only its body. car_bind_state (0x80021888) sets the model&#39;s scale (cvs +0x14): 8192 under 2 or 32, 1365 (a third) under 4, else 4096. These draws use it:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;draw&lt;/th&gt;&lt;th&gt;cheat&lt;/th&gt;&lt;th&gt;change&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;car_draw (0x80022064)&lt;/td&gt;&lt;td&gt;2 or 4&lt;/td&gt;&lt;td&gt;the body node&#39;s rotation (model +0x4), each column times the scale; wheels on the car go through it&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;car_draw&lt;/td&gt;&lt;td&gt;32&lt;/td&gt;&lt;td&gt;each wheel node&#39;s own rotation (model +0x8 nodes, 72 bytes each); a wheel off the car (cvs +0x1e8) skips the body&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;shadow (0x80029728)&lt;/td&gt;&lt;td&gt;2 or 4&lt;/td&gt;&lt;td&gt;the car&#39;s matrix, so the box shrinks or grows&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;fxp_parse (0x80022cd0)&lt;/td&gt;&lt;td&gt;2 or 4&lt;/td&gt;&lt;td&gt;each glow&#39;s place, in place at load&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;glows (0x80029fb0)&lt;/td&gt;&lt;td&gt;always&lt;/td&gt;&lt;td&gt;size times the scale&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;beams (0x8002a81c)&lt;/td&gt;&lt;td&gt;always&lt;/td&gt;&lt;td&gt;push times the scale; off under 2 and 4, so under 32 they reach twice as far&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;In the port:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;car::draw::{model_scale, body_scaled, wheels_scaled, scale_columns}&lt;/code&gt; hold this logic.&lt;/li&gt;
&lt;li&gt;The shadow takes the scaled matrix.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;glows&lt;/code&gt; and &lt;code&gt;beams&lt;/code&gt; take the scale. Their old fixed 4096 was this value.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Lamps::scale&lt;/code&gt; scales the glow places, from &lt;code&gt;Race::set_lamps&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;The app&#39;s &lt;code&gt;car_triangles&lt;/code&gt; takes a body scale. Each wheel is flagged as on the car or flying.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;hwtr --track NAME --cheats BITS&lt;/code&gt; starts a race under cheats directly. Shots at DESERT1 show third-size cars under 4 and doubled wheels under 32.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Checked against the original:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;shadows_match_the_original&lt;/code&gt; under cheats 0, 2, 4, 32 and 6.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;lamps_draw_as_the_original&lt;/code&gt; with the model&#39;s scale at one, two or a third. This is where the beams&#39; scale turned up: the first run failed on them.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;fxp_lamps_scale_as_the_original&lt;/code&gt; runs fxp_parse under five cheat bytes.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The car model itself is drawn by the app&#39;s renderer, so its scale has no hle check.&lt;/p&gt;
&lt;p&gt;This closes entry 59: no code sets option 0x80, because its slot&#39;s code 77777777 is taken by slot 0 first. It also closes the scale-cheat items of 82 and 85. Entry 82&#39;s reverb item was done in 83. Their other questions are carried here.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Button codes and the cheats they give</title>
<id>/85-button-codes-and-the-cheats-they-give</id>
<link rel="alternate" href="/85-button-codes-and-the-cheats-they-give"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="ui"/>
<summary>The cheats had no way in.</summary>
<content type="html">&lt;p&gt;The cheats had no way in. The port stored a profile&#39;s cheat word, carried it into the race and saved it to the card, but nothing ever set it. The main menu gathered code buttons and never checked them.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;How the original takes a code.&lt;/strong&gt; On the main menu each pad keeps its last eight code buttons as digits: Square 1, Triangle 2, L1 3, R1 4, L2 5, R2 6. After any press (0x8008d010, 0x8008d14c), 0x8007f87c checks them for that pad&#39;s player:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Cheat codes&lt;/strong&gt; (0x8007f710). Ten slots, one per bit, from 0x800bee6c. The first match sets the profile&#39;s cheats to that bit alone.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The car code&lt;/strong&gt; (0x8007f79c). 12345612 gives the car &lt;code&gt;towjam&lt;/code&gt;.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;A match chimes (effect 52) and fills the buffer with spaces.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The codes.&lt;/strong&gt; Five slots (0, 1 and 7 to 9) hold 77777777, so bits 2, 128, 256 and 512 can&#39;t be had. The reachable ones:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;code&lt;/th&gt;&lt;th&gt;cheat&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;77777777&lt;/td&gt;&lt;td&gt;1&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;16522561&lt;/td&gt;&lt;td&gt;4: small cars&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;34563456&lt;/td&gt;&lt;td&gt;8: flat textures&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;64561234&lt;/td&gt;&lt;td&gt;16: DUDE sounds&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;12124455&lt;/td&gt;&lt;td&gt;32: big wheels&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;63124536&lt;/td&gt;&lt;td&gt;64&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;In the port:&lt;/strong&gt; &lt;code&gt;Tables::apply_code&lt;/code&gt; is 0x8007f87c, and the menu runs it after each press. Entering 16522561 in the app&#39;s front end logs &quot;player 1: code 16522561 taken, cheats 0x4&quot;.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;codes_are_taken_as_the_original_takes_them&lt;/code&gt;: 400 codes against random profiles, comparing the answer, cars and cheats with 0x8007f87c.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;a_code_pressed_on_the_main_menu_is_taken&lt;/code&gt;: presses the small-car code frame by frame on the original&#39;s main menu from &lt;code&gt;menu-main&lt;/code&gt;, and checks player one&#39;s cheats and the cleared buffer. (hwtr-hle&#39;s &lt;code&gt;--peek&lt;/code&gt; seems to read before the frames run. It showed nothing until I drove the frames from a test.)&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;On the way:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;The wheels constant.&lt;/strong&gt; &lt;code&gt;HALF_WHEELS&lt;/code&gt; in car/load.rs multiplied the diameter by 0x2000 (two). Cheat 32 doubles the wheels, so it is now &lt;code&gt;BIG_WHEELS&lt;/code&gt;, and the comment says so.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The small cars&#39; puffs.&lt;/strong&gt; &lt;code&gt;effects_init&lt;/code&gt; (0x8002bdb4) ends by setting 0x800d0db6 to 28 under cheat 4. The puff draw (0x8002f618) takes it off each puff&#39;s growth. Ported as &lt;code&gt;Effects::set_small&lt;/code&gt;/&lt;code&gt;puff_shrink&lt;/code&gt;. &lt;code&gt;the_small_cars_shrink_the_puffs_as_in_the_original&lt;/code&gt; checks the value under seven cheat bytes. It also checks that one puff drawn with 0 and with 28 moves the original&#39;s quad template and the port&#39;s corners by the same 28.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;docs/engine/cheats.md now covers the codes and every bit&#39;s readers.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Cheat 16&#39;s DUDE bank and its effects table</title>
<id>/84-cheat-16-s-dude-bank-and-its-effects-table</id>
<link rel="alternate" href="/84-cheat-16-s-dude-bank-and-its-effects-table"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="audio"/>
<summary>Cheat option 16 (0x80013ff8(16) in the race&#39;s sound load, 0x8001924c) changes two things:.</summary>
