This article was originally written in Japanese and translated into English with AI assistance. Please note that some expressions may carry nuances from the original Japanese.
🇯🇵 日本語版はこちら / Japanese version
Series: utf8conv Development Journal (Part 10)
Last time, I wrote up to the point where, on the evening of September 14, 2026, I ran a conversion from the screen with my own hands.
The machine I normally use for development is a Mac, but this time is the story of the 2 days on a Windows PC that I own separately, from when I opened utf8conv’s window to when the way its screen is built changed.
On the Windows PC, the Window Opened
On the afternoon of September 16, to verify how it runs, I launched utf8conv on the Windows PC that I own separately. It was not run in a form bundled up for handing out to many people; it was run by starting the files still being built (the source) as they were. All I checked was whether the window opened. I did not run file conversion, which is utf8conv’s basic function.
The window opened from the source code without any problem. The window’s title read “UTF-8 Converter”.
That PC did not have uv, the tool that sets up Python, so I installed it on the spot. The Python that utf8conv actually used was 3.12.14, set up by uv.
I also found one difference from the Mac. tkinter (the part that puts up the screen, which I wrote about in the third installment of this series) draws the screen using a mechanism inside it called Tk. The version of that Tk was 9.0 on the Mac and 8.6 on this Windows PC.
The Window Was Spilling Off the Bottom of the Screen
utf8conv’s window did open, but the first window that opened was spilling off the bottom of the screen. The bottom of the log area, the lowest part of the window, was hidden behind the taskbar (the strip that runs along the bottom of the screen).
The screen is 1920×1080, and display scaling is 125%. When I maximized the window, everything fit, right down to the bottom of the log area. It was not that the contents could not fit; it was that the size of the window when it first opens was larger than the height available.
Shrinking the Log Was Not Enough
At first, I intended to fix it by cutting the log area down to 10 lines so that it would fit within the screen.
The height of a window is measured in pixels. A pixel is the size of one of the dots that make up the screen. The AI in the role that builds to the brief made a procedure for measuring the size of the window on Windows, and I ran it on that PC. The parts other than the log came to 671 pixels on their own, and even with the log at 10 lines, it was 79 pixels short.
When I measured on the Mac as well, the same problem turned up. macOS was shrinking the window to a height of 921 pixels, and the line at the very bottom, the one that reports the status, had become 1 pixel. Windows lets the window spill off the screen; macOS shrinks the window down to a height that fits on the screen. What was shrunk was being taken from the parts at the very bottom. Only the way it looked was different; it was the same problem.
Is This an App That Needs to Fill the Whole Screen?
I had 5 proposals made for changing how the parts inside utf8conv are arranged. I dropped 2 of them because their width went beyond the screen. And 3 were long in the vertical direction, at a size that covered the entire screen.
Comparing the proposals as images, I arrived at a question: “Is this really an app that needs to fill the whole screen?”
Take, for example, the video editing software FCP X. The library of files you work with is at the top left of the screen, the playback area is at the top center, the settings area is at the top right, and the timeline is at the bottom; its functions are divided up like that, and you carry out the work of editing video by using all of them at the same time, or by turns.
utf8conv, on the other hand, is only about specifying a directory, specifying the language type, running it, and having the log displayed. It has nothing like functions at a level that would use the whole screen at once, and there is no reason it has to.
A wide log area is convenient for checking the log, but I think that the moment the number of target files grows, it becomes impossible to check all of it, so I felt that building that part up was not quite right, either.
As a result, I decided to rework the screen layout.
This software is not the type, like video editing software, where you operate continuously across several functions; it is the type where you “decide the settings one at a time, in order, and then just run it and wait.” So it is not the case that every function and every setting has to be shown all at once.
So I decided to rebuild it as a collapsible screen. Only 1 section is open, and the other sections are shown folded down to 1 line each.
Doing it this way, besides saving space, also produces the effect of “limiting what you need to look at.” Even if several clicks are needed, because what you need to look at is limited, my view is that it will not become a burden on the user’s cognitive capacity either.
