The swamp of a modified keyboard (Reiwa NeXT keyboard V2)

The swamp of a modified keyboard (Reiwa NeXT keyboard V2)

This article is the 22nd day of the advent calendar of Studio Aquatan 2023.

Hi, this is Prof. omzn (keyboard modification studies).
This is a continuation from a week ago.

Reiwa NeXT keyboard V2

Actually, I have an unmodified NeXT keyboard, but this one is junk and does not work even if it works.
I decided to modify this one, too.

The eBay price of the NeXT keyboard as of December 2023 is as shown below.

I’m so nervous about wasting it…

Now, we will realize the following three points.

  • No pattern cutting
  • Change the physical keys to compatible ones
  • Make them light up

Key layout

Similar to V1, but the Return key is inverted L-shaped, and the backslash key is pushed out to the numeric keypad.
The backslash key is pushed out to the numeric keypad.
The engraving looks like old Macintosh.

Physical keys

This keyboard uses ALPS black keys.
The ALPS axis is no longer manufactured by ALPS Electric.

Key Matrix Analysis

The matrix pattern of the original NeXT keyboard is analyzed. With a tester in hand, we follow the pattern on the back side.
This time, I will make the keyboard without cutting the pattern. We will also make a board on which Pi Pico will be placed, and fit it into the place where the original controller was removed.

The following figure shows the result of the analysis.

The keys close to each other are compactly arranged, and the matrix looks like it could fit in an 8×10… but only the modifier keys are sticking out. What’s this? If you look closely, you can see that the line marked R10 is GND. So, these keys are not matrix keys, but direct keys.

Since we are not using a microcontroller with that many pins, we integrate the modifier keys into the matrix by following the GND pattern and removing the jumper from GND to make it a single bus.
The final matrix is as follows. The modifier is integrated in the rows and columns for the convenience of the Pi Pico board.

We will make a board to convert this matrix to Pi Pico. This is very convenient because the modifier keys can be integrated only on this board.

After designing the board, I ordered it from Elecrow. The cost of the board is $1.00. Shipping is about $15 and it takes less than a week. I decided that it would be better to solve the problem with money than to try hard to arrange the line where the bug would enter.


I looked around for a way to make the LEDs glow from behind the keys, but in the end I decided to replace all the switches.

The board has a single-sided pattern, so it is easy to remove the solder.
However, the copper foil pattern itself is brittle and easily peels off since it is an old board.
With great care, we removed the solder and the switches.





I got Matias Click Switch Gray and White from FILCO Direct. Only the numeric keypad is made to click with White.

The jumper on the surface of the switch will be removed and re-mounted on the back side of the switch, because the LED strip will be in the way.
The consumption speed of the desoldering wire is extraordinary.
In the end, I had to remove more than 90% of the mounted parts.

In addition, the Caps Lock LED is changed to a white LED.

Next, I installed LED strips to make the LEDs shine. 10mm wide LED strips are commonly used, but they seem to interfere with the keyswitch (just barely), so I got 5mm wide ones this time.

First row…

2nd row…

All rows…


This is the end of the work on the board, so let’s solder the keyswitch.
Once the keyswitch is soldered, the LED strip cannot be touched, so it is necessary to double-check for poor contact.


When I finished soldering the keyswitches, the key matrix conversion board arrived from Shenzhen.

Connectors for I2C and WS2812 were also pulled out.
This is used to bridge the Raspberry Pi Pico and the keyboard board.
As long as there are no bugs on the board, the wiring work can be completed without connecting a single wire.

Test lighting! Good!

The transparent OLED is also attached via I2C.
This is exactly the same as V1, so it is easy to do.

All that is left is to assemble it.


It’s done!

The code of QMK is available in GitHub.

Bugs and failures along the way

  • At first, I misunderstood how to light the LEDs on the main board, and thought the 5V line was GND. I made the first conversion board as it was, so I had to do a major re-do and ended up reordering.
  • After installing the LED strips and soldering all the keyswitches, I did a key matrix test and found that one line was unresponsive. The tester found that only when the LED strip was connected, 5V was connected to the line in question. This was because when the jumper wire was replaced, a small portion of the jumper wire was in contact with the 5V of the LED strip.