<content type="html">&lt;p&gt;Cheat option 16 (0x80013ff8(16) in the race&#39;s sound load, 0x8001924c) changes two things:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;The bank.&lt;/strong&gt; &lt;code&gt;DUDE&lt;/code&gt; is opened as VAB 0 instead of &lt;code&gt;MAINSFX2&lt;/code&gt;. Every track&#39;s archive has it: one program of eight tones.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The table.&lt;/strong&gt; It calls 0x8001a73c with 1 instead of 0. That function fills the effects table (0x8011aec0, 16 bytes per id) from the executable&#39;s records (0x800bd5f8: tone, program, note, id). With 1, every effect gets program 0, its tone if under 8 (else 3), and a note of that tone plus 60. The note is taken from the tone already stored, so a substituted 3 gives 63.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;strong&gt;In the port:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;hwtr_game::snd::effects_table(byte, dude)&lt;/code&gt; builds the table both ways. It moved out of the app&#39;s &lt;code&gt;Effects::new&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;The race loads &lt;code&gt;DUDEVH&lt;/code&gt;/&lt;code&gt;DUDEVB&lt;/code&gt; when &lt;code&gt;setup.options &amp;amp; 16&lt;/code&gt;. The options come from the profile&#39;s cheats through the front end, so the cheat entered at the password screen reaches the race.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt; &lt;code&gt;effects_table_matches_the_original&lt;/code&gt; (hle tests/effects_table.rs) calls 0x8001a73c(0) and (1) on a race state and compares all 61 entries, ids included. A guard makes sure the cheat changes more than 40 of them, so the test can&#39;t pass with two equal tables.&lt;/p&gt;
&lt;p&gt;Entry 83&#39;s dude question is answered. Its reverb question is carried here.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The tunnel reverb: libspu&#39;s studio large, deepened in zones with flag 0x80</title>
<id>/83-the-tunnel-reverb-libspu-s-studio-large-deepened-in-zones</id>
<link rel="alternate" href="/83-the-tunnel-reverb-libspu-s-studio-large-deepened-in-zones"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="audio"/>
<summary>0x80015d04, carried since entry 33 as &quot;the master volume for car 0&#39;s flag 0x40&quot;, turned out to be the race&#39;s reverb: a tunnel echo the port did not have.</summary>
<content type="html">&lt;p&gt;0x80015d04, carried since entry 33 as &quot;the master volume for car 0&#39;s flag 0x40&quot;, turned out to be the race&#39;s reverb: a tunnel echo the port did not have.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;What it is:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;The call.&lt;/strong&gt; 0x80015d04 is &lt;code&gt;iface_sound+0x34&lt;/code&gt;. The race&#39;s sound frame (0x800354ac) calls it for each car, but only in a race with one human driver (0x800d2622, counted as the sound loads).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The level.&lt;/strong&gt; The fifth output of 0x80045a84 (the car&#39;s sound input): &lt;code&gt;(car.flags &amp;lt;&amp;lt; 6) &amp;amp; 0x1000&lt;/code&gt;, so 4096 while the car&#39;s flag 0x40 is set. That flag comes from zone flag 0x80.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The target.&lt;/strong&gt; For car 0, unless the sound is paused (0x800d24d3), &lt;code&gt;level × 100 / 4096&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The step.&lt;/strong&gt; The depth (0x800d0c40) moves toward the target by at most 4 a frame (0x80015cc0).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The setter.&lt;/strong&gt; 0x800a66d4 is not &lt;code&gt;SsSetMVol&lt;/code&gt;. It fills a libspu reverb attribute with mask 6, depth left and right, scaled &lt;code&gt;depth × 0x7fff / 127&lt;/code&gt;: &lt;code&gt;SsUtSetReverbDepth&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The voices.&lt;/strong&gt; 0x800a6b94 feeds the first 20 voices to the reverb.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The set-up.&lt;/strong&gt; The race&#39;s sound load (0x8001924c) calls &lt;code&gt;SsUtSetReverbType(4)&lt;/code&gt;, then &lt;code&gt;SsUtReverbOn&lt;/code&gt;. The registers it leaves in the hle&#39;s SPU are psx-spx&#39;s &quot;studio large&quot; preset, with mBASE at 0xf204.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The pause.&lt;/strong&gt; 0x80019f7c turns a sounding reverb off: no depth, no voices, &lt;code&gt;SsUtReverbOff&lt;/code&gt;. Going on turns it back on.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Port:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Game side.&lt;/strong&gt; &lt;code&gt;engines::Reverb&lt;/code&gt; holds the depth, sounding and paused, with &lt;code&gt;step&lt;/code&gt; (0x80015d04) and &lt;code&gt;pause&lt;/code&gt; (0x80019f7c). &lt;code&gt;Change::ReverbDepth&lt;/code&gt; and &lt;code&gt;Change::Reverb&lt;/code&gt; carry it to the app. &lt;code&gt;snd::STUDIO_LARGE&lt;/code&gt;, &lt;code&gt;STUDIO_LARGE_BASE&lt;/code&gt; and &lt;code&gt;reverb_depth&lt;/code&gt; hold the preset and the depth scaling.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;App.&lt;/strong&gt; Turns the reverb on as the race&#39;s sound loads, steps it each frame with one player, pauses it with the race, and drops it with the race&#39;s sound.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Sound chip.&lt;/strong&gt; It had no reverb at all. It now runs the SPU&#39;s reverb as psx-spx describes the hardware: every other sample, from the voices fed to it, through reflections, four combs and two all-pass filters, into a ring the size of the work area, out at the depth.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Checked (hle tests/reverb.rs):&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The 32 registers, mBASE and the reverb enable bit match what the original&#39;s race leaves.&lt;/li&gt;
&lt;li&gt;400 steps in and out of a zone match 0x80015d04&#39;s depth and the output register it writes.&lt;/li&gt;
&lt;li&gt;The pause both ways matches 0x80019f7c.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;A unit test checks that an impulse rings and dies away. In shot runs with sound, the reverb comes in on DESERT1–2, GLACIAL1–3, HAUNTED3 and VOLCANO2 within 40 seconds of driving.&lt;/p&gt;
&lt;p&gt;The reverb&#39;s arithmetic follows psx-spx. The hle models the SPU&#39;s registers, not its sound, so it is not checked against hardware; the hardware&#39;s half-band filters on the reverb&#39;s input and output are left out.&lt;/p&gt;
&lt;p&gt;While reading 0x8001924c I also found cheat bit 16, which loads a bank named &quot;dude&quot;; it is not ported yet.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Entry 75&#39;s remaining questions, carried forward</title>
<id>/82-entry-75-s-remaining-questions-carried-forward</id>
<link rel="alternate" href="/82-entry-75-s-remaining-questions-carried-forward"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="tooling"/>
<summary>Entry 75 carried a list of questions it had not answered.</summary>
<content type="html">&lt;p&gt;Entry 75 carried a list of questions it had not answered. Six of them have been answered since:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;question&lt;/th&gt;&lt;th&gt;answered by&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;action 12&#39;s name&lt;/td&gt;&lt;td&gt;79: the HUD on/off button, now ported&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;menu effects at full volume&lt;/td&gt;&lt;td&gt;76: so are the original&#39;s (0x80015908 takes 4096 from 43 of 44 callers)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;when the music volume applies&lt;/td&gt;&lt;td&gt;76: on boot, after a card read, and from the options and pause menus&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;the turbo&#39;s own sound&lt;/td&gt;&lt;td&gt;79: there is none; only an empty turbo sounds (effect 44)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;the wreck flash&#39;s blend mode, and drawing it&lt;/td&gt;&lt;td&gt;80: mode 0, drawn as a flat quad under the HUD&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;who sets fx_enable&lt;/td&gt;&lt;td&gt;81: 0x80028b34, from the views and the scale cheats&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;This entry resolves 75. The rest of its list carries here, sharpened where the work above learned more.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The scale cheats.&lt;/strong&gt; I looked at them while doing fx_enable. &lt;code&gt;car_bind_state&lt;/code&gt; (0x80021888) sets the model&#39;s scale in cvs +0x14:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;option 2 or 32: 2.0&lt;/li&gt;
&lt;li&gt;option 4: a third&lt;/li&gt;
&lt;li&gt;otherwise: 1.0&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;code&gt;car_draw&lt;/code&gt; (0x80022064) and the shadow use it. The port scales the cars&#39; handling under the cheats but draws every car at its normal size.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Question 66 is still open.&lt;/strong&gt; On the grid, car 0&#39;s body sits 141,000 to 146,700 above its wheels&#39; contacts. Its origin height is 117,590 and its ride height 20,480, which sum to 138,070; the remainder looks like spring sag. That fits the third half-axis being the drop from the body&#39;s centre to the wheels&#39; mounts, but it is not shown, so 66 stays open.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Which effects a race has: fx_enable from the views and the scale cheats</title>
