Month: August 2026

This week was mostly about one thing: finishing the remaining Gus game bugs.

Gus Paint

One of the bugs was in the inside() Lingo function, which was causing the Paint and Play buttons in Gus Paint to crash. I found the fix, but it was slightly wrong in its assumptions. I figured that out a couple of days later when I looked at it again and made the required change.

The game passes a proper POINT and the rect of cBackground, but cBackground is temporarily VOID during the first few frames. This meant inside() received (POINT, VOID) and fell through without returning a value, causing a fatal error.

Fixed it to return FALSE for invalid arguments instead of crashing. Sev also suggested using the existing TYPECHECK macros for handling the different argument types.

There were some other smaller bugs I worked on, like a textbox getting disproportionally larger incrementally with text input. funniest bug so far. fixed it. This one along with some more bugs, will be listed in the PR.

I spent a lot of time simply going through the games and reproducing things that looked wrong.

Over the last few weeks I have gradually gotten better at narrowing those problems down. Instead of immediately assuming that the game itself is doing something unusual, I can now usually trace the failing Lingo call into the engine and figure out what assumption ScummVM is making that the original Director did not.

Looking back

This week was mostly about cleanup: playing through games, finding the remaining failures, tracing them into the engine, and getting them into a state where they can be fixed properly.

And with that, GSoC is coming to an end.

Over the last 12 weeks I have worked on a fairly wide range of things in the Director engine: the ImGui visual debugger, movie cast members, film loops, Lingo debugging and profiling, ImageDiff and buildbot integration (-_-), game detection, and a lot of small engine and game compatibility fixes.

It has been a pretty different experience from working on a normal project. A lot of the time I was dealing with code and file formats that are decades old, trying to figure out not just what the original game is doing, but what Director itself must have been doing underneath it.

I’ll soon be posting a separate write-up covering the work I did over the full 12 weeks.

Because they need some testing, I”ll push the PR for the bugs soon. (Updating soon)

Upadate: https://github.com/scummvm/scummvm/pull/7854

that’s it for week 12 : )

So the highlight of this week is, as the title suggests, the Lingo tracer. I mentioned it in a recent blog. When I shared a screenshot with sev of the prototype I made, he said it seemed like a good idea and green lit it. This week was spent mostly on it, and getting it working was not that hard. The hard part was making it easy to use and understand.

The tracer records Lingo as it runs: every handler that gets called, when the score changes frames, and when a script is frozen and later resumed. It draws all of that as a flamegraph inside the ImGui debugger, so you can see which handlers ran, how they nested, and how long each one took. There is also a small stats table that sums the total and self time per handler, so the expensive ones are easy to spot. It stays off until you tick Capture, so it costs nothing when you are not using it.

Most of my time went into the reading experience. Frames and freeze/thaw markers sit in a thin strip at the top so the flamegraph below stays clean, tiny sub-pixel slices get merged so a busy trace does not turn into noise, and hovering one call highlights every other call with the same name.

fix

The other fix this week was a sound bug in gusmuse. Entering or leaving an area stacked the music tracks on top of each other, and the previous one never stopped.

In Director the sound channels are one shared resource between the Stage and any movie-in-a-window, but in ScummVM each window had its own sound manager, so a track started from one window could not stop a track from another.

I routed every window’s sound through the Stage’s manager, and now the channels are truly global.

That’s it for the week : )

Mostly a normal week: two engine bugs, and a lot of sharpening the ImGui debugger.

Bugs.

Fixed the Cybertown cyberbud trails vanishing. The dirty channel check was comparing the raw _thickness byte, so a flipped tween bit falsely dirtied the channel and erased the trail. Fixed the Sink-or-Float “Prof. Gus’ face” (Talked about it here: week 5): a digital-video sprite the score sizes to 0×0 (an invisible timing clock) was being blown up to its native 240×180, so now it stays invisible. Also null-guarded a crash in endOfVideo().

Debugger (DT).

A sweep of bugs and quality of life fixes: killed ImGui::Text format string crashes in the script viewer, added some keyboard shortcuts, cached the cast browser rows, made the cast list sortable, added a Vars filter and copy value, refreshed stale thumbnails, added a movie cast member score viewer and a lot of other changes.

Lingo profiler.

I also spent part of the weekend on an experimental feature (for fun), a profiler for Lingo. It is still early, and I will come back to it later. For now the focus is wrapping up the Gus games.

Gus games status

• gustown, guscarn: done
• gusolis: 90% done
• gusmuse: in progress
• guspaint, guspark: remaining (still need to check how many bugs survive all the recent changes)

PRs

https://github.com/scummvm/scummvm/pull/7780
https://github.com/scummvm/scummvm/pull/7781
https://github.com/scummvm/scummvm/pull/7779