It is fun to revive old keyboards by modifying them.
In particular, keyboards with mechanical switches from the 1990s are worth reviving because they have a good feel when you hit the keys.

If you find a good keyboard in a junk shop, you can get it immediately!

Merry Christmas and a happy new year!

The way to a modified keyboard (Reiwa NeXT keyboard V1)

The way to a modified keyboard (Reiwa NeXT keyboard V1)

This article is the 15th day of the advent calendar of Studio Aquatan 2023.

This is Prof. omzn (keyboard modification studies).

Heisei Version of the NeXT Keyboard

When I was still young, in the 2000s, I enjoyed modifying old computers to convert them into the latest PCs.
I modified NeXT cube, NeXT station, Macintosh LC II (I think), Mac G4 cube, etc. to make a mini PC.
Despite all the hard work, we produced a terrible machine that could not deliver much performance, and would run out of control quickly because of its broken thermal design.
I feel bad for the old computers that were sacrificed.

The NeXT keyboard was something like a Mac-style ADB, but at the time it was just a useless keyboard.
Then, young omzn had a flash of inspiration.

“I think I can get a controller from a junk PS/2 keyboard lying around here and port it to the NeXT keyboard.”

He immediately investigated the keyboard mechanism, and found that the keyboard had a key matrix decoder on it, and, unfortunately, the key matrices were all different.

“I want to make the most of this PS/2 keyboard I’ve taken apart…”

Here, omzn had another flash of inspiration.

“I see, I can rewire the NeXT side to fit this PS/2 keyboard.”

Thus was born the Heisei version of the NeXT keyboard.

I thought it would be OK to use PS/2, which was the standard at the time, but the era soon gave way to USB, and the NeXT keyboard was used for nearly 20 years with the USB conversion. However, because it was originally a PC/AT keyboard, it did not work well with Macs, and I had to use it with key-swapping software. In addition, there was no way to access the function keys, which sometimes caused me a great deal of trouble.

Therefore, for 20 years, I had been thinking, “Someday, I would like to remake this keyboard into a modern keyboard….”

Reiwa NeXT Keyboard V1

In the fall of 2023, Yuyu showed me a keyboard that he had made by himself, which ignited my heart, and I started to aim for a complete transition to a self-made keyboard.

Basically, what we need to do is the same as before, which is to make a matrix decoder that matches the physical matrix. QMK provides an easy mechanism to do this.


The keyboard looks like a normal keyboard, except that the Function keys are missing, and the backquote is on the numeric keypad, but that’s OK.

Physical Switches

The NeXT keyboard uses ALPS cream axes. The light click feeling is very addictive.
(For more information about ALPS axis, please visit this site.)

Matrix Analysis

The above of the following figure shows the matrix of a PS/2 keyboard that I had posted on my web page 20 years ago.
The matrix was so extravagantly constructed that it would have required 25 matrix pins as it is.
I was able to reduce the number of columns by three, since the columns can simply be merged where they do not conflict with each other even if they take ORs (see the bottom of the figure). (bottom of the figure).

This still requires 22 GPIOs.
The Pi Pico has 26 GPIOs, so there are still 4 GPIOs left over for this application. We can increase the number of functions by allocating them to LED driving and I2C.

Hardware creation

Put Pi Pico on the board.

Cut the razor cable connected to the PS/2 controller, and connect it to Pi Pico’s GPIOs.
This is the first prototype, so we soldered the cable as it is, but it would be better to make the cable removable with a socket.

Install the USB terminal

3D print the base of the Pi Pico and make the USB port. The photo shows the prototype in red, but the final model will be in black.

The final shape looks like this. Compared to before modification, it looks much cleaner now that the extra board is gone.

Configuration of QMK

Configuration of the keyboard

  • Define MATRIX_HAS_GHOST for a matrix without diodes, like this keyboard.
  • The LED of CapsLock is controlled by GP0.


The keymap is written in the same way as the previous matrix.