<id>/81-which-effects-a-race-has-fx-enable-from-the-views-and-the</id>
<link rel="alternate" href="/81-which-effects-a-race-has-fx-enable-from-the-views-and-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="render"/>
<summary>Who sets fx_enable (0x800d0d98), carried in 75. The race load sets it, at the end of 0x80028b34 (called through 0x80013f50 with the view count, 0x800d0bc5):.</summary>
<content type="html">&lt;p&gt;&lt;strong&gt;Who sets fx_enable (0x800d0d98), carried in 75.&lt;/strong&gt; The race load sets it, at the end of 0x80028b34 (called through 0x80013f50 with the view count, 0x800d0bc5):&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Its value.&lt;/strong&gt; &lt;code&gt;(views &amp;lt; 2) | 0x1fe&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Cheat options.&lt;/strong&gt; If cheat option 2 or 4 is on (0x800d2468, copied from the race setup&#39;s +0x1c as the renderer starts), bits 0x02, 0x20, 0x80 and 0x100 are cleared.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;What reads each bit.&lt;/strong&gt; I went through every load of the word:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;bit&lt;/th&gt;&lt;th&gt;effect&lt;/th&gt;&lt;th&gt;read by&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;0x01&lt;/td&gt;&lt;td&gt;puffs&lt;/td&gt;&lt;td&gt;0x80030fc8, 0x8003119c&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x02&lt;/td&gt;&lt;td&gt;skid marks&lt;/td&gt;&lt;td&gt;0x8002b888&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x04&lt;/td&gt;&lt;td&gt;sparks&lt;/td&gt;&lt;td&gt;0x8002c32c&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x10&lt;/td&gt;&lt;td&gt;lamp glows&lt;/td&gt;&lt;td&gt;0x80029fb0&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x20&lt;/td&gt;&lt;td&gt;headlight beams&lt;/td&gt;&lt;td&gt;0x8002a81c&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x40&lt;/td&gt;&lt;td&gt;wreck effects&lt;/td&gt;&lt;td&gt;0x80029e10&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x80&lt;/td&gt;&lt;td&gt;the boost flame&lt;/td&gt;&lt;td&gt;0x8002b05c (the whole block, its flicker draws included)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;0x100&lt;/td&gt;&lt;td&gt;(nothing reads it)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;The port had every effect on in every race, as &lt;code&gt;const ENABLED = 0x1ff&lt;/code&gt;. That was right with one view and no cheats. With two views the original has no puffs, and the scale cheats turn off skid marks, headlight beams and the boost flame.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Port:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Effects::enabled&lt;/code&gt; is the word.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Effects::set_enabled(views, options)&lt;/code&gt; is 0x80028b34&#39;s rule, and &lt;code&gt;Race::new&lt;/code&gt; sets it from its cameras and the setup&#39;s options.&lt;/li&gt;
&lt;li&gt;Puffs and skid marks test it.&lt;/li&gt;
&lt;li&gt;The race skips the beams and the flame without their bits, the flame&#39;s random draws included.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;code&gt;the_effects_on_match_the_original&lt;/code&gt; (hle tests/effects.rs) runs 0x80028b34 with views 0 to 3 under ten cheat bytes and compares the word.&lt;/p&gt;
</content>
</entry>
<entry>
<title>A wreck&#39;s screen flash is drawn: a flat quad blended half and half</title>
<id>/80-a-wreck-s-screen-flash-is-drawn-a-flat-quad-blended-half</id>
<link rel="alternate" href="/80-a-wreck-s-screen-flash-is-drawn-a-flat-quad-blended-half"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="render"/>
<summary>A wrecked player&#39;s screen flash is now drawn.</summary>
<content type="html">&lt;p&gt;A wrecked player&#39;s screen flash is now drawn. Before this, the port kept its value but never drew it, because the blend mode was unknown (carried in 75).&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;What 0x8002e128 draws.&lt;/strong&gt; It runs first in the ember update, each frame. For each flash slot that is set:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Primitive.&lt;/strong&gt; A 24-byte POLY_F4 (code 0x2a: flat, semi-transparent).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Size.&lt;/strong&gt; It covers the whole screen, 384 × 240. With two players, each view draws only its own player&#39;s flash, 240 / 2 − 1 high.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Colour.&lt;/strong&gt; A grey of the flash&#39;s value before this frame&#39;s dimming.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Order.&lt;/strong&gt; It goes in ordering-table slot 4, under the HUD&#39;s 1 and 2.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Fade.&lt;/strong&gt; The value then drops by 3; under 130 it is cleared and not drawn. A wreck sets it to 160, so the flash lasts about ten frames.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;The blend mode.&lt;/strong&gt; A flat semi-transparent polygon blends by whatever page the frame left set. I logged the HLE GPU&#39;s page each time a flat semi-transparent quad from the screen&#39;s corner drew, while replaying VOLCANO2&#39;s third racecheck script, which wrecks the player. Every frame of the flash, greys 160 down to 133, came out mode 0: half old, half new. The log was a one-off and is removed.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Port:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Renderer.&lt;/strong&gt; Untextured polygons (mode bit 30, &lt;code&gt;Vtx::FLAT&lt;/code&gt;, an unused page bit), drawn in their colour as it is and blended whole when semi-transparent. A semi-transparent overlay layer, drawn after the world and before the HUD.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Effects.&lt;/strong&gt; &lt;code&gt;flash_drawn&lt;/code&gt; holds this frame&#39;s grey. It is wrapped in &lt;code&gt;Drawn&lt;/code&gt;, frame output that compares equal, so the effects tests that compare whole states still match.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;App.&lt;/strong&gt; Draws a flash quad per shown flash.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt; a temporary forced wreck in a DESERT1 shot (not committed) shows the screen washed grey under the HUD at the wreck, and clear ten frames on. The effects tests pass.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Triangle turns the HUD off and on; the turbo has no sound of its own</title>
<id>/79-triangle-turns-the-hud-off-and-on-the-turbo-has-no-sound-of</id>
<link rel="alternate" href="/79-triangle-turns-the-hud-off-and-on-the-turbo-has-no-sound-of"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="input"/>
<summary>Two questions carried in 75: what action 12 is, and what the turbo&#39;s own sound is.</summary>
<content type="html">&lt;p&gt;Two questions carried in 75: what action 12 is, and what the turbo&#39;s own sound is.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Action 12 is the HUD on/off button&lt;/strong&gt;, Triangle by default. I scanned all 17 calls to the level getter 0x8001b170. Only 0x80034940 asks for action 12, at the end of each player&#39;s race pad read, and passes its level to 0x8006545c:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;No change.&lt;/strong&gt; The level is kept per player at 0x800d0e60; if it hasn&#39;t changed, nothing happens.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;A press.&lt;/strong&gt; The player&#39;s show mask (0x800d0e58) goes to 0 if it shows everything the HUD shows when on (0x800d0e5c, set with it at the race&#39;s setup). Otherwise it goes back to that full mask.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Any change, press or release.&lt;/strong&gt; Every line of the player&#39;s stunt announcement is marked done, both halves (0x800beae4 + 32 a line, +0xd), cutting short one under way.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The original&#39;s controls screen labels the button &quot;HUD On/Off&quot;. The port had no such button, so Triangle did nothing in a race.&lt;/p&gt;
