Repository navigation
Keep a pinned fishing float from turning NaN, in a port replacement - #225
Closed
SirFancyBacon wants to merge 1 commit into
Closed
SirFancyBacon wants to merge 1 commit into
SirFancyBacon wants to merge 1 commit into
Conversation
While the float is pinned (FishLineSetUki with a weight of 0 or more, as on every cast), FishLineStep copies uki into both point[18] and ukip[0] at the end of each relaxation pass, so on the next the length between them is 0 and length * (1 - weight) / length is 0/0. The EE's floating point gives a finite result and the float stays still; a host's gives NaN, which runs through the float and the line into Toan's position. On macOS ARM the game crashes in CCollisionMDT::PickUpNearPoly from EdMoveChara as the line is cast; on Linux x86_64 the view turns to a single colour and leaving the fishing menu crashes. FishLineStep is replaced in port/src/fishing.cpp by retail's body with a zero length taken as 1, the same zero move the EE makes, and nothing else changed; ps2/src is untouched. Tororoi found the cause and wrote the guard and FishingLine.APinnedFloatStaysFinite in TheMoonPeople#219, which this carries over as a replacement because ps2/src is not edited for the port. docs/PC.md lists it under a new "Arithmetic the EE does differently". Co-authored-by: Thomas Cantwell <cantwell.tom@gmail.com>
Contributor
|
I recommend merging PR 230 instead |
Adubbz
added a commit
that referenced
this pull request
Oct 27, 2026
…e open PRs The port no longer compiles ps2/: the game's code is forked into port/src/game and port/include/game, and PORT_OVERRIDE is gone. ps2/ stays the byte-matching decomp. - PS2-only code removed from the fork: MWCC pragmas, code the PC link discarded, SDK calls that do nothing on PC, dead conditionals. - All port code brought to AGENTS.md and C++26: anonymous namespaces, nullptr, C++ casts, std containers and strings in place of fixed buffers and C strings, enum class, no goto, meaningful names, minimal comments, documented headers. - Every function in a logical file: the platform layer is in port/src/<subsystem> (app, data, platform, render, input, ui, options, title, battlemenu, dialog, quest, item, text, localize, save); the old replacement units are merged into their game units; no *_port files, stub headers, linknames.cpp or linker aliases remain. - Namespaces: dc::game for the game, dc::<subsystem> for the platform layer, dc::test for the tests. Only main, extern "C" entry points and the symbols Kazarov calls stay global. - CI per platform: pc.yml (Linux, Windows, Flatpak), macos.yml and switch.yml, with the Switch toolchain image cached on ghcr.io. Kazarov compiles pipelines on first use on a CPU rasterizer, so lavapipe no longer runs test processes out of memory. - Merged PRs, simplified and restyled: #186 (autosave on each dungeon floor and resume there), #220, #223 (item framework and gameplay documents), #226 (sound publishing), #227, #228, #229, #230 (the fishing line's springs divide as the EE does; supersedes #219 and #225). Each step was gated on Windows: build, unit tests, and headless screenshots of ten game modes, pixel-identical to master's build. Co-authored-by: Alessandro Suha <plarpoon@zandark.eu> Co-authored-by: Thomas Cantwell <cantwell.tom@gmail.com> Co-authored-by: SirFancyBacon <m.johnsondelta@gmail.com> Co-authored-by: IvanPerrella <ivan.perrella@gmail.com>
Member
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Carries over #219 by @Tororoi as a
PORT_OVERRIDEreplacement, sops2/srcstays untouched. Tororoi found the cause and wrote the fix and its test; they're credited as the author of this fix.Fixes: #224
Requires: #220
The bug
While the fishing float is pinned (
FishLineSetUkiwith a weight of 0 or more, as on every cast),FishLineStepcopiesukiinto bothpoint[18]andukip[0]at the end of each relaxation pass. On the next pass the length between them is 0, solength * (1 - weight) / lengthis 0/0.CCollisionMDT::PickUpNearPoly(fromEdMoveChara) as the line is cast.The fix
FishLineStepis replaced inport/src/fishing.cppby retail's body with one guard: a zero length is taken as 1, which gives the same zero move the EE does. Nothing else in the copy changes (compared againstps2/src/fishing.cpp, the only difference is the guard), and the two point-pair tables it reads,staticin retail's unit, come with it.docs/PC.mdgets a short section, "Arithmetic the EE does differently", listing it, in the style of the pointer replacements' table.Testing
FishingLine.APinnedFloatStaysFinite, from Keep a pinned fishing float from turning NaN #219, fails on Linux x86_64 without the change and passes with it. The other fishing tests pass.ps2/srcis unchanged, so the PS2 builds are unaffected.