  • Use NeXT Power key as Fn key.
  • Page up, Page down for Volume up, Volume down (Fn key is the original function)
  • Brightness up, Brightness down as Home, End (press Fn for original function)
  • Fn + number” for “F1~F10
  • Left Command key alone for “Eisuji”, Right Command key alone for “Kana
  • The left Command key alone is “Eisuji”, the right Command key alone is “Kana”, and so on.
  • The Esc area is somewhat special, with a single press of Esc, Shift + Esc for “~”, and Shift + Alt + Esc for “`”.

The correspondence between the GPIO and the matrix to be used is written in info.json.

Creating graphical key layout with KLE

Create a layout in KLE for GUI when you change the key layout in VIA or Remap.

The key is to write the coordinates of the matrix in the Legend of the keytop.
The matrix of the PS/2 controller does not match the physical layout at all.

Setting up VIA

Create via.json file. This is used to change the keymap in GUI with tools such as Remap.
The contents of this file contains the key arrangement data created by KLE, the keyboard name, Vender ID, Product ID, and matrix of matrix.

The keyboard is ready to use!

After burning the firmware, the keyboard should be ready for use.

qmk flash -kb next_keyboard_v1 -km via

The source code is available on GitHub.

(Extra) Adding an OLED panel

The keyboard by itself is a little bit lonely, so I added a HUD-like display using a transparent OLED. This is a bit of a long story.

Actually, this story was just a prelude.
The main story will be in the next article.

See you!

Build Your Own Leopard Gecko Incubator

Build Your Own Leopard Gecko Incubator

This is the 8th day of Studio Aquatan 2023 Advent Calendar.

Hello, this is omzn, the lizard professor.

I have been keeping Leopard Gecko since last year, and have been managing it by borrowing a part of Akuatan system. (Click to enlarge) This system is now much larger than the aquarium.

there are four cages for lizards…

Today’s topic

This year, my leopard lizard started to lay eggs, so I made an incubator after much trial and error.
The first eggs hatched in my incubator, so I am recording them here.

Leopard Lizard Eggs

The leopard gecko generally lays two eggs per brood.
The eggs are laid in an egg-laying site created by digging up the bottom floor of the cage and then backfilled, so they can be dug up and transferred to the incubator. After the eggs are retrieved, the laying female still tries to hide them by covering the area where she laid them with soil, but it is a futile effort….

The eggs have soft eggshells that feel fluffy when touched. If the egg is accidentally rolled or dropped, it will die, so I mark with a magic marker where the egg was on the top when I dug it out. After that, always store the eggs with this mark on the top.
The incubator is placed in a Tupperware container lined with moss. Check manually from time to time to see if the moss is keeping the eggs moist.

Management of Leopard Lizard Eggs

It is recommended that the eggs of the Leopard Lizard Monster be maintained at a temperature of 26°C to 34°C and a humidity of 80% to 90%. Depending on the temperature, it is said that the eggs hatch in 40 to 70 days.
Since it is quite a long process, it is best to keep the eggs in a quiet place with stable temperature and humidity. The eggs hatched in 53 days after being kept at 28°C. Two more eggs hatched around 53 days later.


If the humidity is not sufficient, the eggs will dry out and die. If the humidity is too high, the eggs will dry out and die, and if the humidity is too low, they will become moldy.
Therefore, I would like to keep the inside of the incubator at a constant humidity level. As for humidification, the incubator will be humidified on its own if we place something containing moisture (such as water moss) inside the incubator without doing anything in particular, so we will build the incubator with the policy of reducing the humidity that has risen too high.


Temperature-dependent sex determination (TSD) determines sex according to the temperature during incubation[1].
(Note that if the incubation temperature is kept below 28°C, almost 100% of the eggs will be female.)
Therefore, temperature control is essential if the sex is to be fixed.


Temperature-dependent determination of leopard geckos (quoted from [1])

Regardless of the sex, the temperature should not fall below 26°C, so some kind of heat-retention device is necessary. Here, I use a sheet heater to keep the temperature on the higher side. In my breeding environment, the room temperature is always around 26°C because the entire building is air-conditioned 24 hours a day. Therefore, in environments where the room temperature is too high and cooling is necessary, cooling must be taken into consideration.