&lt;p&gt;The port now has it:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Controls::hud&lt;/code&gt; is action 12&#39;s level.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Hud::hud_button&lt;/code&gt; does 0x8006545c.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Race::step&lt;/code&gt; calls it for each player after applying the controls, as 0x80034940 does.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;code&gt;the_hud_button_matches_the_original&lt;/code&gt; (hle tests/hud.rs) runs random presses and releases on both players through 0x8006545c and the port. It compares the masks and every half&#39;s done flag after each one: 20 rounds of 60, 487 toggles. In a DESERT1 shot the HUD is gone after one Triangle press.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The turbo has no sound of its own.&lt;/strong&gt; A turbo (0x80049440) makes one indirect call, 0x8012fe04, which is 0x8002af60: the boost flame, and it calls nothing. car_update&#39;s turbo branch sounds only when there is no turbo left: effect 44 through &lt;code&gt;iface_sound&lt;/code&gt;, which the port already plays. Entry 33 had counted a &quot;boost&quot; among the car&#39;s voices. The voices at 0x8011a840 +0x4c to +0x52 are the tyres, the scrape, the crash and the impact.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Two write-only questions: the fastest lap&#39;s car and the flying table&#39;s in-use byte</title>
<id>/78-two-write-only-questions-the-fastest-lap-s-car-and-the</id>
<link rel="alternate" href="/78-two-write-only-questions-the-fastest-lap-s-car-and-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="decomp"/>
<summary>Two small questions, answered from the executable alone.</summary>
<content type="html">&lt;p&gt;Two small questions, answered from the executable alone.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Who reads the fastest lap&#39;s car (0x800d0e69, from 67).&lt;/strong&gt; Nobody. There are two stores and no load:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;0x80061e04, the HUD&#39;s race setup, writes 0xff (none).&lt;/li&gt;
&lt;li&gt;0x80064ce4, the standings, writes the fastest car.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Nothing else touches it:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Loads:&lt;/strong&gt; no gp-relative load and no address built from &lt;code&gt;lui 0x800d&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Neighbours:&lt;/strong&gt; the only base pointer nearby is 0x800d0e60, a two-entry per-player array.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Pointers in data:&lt;/strong&gt; a scan of the executable finds none to 0x800d0e60–0x800d0e69.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;So it is a write-only byte. The port keeps nothing for it.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The flying table&#39;s in-use byte (from 71).&lt;/strong&gt; Only 0 or 1 ever lands in a record&#39;s +0:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;the table&#39;s set-up at race load (0x8007c7a0) and clearing (0x8007c824) write 0&lt;/li&gt;
&lt;li&gt;the throw (0x8007c9b0) writes 1&lt;/li&gt;
&lt;li&gt;taking a car&#39;s wheels away (0x8007da78) writes 0&lt;/li&gt;
&lt;li&gt;a snapshot put back (0x8007ec08) copies a byte from a snapshot, and only in-use records are saved&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The snapshot code&#39;s branch for kind 1 never meets another kind.&lt;/p&gt;
&lt;p&gt;Docs: results.md (standings, layout) and flying-wheels.md (the in-use byte).&lt;/p&gt;
</content>
</entry>
<entry>
<title>The sound chip takes ADPCM from rrt-kit; no emulator in the game binary</title>
<id>/77-the-sound-chip-takes-adpcm-from-rrt-kit-no-emulator-in-the</id>
<link rel="alternate" href="/77-the-sound-chip-takes-adpcm-from-rrt-kit-no-emulator-in-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="build"/>
<summary>retro_rt moved PS-ADPCM and the SPU&#39;s Gaussian interpolation from rrt-emu into rrt-kit (rrt_kit::adpcm::spu, unchanged, with its tests), and added a CD-XA decoder (rrt_kit::adpcm::xa).</summary>
<content type="html">&lt;p&gt;retro_rt moved PS-ADPCM and the SPU&#39;s Gaussian interpolation from rrt-emu into rrt-kit (&lt;code&gt;rrt_kit::adpcm::spu&lt;/code&gt;, unchanged, with its tests), and added a CD-XA decoder (&lt;code&gt;rrt_kit::adpcm::xa&lt;/code&gt;). I updated rrt from c693b31 to 7325d77.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The sound chip off the emulator crate.&lt;/strong&gt; The app&#39;s sound chip (crates/hwtr/src/spu.rs) decoded its samples with &lt;code&gt;rrt::emu::spu&lt;/code&gt;. That was the only thing linking rrt-emu, the originals&#39; CPUs and HLE, into the game binary, against the rule that the hwtr app is the port alone. It now uses &lt;code&gt;rrt::kit::adpcm::spu&lt;/code&gt;, the same functions moved. The app drops rrt&#39;s &lt;code&gt;emu&lt;/code&gt; feature: &lt;code&gt;cargo tree -p hwtr&lt;/code&gt; no longer lists rrt-emu. hwtr-hle never used it.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;XA.&lt;/strong&gt; Nothing to use it for. The disc has no XA: no file is Form 2 (docs/disc/layout.md), the music is CD-DA and the movies are EA&#39;s WVE. If a later disc or revision turns out to carry XA, the decoder is there.&lt;/p&gt;
&lt;p&gt;Checked: the workspace tests pass, and a DESERT1 race runs with the sound chip on (&lt;code&gt;HWTR_SHOT_SOUND=1&lt;/code&gt;).&lt;/p&gt;
</content>
</entry>
<entry>
<title>Sound levels: the menu&#39;s effects, the music volume and CdMix</title>
<id>/76-sound-levels-the-menu-s-effects-the-music-volume-and-cdmix</id>
<link rel="alternate" href="/76-sound-levels-the-menu-s-effects-the-music-volume-and-cdmix"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="audio"/>
<summary>I went after three open questions about sound levels.</summary>
<content type="html">&lt;p&gt;I went after three open questions about sound levels. Two turned out to be how the original works; the third was a real gap in the port.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Menu sounds at full volume (from 37, carried in 75).&lt;/strong&gt; The menu&#39;s sound player is 0x80015908, behind &lt;code&gt;iface_game+0x1c&lt;/code&gt;. It scales only by the level it is given (4096ths, made 127ths) before &lt;code&gt;SsUtKeyOn&lt;/code&gt;. I checked all 44 call sites: 43 pass a fixed 4096. The remaining one is the Sound FX slider&#39;s release (0x80092f50), which plays at the new level. Moving the slider (0x80092fd0, 0x8009311c) sets nothing global for effects. So the original&#39;s menu sounds ignore the slider too, and the port already matched it.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;0x800a5c58 is CdMix (32).&lt;/strong&gt; Entry 32 took it for libsnd&#39;s master volume. It calls 0x800abc08, which writes the CD drive&#39;s attenuation registers (index 2, then 3, then apply 0x20). That is libcd&#39;s &lt;code&gt;CdMix&lt;/code&gt;: the music level for the left into the left and the right into the right, nothing across.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;When the music volume applies (from 36, carried in 75).&lt;/strong&gt; 0x80014b6c has six callers:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;the defaults (0x8008a184)&lt;/li&gt;
&lt;li&gt;a successful card read (0x80088828; a failed read sets nothing)&lt;/li&gt;
&lt;li&gt;the options screen&#39;s two sliders&lt;/li&gt;
&lt;li&gt;the pause menu&#39;s music line&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The port did only the options screen and the pause menu, and set the CD level when a race started. So after loading a save with quieter music, the menus played at the default level. Both boot paths now ask for it.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The arithmetic.&lt;/strong&gt; The port computed &lt;code&gt;m × 127 / 255&lt;/code&gt;. The original rounds down twice, &lt;code&gt;((m &amp;lt;&amp;lt; 12) / 255 × 127) &amp;gt;&amp;gt; 12&lt;/code&gt;, which comes out one lower for settings 249, 251 and 253. &lt;code&gt;hwtr_game::cd::music_level&lt;/code&gt; now does it the original&#39;s way. &lt;code&gt;the_music_level_matches_the_original&lt;/code&gt; (hle tests/sound.rs) checks all 256 settings against the level 0x80014b6c keeps at 0x800d2488.&lt;/p&gt;
&lt;p&gt;New reference page: docs/engine/sound-levels.md.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Closing the questions later work answered</title>
<id>/75-closing-the-questions-later-work-answered</id>
<link rel="alternate" href="/75-closing-the-questions-later-work-answered"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="tooling"/>
<summary>The site counted 70 of 74 entries as open.</summary>