Into the same rebuild, I also put other design-change items.
- A cancel button that stops a run partway through
- Displaying that a Dry Run, which runs it as a trial without converting, has been done
- Removing the 「オプション」 (Options) section
- Fixing the log area at 20 lines
- Since the log does not need to be visible while you are adjusting the settings, hiding the log part, which takes up space
- Adding new ▲▼ buttons for moving between the settings of each section
The Collapsible Screen Worked on Windows
The next day, on the morning of September 17, the collapsible screen was ready. At midday that day, I checked it on the Windows PC. The checks that measure the size and position of the window’s parts by machine came to 58 items, with 0 failures.
I also tried the cancel button. When I pressed cancel partway through actually converting 200 files, it stopped at 99. When I ran it again, the 99 already done were skipped, and the remaining 101 were converted. 200 in total.
Even so, the window was still spilling off a little. Although the size of the window fit, because of where it was placed, its bottom edge overlapped the taskbar by about 37 pixels.
Just 1 Set of Arrows
Looking at the screen, there was one more thing I pointed out. The ▲▼ buttons for moving back and forth between the sections of the settings screen were attached to every section. Because the structure was divided into 5 sections in all, 5 sets of ▲▼ were displayed. There was no need to show ▲▼ buttons fixed to each item; preparing 1 set shared across all the screens keeps the buttons in the same position and is easier to understand, so I decided to let 1 set of ▲▼ buttons stand for them all. Also, the ▲ and ▼ buttons were lined up side by side, needlessly taking up horizontal space, so I decided to stack them vertically at the right edge of the window.
At the same time, I also decided where the window sits when it launches: horizontally, in the center of the screen; vertically, 20 pixels from the top edge.
I checked the version with these fixes applied on the Windows PC once more. The bottom edge of the window was at 717 pixels from the top of the screen, inside the height excluding the taskbar (816 pixels). The ▲▼ were in the same position on every one of the 5 screens.
The ▲▼ Can Now Be Pressed with Keys, Too
That night, I made it possible to press ▲▼ with keys as well. On Windows, Alt+N goes to the next one and Alt+B goes back. On the Mac, Cmd+] goes to the next one and Cmd+[ goes back. The reason I did not use Cmd+N and Cmd+B on the Mac is that they overlap with the keys for “New” and “Bold.”
I also made it so that when you hold the mouse over a button for 0.5 seconds, a short explanation appears.
I went back from Windows to the Mac and confirmed that, on the Mac keyboard as well, Cmd+[ and Cmd+] move between sections. Alt+N and Alt+B on Windows I did not check at that time.
The README Described a Feature That Did Not Exist
I wrote the result of the window opening on Windows into the brief that lays out the procedure for building the distributable.
On the same day, September 16, I found that the README (the manual that describes how to use it) described a feature that did not exist. The README said “they can be deleted all at once,” but when I searched the program’s files, that feature was nowhere to be found.
I decided to remove that one sentence from the README. As for deletion, I have a basic feeling of “I do not want to touch anything as frightening as deleting,” so rather than having the software make irreversible changes, I am fixing things in a direction that keeps them as reversible and safe as possible.
Next time, I will continue from here.
About Soul Resonant Works
Soul Resonant Works is a solo venture developing seven local AI systems.
Starting from zero programming experience, the development is progressing through collaboration with AI.
🌐 Soul Resonant Works:
→ https://sr-works.net/en/index.html
📝 This blog publishes the entire development process as a serialized journal.
CubePlot (free version available)
CubePlot is the first product from Soul Resonant Works — it turns a CSV into a 3D scatter plot you can rotate and zoom. No install, no sign-up: it’s a single HTML file you open in your browser, and it works offline. CubePlot itself does not send the data you load outside your machine — external network traffic is blocked at the browser level (CSP). Start with the free version.
▶ Product page: https://sr-works.net/en/cubeplot/
▶ Get it (commercial license, USD $39 + tax where applicable): https://soulworks8.gumroad.com/l/pzcij
If you found this article useful, please share it.