Installation of the incubator

According to the above management policy, a mechanism to control the temperature and humidity inside the incubator is to be built.
The required components are as follows

  • M5 HUB Switch D (2-channel 100 VAC relay)
  • M5ATOM Lite (M5 ATOM S3, M5 ATOM S3 Lite, or any other type)
  • DC fan (5V drive)
  • ENV III Unit (SHT30 temperature/humidity sensor)
  • MOSFET (2SK4017)
  • 10kΩ resistor x 2
  • Sheet heater
  • DAISO shoe case

The main body of the incubator is a small shoe case from Daiso. A regular sized shoe case would be fine, but it may be too wide to hold the incubator. Also, the small size may be available only at larger Daiso stores. The small size can hold up to 10 eggs, and since it is 100 yen, there is no hesitation in processing the material, which is very good.

For humidity control, a fan is mounted.

Send PWM to the MOSFET through pin 25 of M5ATOM Lite to turn the fan in steps of 0 to 255. (Actually, the fan will not turn unless the PWM is set at about 65 because electromotive force is required.)
The fan is installed in a hole in the incubator lid, and the fan is turned in a direction that releases the internal moisture to the outside.
The humidity is measured by the ENV III Unit’s SHT30. The temperature and humidity can be easily obtained with this temperature/humidity sensor.
The fan is driven by PID control using humidity as an input. Since humidity is sensitive to fan operation, the humidity acquisition interval is as short as 2 seconds. In reality, it would be fine to simply turn the fan ON and OFF at a threshold value, but the following code is used for study purposes.

int Fan::manageByHumid(float h) {
  const float dt = 2;
  const float kp = 10, ki = 1, kd = 10;

  static float diff_p = 0, diff_c = 0, integral = 0;
  float p, i, d;
  int prev_power = fan();

  diff_p = diff_c;          // 湿度の差 (%)
  diff_c = h - _target_humid;  // 湿度の差 (%)
  integral += (diff_c + diff_p) / 2.0 * dt;
  integral = integral > 50 ? 50 : (integral < -50 ? -50 : integral);

  p = kp * diff_c * (diff_c < 0 ? 5 : 1); // 1 % で pwm 10 :負の時は x 5
  i = ki * integral;                //
  d = kd * (diff_c - diff_p) / dt;  // 1% で pwm 10
  float power = p + i + d;
  int ipower = constrain((int)power, 0, 255);
  DPRINTF("humid: %.1f, target: %.1f, power: %.1f (%3d)\n", h, _target_humid, power, ipower);
  return (ipower && !prev_power || !ipower && prev_power);

The temperature is controlled by driving the seat heater with an AC relay.
Using M5 HUB Switch D allows ATOM Lite to run on a 100 V power supply, eliminating the need for extra wiring.
The second GROVE connector on the body of the M5 HUB Switch D has pin 25, so the fan circuit described above can be made at the end of this GROVE connector. Since the temperature change is not as short as the humidity change, the upper and lower thresholds are set to control the ON and OFF of the fan.

The final appearance will look like this.

The software will use a general-purpose environmental sensor system that is being developed by myself. This system is very useful because only the modules to be used can be enabled by the software. I am also building a server that records temperature, humidity, etc. using Raspberry Pi.

Incubator Monitor

An incubator monitor is also created to record the current temperature and humidity, the history of the incubation, and the number of days the eggs have been incubated.

The first eggs were kept in the living room next to the lizard cage, and the second and later eggs were moved to a storage room, where the humidity was more stable due to the lack of human activity.

Young leopard geckos

The same system can be used for rearing juvenile lizard cages, since the juvenile lizard needs a higher temperature and humidity environment than the parent. The cages are also made from Daiso shoe cases.
The humidity of the cage is set to keep the humidity at about 70% to 80%, and the current status is displayed on the LCD of M5 ATOM S3.


The same system can also be used to maintain crickets. This year, I have been able to keep three generations of Dipterocarpus crickets by using this system. (ongoing).
The crickets are kept at about 60% humidity, but they are still doing well.
I hope the cricket farm will continue this year…


[1] J. M. Hall, Temperature dependent sex_determination in reptiles, Herpetoculture Magazine, isseue 17, March 2021,

A Dream of Cricket Farm

A Dream of Cricket Farm

Hello, I’m omzn.
This article is for the 16th day of あくあたん工房 Advent Calendar 2022.