<content type="html">&lt;p&gt;The site counted 70 of 74 entries as open. Most of those questions had been answered by later work, but never closed. No entry used &lt;code&gt;--resolves&lt;/code&gt;, and a few wrote &quot;None new&quot; or &quot;None for this change&quot; where cairns wants the literal &lt;code&gt;nothing&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;I went through &lt;code&gt;cairns open&lt;/code&gt; entry by entry and checked each question against the code and the reference (grepping for every address it named). Where it was answered, this entry resolves it, citing what answered it. Whatever part is still truly unknown is carried into this entry&#39;s own &lt;strong&gt;Still unknown&lt;/strong&gt;, so nothing is lost by closing. Questions with no answer yet (2, 3, 4, 5, 6, 7, 12, 15, 16, 28, 30, 32, 35, 38, 52, 58, 59, 62, 63, 66, 67, 71, 72, 73) stay open where they are.&lt;/p&gt;
&lt;p&gt;Entries 24, 25, 49, 61, 70 and 74 meant &quot;nothing new&quot; and now say &lt;code&gt;nothing&lt;/code&gt;.&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;entry&lt;/th&gt;&lt;th&gt;answered by&lt;/th&gt;&lt;th&gt;carried forward&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;1&lt;/td&gt;&lt;td&gt;35: SLUS_009.64 plays the movies, shows PSXLEGAL then PSXRFA1, then runs CCCPSX.EXE&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;9&lt;/td&gt;&lt;td&gt;TIMs carry their own VRAM rectangle; 32: each sample&#39;s pitch comes from the note keyed&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;10&lt;/td&gt;&lt;td&gt;11 (the controls module reads libpad), 29 and 63 (vibration and jolts)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;11&lt;/td&gt;&lt;td&gt;11 and controls.md: the curves (dead zone 10), the action table; 63: the vibration path&lt;/td&gt;&lt;td&gt;action 12 (Triangle)&#39;s name; this entry names 9 to 11 in controls.md&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;14&lt;/td&gt;&lt;td&gt;15: x, y, z, w&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;17&lt;/td&gt;&lt;td&gt;32 (sound)&lt;/td&gt;&lt;td&gt;analog stick against the digital pad in a race, unchecked&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;18&lt;/td&gt;&lt;td&gt;car-object.md: +0xec gravity&#39;s direction, +0x770 the origin; computer cars run their driver (ai.rs); the hook-cost question went with the emulator (25)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;19&lt;/td&gt;&lt;td&gt;car-object.md: +0x6b4 handling flags, +0x86a/+0x86b air control, surface 6&#39;s drag; the tuning bytes are TUNINGPRM&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;20&lt;/td&gt;&lt;td&gt;41/42 (0x8006b754, the moving volumes), 65 (0x8007c894, flying wheels), car-object.md (air control 0x8003d71c)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;22&lt;/td&gt;&lt;td&gt;recovery (race.rs &lt;code&gt;reset_car&lt;/code&gt;, 0x80041384) and the player&#39;s effects (0x8003cb74, 48 and 50) ported&lt;/td&gt;&lt;td&gt;what damp_spin&#39;s 8 and 25 stand for&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;23&lt;/td&gt;&lt;td&gt;68 (start placement, driver setup); the object (0x8004c478) and zone flags in the port; handling.rs and car-object.md (+0x04 steering lock, +0x6c the computer&#39;s pace, wheel +0x10 steering angle)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;26&lt;/td&gt;&lt;td&gt;43 (results, snapshots), hud.rs (countdown 0x8001e060, power-up icon 0x80062534), 50 (stunts), 44 (computer cars), race_over (time limit), laps.rs &lt;code&gt;watch_way&lt;/code&gt; (0x8005c9b4)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;27&lt;/td&gt;&lt;td&gt;66, 41/42, 46, 65, 67, 48 and 63&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;31&lt;/td&gt;&lt;td&gt;44: 0x800d2608 is the attract race&#39;s flag&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;33&lt;/td&gt;&lt;td&gt;48, 51, 60, 62&lt;/td&gt;&lt;td&gt;0x80015d04 (master volume for car 0&#39;s flag 0x40), not ported&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;34&lt;/td&gt;&lt;td&gt;37 (garage previews and bars), 38 (cup screen 0x8008e310), 0x8009a89c ported&lt;/td&gt;&lt;td&gt;the preview&#39;s NCCT lighting (0x80032ae8); the main menu&#39;s stat bars (0x8011dfb4); the menu&#39;s engine sound; who sets 0x800d10b0&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;36&lt;/td&gt;&lt;td&gt;37 (the options screen&#39;s music lines)&lt;/td&gt;&lt;td&gt;the TOC entry past the last track; when the music volume is applied&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;37&lt;/td&gt;&lt;td&gt;40 (controls screen, 0x8009495c)&lt;/td&gt;&lt;td&gt;menu effects still play at full volume&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;39&lt;/td&gt;&lt;td&gt;44 (attract race), 40 (card manager); the front end now reports no unported actions&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;40&lt;/td&gt;&lt;td&gt;44 (attract race)&lt;/td&gt;&lt;td&gt;card screens can&#39;t be checked against hle (no card there); the &#39;card seen&#39; flag (0x800d0f28)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;41&lt;/td&gt;&lt;td&gt;57 (triggered animations), 41/42 (moving volumes), 65 (flying wheels)&lt;/td&gt;&lt;td&gt;the pickup touch is taken after the collision step, not inside it&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;42&lt;/td&gt;&lt;td&gt;46 (the knocked prop&#39;s debris), 62 (its sound at a spot)&lt;/td&gt;&lt;td&gt;no hle check of a knock&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;43&lt;/td&gt;&lt;td&gt;71 (flying wheels in snapshots), 70 (0x800d0e54), 56 (a put-back wreck&#39;s look)&lt;/td&gt;&lt;td&gt;what the draw-mode byte 0x800d246c changes&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;44&lt;/td&gt;&lt;td&gt;69 (0x800836e8)&lt;/td&gt;&lt;td&gt;trackside camera flags past bit 1; the front end&#39;s setup against hle&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;45&lt;/td&gt;&lt;td&gt;46 (debris, embers, smoke), 47 (boost flame), 45 (puffs)&lt;/td&gt;&lt;td&gt;the wreck&#39;s screen flash is kept but not drawn; the billboard units; who sets fx_enable (0x800d0d98)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;46&lt;/td&gt;&lt;td&gt;65 (seen on screen; 0x8007c9b0)&lt;/td&gt;&lt;td&gt;the flash&#39;s blend mode; the volume record&#39;s +8 position&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;47&lt;/td&gt;&lt;td&gt;car-wheels.md: cvs +0x18 is twice the last wheel&#39;s lift&lt;/td&gt;&lt;td&gt;the turbo&#39;s own sound&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;48&lt;/td&gt;&lt;td&gt;60 (+0x48), 62 (world sounds, VAB 2)&lt;/td&gt;&lt;td&gt;the two-player mixer (as 62); prop velocity (as 63)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;50&lt;/td&gt;&lt;td&gt;51 (commentary and the ten-turbo line)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;51&lt;/td&gt;&lt;td&gt;62 (VAB 2)&lt;/td&gt;&lt;td&gt;what the dialog bank&#39;s tones say&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;53&lt;/td&gt;&lt;td&gt;57 (trigger zones)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;54&lt;/td&gt;&lt;td&gt;car-lights.md: +0xbc glow strength, +0xc4 light targets&lt;/td&gt;&lt;td&gt;cvs +0xc0&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;55&lt;/td&gt;&lt;td&gt;58 (+0x130, the light colours)&lt;/td&gt;&lt;td&gt;the model scale cheats (options 2 and 4)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;56&lt;/td&gt;&lt;td&gt;58&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;57&lt;/td&gt;&lt;td&gt;62 (looped trigger sounds, their follow and stop)&lt;/td&gt;&lt;td&gt;a differential check of the load-time marking&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;60&lt;/td&gt;&lt;td&gt;62 (the world&#39;s records silenced); 0x80014a0c ported&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;64&lt;/td&gt;&lt;td&gt;73 (racecheck&#39;s matrix over all tracks includes wrecks)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;65&lt;/td&gt;&lt;td&gt;71 (the snapshot records)&lt;/td&gt;&lt;td&gt;the rest of the 680-byte record (as 71)&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;68&lt;/td&gt;&lt;td&gt;73 (all 11 tracks)&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;69&lt;/td&gt;&lt;td&gt;70&lt;/td&gt;&lt;td&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;I also fixed two stale pages found on the way:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;rigid-body.md still called 0x8006b754 and 0x8007c894 unidentified.&lt;/li&gt;
&lt;li&gt;controls.md now names actions 9 (reset), 10 (turbo) and 11 (change view).&lt;/li&gt;
&lt;/ul&gt;
</content>
</entry>
<entry>
<title>One tap steps the garage&#39;s car once</title>
<id>/74-one-tap-steps-the-garage-s-car-once</id>
<link rel="alternate" href="/74-one-tap-steps-the-garage-s-car-once"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="ui"/>
<summary>Reported in play: one tap of up in the garage stepped the car two or three times.</summary>
<content type="html">&lt;p&gt;Reported in play: one tap of up in the garage stepped the car two or three times.&lt;/p&gt;
&lt;p&gt;Reproduced with the front-end example: one 4-frame up press in the garage stepped at frames 1000 and 1003.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Cause.&lt;/strong&gt; When a car&#39;s model arrives, the original clears pad 0&#39;s held-step delay (0x8008b9d0, called from the garage draw 0x80091cf0; the main menu&#39;s car row and the cup screen do the same). In the original that is harmless: it reads the car from the CD only a second and a half after the last step, so the button is always up by then. The port has every model in memory and shows the new car within a frame or two, while the tap is still held. With the delay back at 0, the next blank stepped again.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Fix.&lt;/strong&gt; &lt;code&gt;Front::car_arrived&lt;/code&gt; clears the delay only when the pad is not stepping, i.e. with the button up. Letting go has already cleared it then, as in &lt;code&gt;held_repeat&lt;/code&gt;. Used in the garage, the main menu and the cup screen.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Checked:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;One tap now steps once.&lt;/li&gt;
&lt;li&gt;Held, the car still steps at 0 ms, then after 500 ms, then every 250 ms (frames 1000, 1030, 1045, 1060...), as 0x8008ba30 does.&lt;/li&gt;
&lt;/ul&gt;
</content>
</entry>
<entry>
<title>The ground stage&#39;s best distances run across the cars; racecheck on all 11 tracks</title>
<id>/73-the-ground-stage-s-best-distances-run-across-the-cars</id>
<link rel="alternate" href="/73-the-ground-stage-s-best-distances-run-across-the-cars"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="physics"/>
<summary>I took racecheck past DESERT1: it now has race states for all 11 tracks.</summary>
<content type="html">&lt;p&gt;I took racecheck past DESERT1: it now has race states for all 11 tracks. Running it there found one real port bug in the ground stage and three more harness gaps.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;New states&lt;/strong&gt; (work/states/&lt;track&gt;-race.bin, each a race just after the start):&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Unlocked tracks.&lt;/strong&gt; From a boot run, the menus are driven to the main menu&#39;s TRACK row. Right is pressed once per track (120 frames apart, held 4 frames), then up and X. One up is always lost while a preview loads, so the X opens the garage. A second X returns to the menu on CARS, and up + X starts the race. A fresh profile has six tracks: Desert 1-3 and Glacial 1-3.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Locked tracks.&lt;/strong&gt; For Volcano 1-3 and Haunted 2-3, the player&#39;s profile (pointer at 0x800d2764) gets track bits 0xfff poked at +0x1c before the presses.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;The port bug: ground (0x800536b4).&lt;/strong&gt; The original keeps its best nearest and floor distances (s5 and a stack slot) from 0, once, across all the cars, not per car. A car&#39;s first candidate takes its place regardless, so this mostly doesn&#39;t show. But a player&#39;s car whose wheels gave it a nearest surface and no floor (steep wheels, as on Snake River Mine&#39;s banks) measures its zones&#39; candidates against the previous car&#39;s best. The port reset the best per car. It now carries it across cars. On DESERT2&#39;s first script this took 89 differing steps to 0.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;ground_matches_the_original&lt;/code&gt; now covers it:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;nine states instead of four&lt;/li&gt;
&lt;li&gt;two steps in three, the player&#39;s wheels set on a steep plane with gravity up&lt;/li&gt;
&lt;li&gt;the car searched before it raised&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;It fails on the old code and passes on the new.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Harness gaps in racecheck:&lt;/strong&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Grace over.&lt;/strong&gt; When a reset&#39;s grace runs out, &lt;code&gt;cars_update&lt;/code&gt; puts the car back in the collision (object flag 1 cleared). The world taken at collision entry (item 3) already has it.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Power-ups.&lt;/strong&gt; A power-up taken in the pair search is applied by the race. A step where the original&#39;s car changed &lt;code&gt;power_up&lt;/code&gt; is now counted, not checked, like a reset.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The world at collision entry.&lt;/strong&gt; A computer car&#39;s update can take its object out of the collision or reset it to another zone before the collision runs. The port&#39;s collision step now starts from the original&#39;s whole collision world as 0x8004de6c begins, not the world at the start of the step.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Flying wheels.&lt;/strong&gt; As in &lt;code&gt;Collision::update&lt;/code&gt;, racecheck now runs &lt;code&gt;add_flying&lt;/code&gt; first.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Events.&lt;/strong&gt; The port-side &lt;code&gt;jolted&lt;/code&gt; flag is cleared before diffing, with the other event queues.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;A trigger fired inside the original&#39;s step can create kind-0 objects. The port&#39;s race creates them after the step, so object numbers differ. That only renumbers objects, and racecheck reports the counts when they differ.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Result.&lt;/strong&gt; All 11 tracks × 4 input scripts (steering, reset spam, square, triangle with long turns): 26,908 steps, 26,867 the same, 0 different. 41 steps with a reset or a power-up are not checked.&lt;/p&gt;
</content>
</entry>
<entry>
<title>racecheck: check whole steps between cars_update and the collision&#39;s return</title>
<id>/72-racecheck-check-whole-steps-between-cars-update-and-the</id>
<link rel="alternate" href="/72-racecheck-check-whole-steps-between-cars-update-and-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="test"/>
<summary>racecheck reported differences with steering and turbo scripts: 5, 12 and 32 steps on the drive, speed and air states.</summary>
<content type="html">&lt;p&gt;racecheck reported differences with steering and turbo scripts: 5, 12 and 32 steps on the drive, speed and air states. That held at the earlier commit 2b4ca54 too, so it was not a regression. All of them turned out to be the harness, not the port.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Steps cut by the vertical blank.&lt;/strong&gt; &lt;code&gt;Hle::frame&lt;/code&gt; stops at the vertical blank wherever the main loop is. A race step can therefore straddle two frames: at steps 85, 86 and 100 the step counter moved but the collision stages ran in the next frame. racecheck had compared a half-done original step with the port&#39;s whole one, and the original&#39;s wheel contacts and ground looked one step behind. racecheck now only checks a frame in which the cars&#39; update (0x8004064c) began and the collision update (0x8004de6c) returned.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The step&#39;s tail.&lt;/strong&gt; The previous step&#39;s places, wrong-way watch, camera and clock tick could run at the start of the next frame. The start state is now memory as &lt;code&gt;cars_update&lt;/code&gt; begins, and the end state is memory as the collision update returns, both captured by hooks, instead of the frame&#39;s edges. This fixed &lt;code&gt;contact_time&lt;/code&gt; off by one clock tick.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The wrong-way watch.&lt;/strong&gt; 0x8005c9b4 runs inside &lt;code&gt;cars_update&lt;/code&gt;, after the places. racecheck now runs the port&#39;s &lt;code&gt;laps::watch_way&lt;/code&gt; after the car update.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Port events in the diff.&lt;/strong&gt; The &lt;code&gt;Pending&lt;/code&gt; event queues the port keeps for the race (sounds, lines and others) printed in the debug diff and misaligned it. They are cleared before diffing.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Resets.&lt;/strong&gt; A reset step is counted, not compared: racecheck does not model the reset path.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;On a difference, racecheck now also reports:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;the car as the collision began against the port&#39;s car after its update&lt;/li&gt;