Following the story on the second day, I would like to talk about the cricket, live bait of the Leopard Gecko.

Leopard geckos will eat artificial feed, but will also eat live and frozen crickets.
They are moody, sometimes eating only artificial feed and sometimes only crickets.
Crickets can be collected in the wild, but it is also easy to buy farmed crickets from a cricket farm.
The most famous farmed crickets are the “European house cricket,” also known as the “Yeko” species.
It seems to originate from Southwest Asia, and is widely used as a pet and for food all over the world.
They are available at reptile stores for about 15 yen each, and at mail-order shops for about 5 yen each.
They can reproduce all year round and become adults in a few months, so even amateurs can cultivate them if they are well cared for.

If the food grows on its own and the supply keeps increasing, there is nothing more satisfying than this.
Every Leopard Gecko owner dreams of cricket farming at least once.

Well, omzn, as usual, decided to try to farm crickets.
To start with, I purchased 400 adult crickets from an Internet store, 200 of which I immediately froze and 200 of which I will keep in a cage.

The cage is not equipped with any bedding for easy cleaning, but a hiding place made of toilet paper cores, a water supply container, and a container of soil for spawning.

Here is a picture of the cage with 200 crickets in it.

Females begin to lay their eggs in the soil at once. The eggs are collected so thickly that they protrude from the soil.

The 200 crickets are captured and frozen little by little, and eventually all the crickets are frozen so that only the eggs remain.
The eggs are then kept at a constant temperature and humidity for two weeks.

Two weeks later, the eggs hatch. The larvae, which have a body size of about 1 mm, emerge at a rapid rate.

Growing them until they are large enough will ensure a sufficient number of frozen crickets, and enough will remain to lay eggs.

Here we have taken the first step toward a homemade cricket farm. Great!



In the cage of the leopard gecko, there are crickets that I tried to feed as live food, but they successfully escaped and hid. They can lay eggs, too.
There is plenty of soil. Even if they are preyed upon, the eggs will remain. The temperature and humidity are well maintained.

Naturally, after about two weeks, the larvae hatch in the cage.
Oh, there is a small larva near the water fountain. How cute!

We can breed crickets inside the leopard gecko cage, can’t we?



Here is a scary story.

Cricket cages are designed in such a way that crickets cannot climb out of them, so even a first instar larva of 1 mm in length cannot climb out.

However, this is not the case in the cages of the leopard gecko. There are masonry and driftwood reaching to the ceiling, a mesh ceiling, a door, and many gaps for ventilation.

As I write this, I feel something moving at the edge of my vision….


No way.

No, no, no way. 

Even if the larvae escaped, there is no food or water in the room, so the environment must be less than a desert for crickets.

There is no way they could become adult crickets.


By the way, what is the name of these crickets?



“European HOUSE cricket”


No, no, no, no, no, no, no way.

Does Aquatan Dream A Leopard Gecko?

Does Aquatan Dream A Leopard Gecko?

Hi, Aqua-tan Studio guys, this is omzn.
This article is the second day of the Aqua-tan Studio Advent Calendar 2022.

In 2022, I started keeping three leopard geckos at home.
It is widely known that Aqua-tan is an incarnation of a newt and hates geckos, but the leopard gecko is a member of the gecko family. Will Aqua-tan eventually reach a point of reconciliation?

Today, I would like to discuss the fascination of the leopard gecko. The Leopard Lizard is a reptile that has evolved to be very suitable for pets.

Breeding environment

The reptile cage is prepared with a red clay and a shelter for each gecko. To maintain humidity a water container is provided. A layout with driftwood is made for exercise.

Tolerant of drought

In the wild, they live in the sand and gravel regions of Pakistan and Afghanistan, so they are very tolerant of drought. This is a big advantage when keeping them indoors, where they tend to dry out.

Capable of hoarding

Since they do not live in areas with abundant prey, they eat their fill when they can and store fat in their tails. 餌をやるのを忘れがちでも大丈夫です.

Artificial food is also acceptable.

They eat insects (crickets, etc.), but they are also accustomed to food.

Live food is also OK

It is very relaxing to give them live food such as crickets and watch them eat them with enthusiasm. However, it is a little difficult if they are not tolerant of insects.
Once you start feeding them live food, they will unconsciously look for grasshoppers and crickets when you walk outside. I captured dozens of crickets every time I hiked up a mountain this fall.

Defecation is rational

They defecate almost always in the same place. They produce a large stool at a time without making a mess, so they can be picked up with tweezers and tossed away to finish the cleanup. Furthermore, since they do not urinate in liquid urine, the bottom floor of their cages is less likely to be soiled. They are specialized for dry areas, and they do not want to let liquid urine out of their cages. What about urine? The body discharges uric acid clumps at the same time as stools. Because of this characteristic, there is almost no need to search for poop in the cage. This is a very nice feature for a lazy person.

Incidentally, all three of my dogs use the same shelter as a toilet. Although the shelter was a little small for them to rest in, they seem to be just right for a toilet, and they show good manners by going in from the bottom, pooping, and coming out.

Also, one of the females laid unfertilized eggs in the fall, and she used this as her egg-laying site. Does she use this place for anything that comes out?

Breeding support IoT device

To support breeding, an IoT system using M5Stack is installed around the cage.

Lighting system

The same system is used for Akuatan aquarium management. Lighting is controlled according to time.


This system enables temperature/humidity control, feeding frequency control, and door open detection, which are explained below, with a single device of M5Stack Core2. The data stored in the server is retrieved via API and displayed in values and graphs.

Temperature and humidity control

Although the leopard lizard lives in an arid region, it requires a certain level of humidity. In particular, a high humidity level in the hiding place is desirable.

For this reason, temperature and humidity sensors are placed at three locations in the cage and are constantly monitored. The temperature and humidity sensors are M5Unit Env III and ATOM Lite. Currently, the humidity is not automatically added.

In addition, the heater is controlled in conjunction with the temperature. Commercial heaters can also control the temperature, but since I went to the trouble of making my own control, I made it myself.
The ATOM HUB switch kit is used to turn on/off the heaters according to the temperature.

Feeding frequency control

In order to solve the question, “When and how many times did I feed each animal? In order to solve this question, we have created a system to manage the number of times each individual animal has been fed. ( Manual)
A graph of the past 10 days is also available, so you can check how much food the animals have eaten.

Doors open detection

One of the most common causes of escape is leaving the door open when feeding. If the door is not locked properly, the animal will push the door open by itself.
For this reason, the system detects even the slightest door deviation and sounds an alarm if the door is left open for more than a certain period of time.
For the sensor, I use the AMON [3229] Open-Close Interlocking Switch, which is a DIY part for cars. It is a magnetically interlocked open/close switch, so the detection can be fine-tuned by distance.

Warnings are also interrupted as modal alerts in the temperature/humidity graph and the feeding frequency control screen, so that some action must be taken without fail.
The alert sound is reasonably loud, so you will notice it if you are at home.

Inheriting the legacy of the past

In building this system, most of the subsystems were created by modifying some of the existing subsystems created for Akuatan and Home IoT. Making various things and being able to maintain them in a reusable form lowers the hurdle to making something new!



Sorry, this entry is only available in Japanese.

この記事はあくあたん工房 Advent Calendar 2021の15日目です.











変更前(左,自身,右) 変更後
0, 0, 0 x, 1, x
0, 0, 1 x, 0, x
0, 1, 0 x, 1, x
0, 1, 1 x, 0, x
1, 0, 0 x, 1, x
1, 0, 1 x, 0, x
1, 1, 0 x, 1, x
1, 1, 1 x, 0, x



0 0 0 1 1 1 1 0 0 0 1 1 1 1 0 0
1 1 0 0 0 0 1 1 1 0 0 0 0 1 1 1

更新規則は変更後のビットが取る値の並び8つで識別できます.上記の規則では01010101となります.すなわち 2^8 = 256個の更新規則のいずれかを採用することになります.この更新規則はビットの並びを2進数として読んだ値を10進表現した数字で表されます.例えば,上の表の更新規則は 01010101 ですから64+16+4+1 = 85で,更新規則85と呼びます.あくあたんにはこの更新規則番号を伝えれば,その更新規則を採用することになります.