&lt;li&gt;the whole world before the stages&lt;/li&gt;
&lt;li&gt;each stage (zones, wheels, ground) run from the original&#39;s world against the original&#39;s result&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;One gotcha found on the way: &lt;code&gt;Machine::check&lt;/code&gt; keeps one check per address, so a second hook on the same function replaces the first.&lt;/p&gt;
&lt;p&gt;Result, with every whole step checked:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Script &lt;code&gt;0:x:600,100:right:40,250:left:60&lt;/code&gt;: 412 + 606 + 612 + 612 steps, all the same.&lt;/li&gt;
&lt;li&gt;Script &lt;code&gt;0:x:900,50:left:120,300:square:60,400:right:200&lt;/code&gt;: 411 + 608 + 612 + 612, all the same.&lt;/li&gt;
&lt;li&gt;Script &lt;code&gt;0:x:900,0:r1:900,...&lt;/code&gt;: everything the same except 17 resets, which are not checked.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The port&#39;s race step matches the original on every step checked, including today&#39;s computer-car walls stage.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The results&#39; snapshots keep the flying wheels (0x8007e6e0, 0x8007ec08)</title>
<id>/71-the-results-snapshots-keep-the-flying-wheels-0x8007e6e0</id>
<link rel="alternate" href="/71-the-results-snapshots-keep-the-flying-wheels-0x8007e6e0"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="ui"/>
<summary>The results&#39; snapshots now keep the flying wheels, so the slideshow shows a wrecked player&#39;s wheels where they were at each moment.</summary>
<content type="html">&lt;p&gt;The results&#39; snapshots now keep the flying wheels, so the slideshow shows a wrecked player&#39;s wheels where they were at each moment.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Saving (0x8007e6e0).&lt;/strong&gt; Each in-use slot of the table at 0x801323f4 is written in slot order as a 16-byte record:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;the in-use byte as a u16 (1 for a wheel)&lt;/li&gt;
&lt;li&gt;the car and the wheel&lt;/li&gt;
&lt;li&gt;the body rotation&#39;s first two rows in 127ths, clamped as for cars&lt;/li&gt;
&lt;li&gt;the body place &amp;gt;&amp;gt; 12&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;With no wheel flying, the part takes 4 bytes and writes nothing. If the records do not fit, the whole snapshot is not taken.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Putting back (0x8007ec08).&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;With nothing kept, every slot is freed.&lt;/li&gt;
&lt;li&gt;Otherwise record i goes into slot i from the first, so the wheels close up. Each slot gets the in-use byte, the rotation (third row the cross product, as for cars), the place &amp;lt;&amp;lt; 12, and the car and wheel. Nothing else in the slot is written.&lt;/li&gt;
&lt;li&gt;The slots past the last record are freed.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Port changes:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Snapshot.wheels: Vec&amp;lt;WheelShot&amp;gt;&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Snapshots::take&lt;/code&gt; and &lt;code&gt;put_back&lt;/code&gt; take the collision&#39;s flying table, and the race passes &lt;code&gt;collision.flying&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;A slot the port had empty is filled with a default body. Only car, wheel, rotation and place matter while the results stand still.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Verified: the snapshot test in hle tests/snapshot.rs now places random wheels in RAM and the port before a snapshot. It compares the snapshot bytes, the wheels section included, and the size. Then it moves the wheels into other slots, puts the snapshot back, and compares every slot&#39;s car, wheel, rotation and place with the original&#39;s table. A third run with no wheels checks the 4-byte empty part.&lt;/p&gt;
&lt;p&gt;Docs: results.md&#39;s layout and put-back sections; flying-wheels.md.&lt;/p&gt;
</content>
</entry>
<entry>
<title>The stunt results stop the race&#39;s clock (0x800d0e54)</title>
<id>/70-the-stunt-results-stop-the-race-s-clock-0x800d0e54</id>
<link rel="alternate" href="/70-the-stunt-results-stop-the-race-s-clock-0x800d0e54"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="ui"/>
<summary>Ported the last piece of the stunt points&#39; results table (0x80064294): stopping the clock.</summary>
<content type="html">&lt;p&gt;Ported the last piece of the stunt points&#39; results table (0x80064294): stopping the clock.&lt;/p&gt;
&lt;p&gt;0x800d0e54 is the time the HUD&#39;s &quot;time left&quot; counts down from. At the race start (0x80061ed8) it is set from the setup&#39;s limit (+0x20), only under flag 4, the race against the clock. Each time the stunt points&#39; table is drawn, it takes player one&#39;s time so far: the race clock less car 0&#39;s lap start (+0x5a8). If that is less than 0x800d0e54 (unsigned), it becomes the new value. The time left therefore reads zero from the moment the results appear.&lt;/p&gt;
&lt;p&gt;In the port this is &lt;code&gt;Hud::stop_clock&lt;/code&gt;, called from the stunt branch of the results drawing on &lt;code&gt;Hud::limit&lt;/code&gt;, which is the port&#39;s 0x800d0e54. Without a time limit, &lt;code&gt;limit&lt;/code&gt; is None and nothing changes; in the original it would lower a value nothing shows.&lt;/p&gt;
&lt;p&gt;Verified with &lt;code&gt;the_stunt_results_stop_the_clock_as_in_the_original&lt;/code&gt; (hle tests/laps.rs). It calls 0x80064294 150 times with random clock, lap start and limit, and compares 0x800d0e54 with &lt;code&gt;Hud::limit&lt;/code&gt;. The limit was lowered in 34 of the cases, and every case matches.&lt;/p&gt;
&lt;p&gt;Docs: results.md, under the standings.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Clear stale port gaps; the attract race drops the front end&#39;s picture</title>
<id>/69-clear-stale-port-gaps-the-attract-race-drops-the-front-end</id>
<link rel="alternate" href="/69-clear-stale-port-gaps-the-attract-race-drops-the-front-end"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="engine"/>
<summary>Went through the remaining &quot;not yet ported&quot; notes.</summary>
<content type="html">&lt;p&gt;Went through the remaining &quot;not yet ported&quot; notes. Three of them had been ported already, and one small gap was real.&lt;/p&gt;
&lt;p&gt;Stale notes, now corrected:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The knocked props&#39; debris (0x8002e27c) is already thrown at race.rs&#39;s knock handling. The stage that steps flying wheels (0x8007c894) is in since worklog 65. The trace is removed.&lt;/li&gt;
&lt;li&gt;Props in the crash check (0x8007e000): flag 2 knocks a prop over and flag 4 bumps the car. Both already go through the port&#39;s knock (0x8006b958), which marks the object, records the volume and draws the debris numbers, and the race plays its sound at a spot. The doc comment now says so.&lt;/li&gt;
&lt;li&gt;Snapshot put-back: the wreck&#39;s charred or cleared look (0x80029e10 / 0x80029f04) is set by the race&#39;s results code after each put-back, so the trace in &lt;code&gt;put_back&lt;/code&gt; was stale.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Real gap, now fixed:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The attract race&#39;s setup (0x8009b254) first calls 0x800836e8. That calls through 0x8011dfc0, which a menu-state peek shows is 0x80015648, the background picture loader. Given no name, it clears the picture flag 0x800d2401. The port now clears &lt;code&gt;Front::background&lt;/code&gt; there, as it already does for the password screen (0x800836c0 with 0, the same loader).&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;A menu-to-race smoke run (shot mode, 4000 frames, cross pressed every second, races cut at 600 frames) runs the attract race, then several DESERT1 races and their results, without errors.&lt;/p&gt;
&lt;p&gt;Still open: the points table lowering 0x800d0e54, camera modes other than 0 to 4, and object kinds other than those listed.&lt;/p&gt;
</content>
</entry>
<entry>
<title>loadcheck holds the computer cars and their drivers to the original</title>
<id>/68-loadcheck-holds-the-computer-cars-and-their-drivers-to-the</id>
<link rel="alternate" href="/68-loadcheck-holds-the-computer-cars-and-their-drivers-to-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="test"/>
<summary>loadcheck now checks the computer cars as well as the players&#39;.</summary>