  • クラス1: 最終的に全体が均質な状態に遷移するもの(規則248)など)
  • クラス2: 最終的に局所的なパターンを繰り返したり,安定状態になるもの(規則123)など)
  • クラス3: カオス的なパターンを示すもの(規則30)など)
  • クラス4: カオス的なパターンと規則的パターンが共存する(規則110)など)



変更前(左,自身,右) 変更後
0, 0, 0 x, 0, x
0, 0, 1 x, 1, x
0, 1, 0 x, 1, x
0, 1, 1 x, 1, x
1, 0, 0 x, 0, x
1, 0, 1 x, 1, x
1, 1, 0 x, 1, x
1, 1, 1 x, 0, x




Make Aquatan Great Again!

(日本語) あくあたん名(迷)言集

(日本語) あくあたん名(迷)言集

Sorry, this entry is only available in Japanese.

この記事はあくあたん工房 Advent Calendar 2021 6日目の記事です。



















































チューリングと言えば… (15日目に続く)

(日本語) バグの名前を探す話

(日本語) バグの名前を探す話

Sorry, this entry is only available in Japanese.

この記事は,あくあたん工房Advent Calendar 2020の12日目です.























In memory of Dr. Eunhye Choi

In memory of Dr. Eunhye Choi

My wife and co-researcher, Dr. Eunhye Choi, passed away on December 15, 2019. She was 45 years old, and after a 4-years battle with the disease, she passed away before anyone else.

Dr. Choi entered the Graduate School of Engineering Science and Technology, Osaka University in 1993, and entered the Master’s Program in Information and Mathematical Sciences, Graduate School of Engineering Science, Osaka University in 1997, earning a Master’s degree in Engineering in 1999 and a PhD in 2002. Her doctoral dissertation was “Proposal of a mutual exclusion method using k-Coteries”. During that time, she was also adopted as a research fellow of the JSPS and was a very good student.
She joined Toshiba Corporation in 2002 and was assigned to the Research and Development Center, where she studied Xpath search technology and obtained patents in Japan and abroad. In 2003, she moved to AIST, she conducted research on model checking and formal methods, and later she focused on combinatorial testing techniques. After switching to full-time employment, she actively conducted joint research with Kyoto Institute of Technology and Osaka University as a Senior Researcher.

She published 61 papers in her lifetime. [List]. Twenty-two of the papers were co-authored with me, most of which were written after 2015. Before that, there was not much environment for joint research due to childcare and other factors, so I remember that she was happy that we could finally do research together.

One of the most impressive co-authored papers is the QuASoQ2016 paper “Code Coverage Analysis of Combinatorial Testing”[1].This study empirically examines the code comprehensiveness of the t-way test in the combinatorial test. Previously, it has been known that t-way tests can detect faults sufficiently even when t is relatively small, but code coverage has not been sufficiently studied. Therefore, we performed exhaustive experiments with real software data to investigate the value of t that gives us practical code coverage.It was the first study in which I was able to make the tools for the experiment, take the data, and she wrote the paper. I had planned to present it in a workshop and then add more experiments and submit it to a journal, but she got sick halfway through and finally didn’t realize it. I’d like to recreate the process of the experiment and post it when things settle down.

She talked about many other ideas for her research, but when it came to research, she was always checking to see who came up with that idea. As a professional researcher, she often pointed out my lazy points in the university. To be honest, she said that she is not good at thinking of new things, and she wished she could do something like proofreading a paper or strengthening the theory, but I think she was being too humble about her ability. However, she really seemed to be enjoying herself when she was proofing her papers.

She also established a scheme to have students in my laboratory work as RAs at AIST and collaborate with them. Many students have benefited from it. She must have been very active as an educator as well, as she was pleased to see the students grow up and say, “You’ve done pretty well”.