<content type="html">&lt;p&gt;loadcheck now checks the computer cars as well as the players&#39;. Each computer car is built the port&#39;s way and compared with the original when &lt;code&gt;cars_load&lt;/code&gt; (0x8003beec) returns:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Car::load&lt;/code&gt; places it on the grid.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;race::computer_start&lt;/code&gt; (0x8007c6fc) moves it to its route&#39;s start.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Ai::add&lt;/code&gt; (0x80079e58) gives it a driver, built from the AI read out of RAM as &lt;code&gt;cars_load&lt;/code&gt; entered.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;loadcheck also compares each computer driver with the original&#39;s.&lt;/p&gt;
&lt;p&gt;The run found two port gaps and two harness gaps.&lt;/p&gt;
&lt;p&gt;Port gaps:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;The computer cars&#39; start placement was already in &lt;code&gt;Race::new&lt;/code&gt; but had never been checked. It matches. It now lives in &lt;code&gt;race::computer_start&lt;/code&gt;, and the stale &quot;not yet ported&quot; note on &lt;code&gt;Car::load&lt;/code&gt; is gone.&lt;/li&gt;
&lt;li&gt;The driver&#39;s respawn line is part of the car record (+0x7d0), zeroed at load. Its choice word therefore reads 0, branch 0, not -1, until the first &lt;code&gt;car_update&lt;/code&gt; keeps a real one. &lt;code&gt;Ai::add&lt;/code&gt; now starts it that way.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Harness gaps:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;model_faces&lt;/code&gt; is filled by the app from the model&#39;s root faces, so loadcheck now does the same.&lt;/li&gt;
&lt;li&gt;The collision&#39;s course has to be set before &lt;code&gt;add_car&lt;/code&gt;, as &lt;code&gt;Race::new&lt;/code&gt; does, or a car whose zone is ahead of the line gets lap distance 0 instead of 69400.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Result with the default script (DESERT1, one player and five computer cars): all six cars and all five drivers are the same as the original&#39;s.&lt;/p&gt;
&lt;p&gt;&lt;code&gt;LOADCHECK_DUMP=DIR&lt;/code&gt; writes a differing driver&#39;s full debug output on both sides, for diffing.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Estimate the unfinished computer cars&#39; times for the standings (0x80061824)</title>
<id>/67-estimate-the-unfinished-computer-cars-times-for-the</id>
<link rel="alternate" href="/67-estimate-the-unfinished-computer-cars-times-for-the"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="race"/>
<summary>When a race ends, the results now estimate times for computer cars still short of the line, as 0x80033aa0 does through 0x80061824.</summary>
<content type="html">&lt;p&gt;When a race ends, the results now estimate times for computer cars still short of the line, as 0x80033aa0 does through 0x80061824. They also order the field the way the original does.&lt;/p&gt;
&lt;p&gt;Before this, every unfinished computer car showed no time and scored nothing.&lt;/p&gt;
&lt;p&gt;The estimate (&lt;code&gt;Laps::estimate&lt;/code&gt;) works like this:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;This lap.&lt;/strong&gt; Take the driver&#39;s race left to run (AI record +0x1a8, &lt;code&gt;Driver::progress&lt;/code&gt;). Make it positive and multiply by 1000, then take away the later laps at the lap length. Hold that at 0, using an unsigned compare. Divide by the car&#39;s pace (handling +0x6c, &lt;code&gt;ai_pace&lt;/code&gt;) and add it to the race clock: that is this lap&#39;s end.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Each later lap.&lt;/strong&gt; The lap length over the pace, plus &lt;code&gt;random(lap length / 20)&lt;/code&gt; from the race&#39;s rand.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Then&lt;/strong&gt; the best lap is updated, and the car is finished with all its laps done.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The ordering is a selection sort with swaps, not a stable sort:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;For each place, it picks the fastest time from that place on.&lt;/li&gt;
&lt;li&gt;A car with no time never displaces one.&lt;/li&gt;
&lt;li&gt;Points 10/8/7/6/5/4 go to the first six with a time.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The port&#39;s &lt;code&gt;Race::standings&lt;/code&gt; now calls the free function &lt;code&gt;race::standings&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;One finding: the lap length the estimate reads is 0x800d0e48. Only the race start (0x80061264) sets it, copying it from the best line, so it is still 0 in the saved states, which are mid-countdown. The tests poke it as the start would. If a race ended before the start, &lt;code&gt;random(0)&lt;/code&gt; would hit the divide-by-zero &lt;code&gt;break&lt;/code&gt;.&lt;/p&gt;
&lt;p&gt;Verified in hle tests/laps.rs:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;finish_estimates_match_the_original&lt;/code&gt;: random laps, times, pace and progress over 3 states. 478 estimates were checked against 0x80061824: the laps record, the returned time and best, and the rand seed.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;standings_match_the_original&lt;/code&gt;: 0x80033aa0 with the snapshot and the sound&#39;s end hooked out. Every place (0x800d2688), every car&#39;s points (0x800d25f8), the laps and the rand seed match, including ties and cars with no time.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Docs: results.md gains a &quot;The standings&quot; section.&lt;/p&gt;
</content>
</entry>
<entry>
<title>Wrecked and finished computer cars meet the walls as boxes (0x800572f0)</title>
<id>/66-wrecked-and-finished-computer-cars-meet-the-walls-as-boxes</id>
<link rel="alternate" href="/66-wrecked-and-finished-computer-cars-meet-the-walls-as-boxes"/>
<updated>2026-10-04T00:00:00Z</updated>
<category term="physics"/>
<summary>Ported the collision stage that keeps wrecked or finished computer cars inside the track (0x800572f0).</summary>
<content type="html">&lt;p&gt;Ported the collision stage that keeps wrecked or finished computer cars inside the track (0x800572f0). Before this, such a car met the walls only through its single centre point, so it could sink halfway into the road or a wall.&lt;/p&gt;
&lt;p&gt;How it works:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Each awake computer car that is wrecked (+0x62c) or has finished (+0x5e3) is treated as a box. The half axes are the body&#39;s rotation columns times the object&#39;s half width, its half length and the car origin&#39;s height (+0x778).&lt;/li&gt;
&lt;li&gt;Only the zone of its point is tried. Plane zones come first, then road zones.&lt;/li&gt;
&lt;li&gt;Each side meets the box corner deepest toward it. Each axis is negated when it points along the normal, and the negation persists into the next side. That matters for ties (an axis square to a side), and the port keeps it.&lt;/li&gt;
&lt;li&gt;Planes: no radius prefilter. Kinds 3 and 4 set 0x2000 and 0x1000 on the car.&lt;/li&gt;
&lt;li&gt;Road sides: every normal is computed. The distance is taken at the centre, less &lt;code&gt;sum |a_i . n|&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Going through sets 0x800 instead of putting the body to sleep.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Port changes:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;code&gt;Collision::computer_walls&lt;/code&gt; in collision/walls.rs runs after &lt;code&gt;walls&lt;/code&gt; in &lt;code&gt;stages&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;Shape { Point, Ball, Box }&lt;/code&gt; replaces the ball option in &lt;code&gt;road_sides&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;&lt;code&gt;press&lt;/code&gt; now takes the flags to mark when a body goes through. The walls stage passes them only for player cars, as before.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Verified with &lt;code&gt;computer_walls_match_the_original&lt;/code&gt; in hle tests/collision.rs: 80 steps over 4 states, with computer cars wrecked or finished at random and half-turned about each axis. About 1180 boxes were tested, 306 of them in road zones, with 495 contacts and 691 go-throughs. Every contact and car matches.&lt;/p&gt;
&lt;p&gt;New doc: docs/engine/track-walls.md (both wall stages).&lt;/p&gt;
&lt;p&gt;Also removed a stale &quot;not yet ported&quot; trace for the knocked props&#39; debris, which was already ported.&lt;/p&gt;
</content>
</entry>
</feed>