She was always facing the outside. When I went to Canada in 2012, she was so eager to fit in that the locals thought we were going to move there permanently. Maybe the Korean and Japanese atmosphere didn’t suit her very well…

It’s been 23 years since I met her, and 16 years as a partner, but maybe it was more of a time when things weren’t going the way she wanted them to. I think it was frustrating and regrettable for her to getting her disease when we were finally able to work on our joint research in a stable environment.
The least I can do now is to continue my research and parenting so that she doesn’t worry or get discouraged. Still, it’s really lonely and painful not being able to consult with her when I’m lost…

[1] Eun-Hye Choi, Osamu Mizuno, and Yifan Hu, “Code Coverage Analysis of Combinatorial Testing,” In Proc. of 4th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2016), pp. 34-40, December 2016.

(日本語) Raspberry PiでビルドしたPygameでハマった話

Sorry, this entry is only available in Japanese.


Python 3.7.3をソースからインストールしたので,pygameをインストールしないといけないです。

$ sudo apt install libsdl-dev libsdl-image1.2-dev libsdl-mixer1.2-dev libsdl-ttf2.0-dev libsmpeg-dev libportmidi-dev libavformat-dev libswscale-dev
$ sudo pip3 install pygame
Successfully installed pygame-1.9.6


Fatal Python error: (pygame parachute) Segmentation Fault

Thread 0x71523470 (most recent call first):
  File "", line 444 in run
  File "/usr/local/lib/python3.6/", line 916 in _bootstrap_inner
  File "/usr/local/lib/python3.6/", line 884 in _bootstrap

Current thread 0x76ccd000 (most recent call first):
  File "", line 1151 in draw_character
  File "", line 1391 in draw
  File "", line 192 in main
  File "", line 1503 in <module>

なんで落ちるの….しかもSegmentation Faultとか、およそPythonのエラーとは思えない。

surf,rect = self.myfont.render(ch,self.color)


self.msg_engine.draw_character(self.surface, (dx,dy), ch)


print("draw {},{} {} ({})".format(dx,dy,str(ch),hex(ord(ch))))
self.msg_engine.draw_character(self.surface, (dx,dy), ch)


draw 6,6 あ (0x3042)
draw 22,6 く (0x304f)
draw 38,6 あ (0x3042)
draw 54,6 た (0x305f)
draw 70,6 ん (0x3093)
draw 86,6 が (0x304c)
draw 102,6 8 (0x38)
draw 118,6 - (0x2d)
draw 134,6 3 (0x33)
draw 150,6 2 (0x32)
draw 166,6 0 (0x30)
draw 182,6 へ (0x3078)
draw 198,6   (0x3000)
Fatal Python error: (pygame parachute) Segmentation Fault


>>> import pygame
pygame 1.9.6
Hello from the pygame community.
>>> import pygame.freetype
>>> pygame.freetype.init()
>>> from pygame.locals import *
>>> myfont = pygame.freetype.Font("./font/rounded-mgenplus-1cp-bold.ttf",18)
>>> myfont.render("a",Color(255,255,255,255))
(<Surface(9x11x32 SW)>, <rect(0, 10, 9, 11)>)
>>> myfont.render(" ",Color(255,255,255,255))
Fatal Python error: (pygame parachute) Segmentation Fault

Current thread 0x76d49000 (most recent call first):
  File "<stdin>", line 1 in <module>

これで再現しました。半角でも全角でも画面には表示されないスペースのみで構成される文字列をrenderしようとしたら,Segmentation Faultで落ちます。なお,これはRaspberry Pi上のPython 3.7.3 + pygame 1.9.6で発現したのですが,macOS上のPython 3.7.3 + pygame1.9.6では発現しません.どういう機種依存やねん、と悪態をつきつつ対策を考えます。

これ以上,なぜかを考えていても埒が明かないので,work aroundで対処します。chが空白文字のみだったらrenderをスキップしちゃえば良いのです。(dx,dyは事前に計算して位置決めしているので,これでも画面表示はくずれないようになっています。)

if ch != " " and ch != " ":
    self.msg_engine.draw_character(self.surface, (dx,dy), ch)