lobby Raspberry Pi not booting

We had multi-day issues booting the Raspberry Pi in the lobby, which would play a video loop. Disconnecting the USB power and re-connecting it, or taking out the micro-sdcard, checking it for errors on Linux Mint and then re-inserting it would make it work.

Suspecting power supply issues, the current idea is to

1. add a separate switch to turn on power to the Pi after the TVs are powered on.

2. change the power cord as well as the USB power adapter

With these two changes, the Pi booted up without trouble today, we would probably need to monitor it for a week or so to be sure that this has solved the problem.

visuals on planetarium dome getting washed out

Copy-pasting from an email thread - 

Just a follow-up on the issues with washed-out visuals we were facing.

When slowly fading in particular scenes, we can clearly see a point where parts of the scene get washed out, with the projector suddenly jumping in brightness. Looks like this projector, Optoma ZK-507-W, has some poorly executed "high-brightness" handling, which we don't seem to be able to disable. As long as we keep large areas below 50% brightness, small bright areas like stars, trails etc can be viewed with good contrast. 

I think it is some sort of local dimming similar to what is used in some TVs, https://www.academia.edu/88311604/Backlight_local_dimming_algorithm_for_high_contrast_LCD_TV but with very poor results.

Perhaps a warning line can be added at https://paulbourke.net/dome/faq/#projectors for this projector, indicating its poor performance.

To which Paul replied,

I  assume you have tried

1. Turning off the DynamicBlack settings in Display ->  Image setting -> Brightness mode menu.

2. Turning off the HDR/HLG in Display -> Image settings -> Dynamic range menu.

Yes for the 2nd, there was no such menu for the first point. Since my source is not HDR, I also tried "Auto" in addition to "Off".

Also, have tried all the different modes, including "Film" - even that has the noticeable brightness jump once a large enough area has a pixel value high enough (perhaps >50%). 

(Interestingly, some planetarium shows have more problems than others.

Eg. Unseen Earth, https://www.unseenearth.eu/ - has some drone shots etc which look terrible on our dome if I don't mask out most of it to <50%

Mars - the Ultimate Voyage - https://www.bellmuseum.umn.edu/mars-ultimate-voyage/ - has gamma settings, lighting etc such that most of the show looks pretty good on our dome.

In our own productions using OpenSpace, https://www.openspaceproject.com/ , there seems to be a difference between scenes rendered on Mac and scenes rendered on PC (Linux). Maybe gamma differences.

These micro-organisms are rendered fine, while I had to darken these scenes rendered on Mac for Jupiter, Saturn and Europa to not become completely washed out.

Interestingly, this scene, rendered on Linux, is fine.)

OpenSpace builds on Github actions getting "terminated" - exit code 143

Some of my build github action workflows were consistently getting terminated after 10 minutes of run time or so. 

gmake[2]: *** Waiting for unfinished jobs....

gmake[2]: *** wait: No child processes.  Stop.

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3607: ext/spice/CMakeFiles/spice.dir/src/common/dafopw_c.c.o] Terminated

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3635: ext/spice/CMakeFiles/spice.dir/src/common/dafps_c.c.o] Terminated

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3663: ext/spice/CMakeFiles/spice.dir/src/common/dafrcr.c.o] Terminated

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3649: ext/spice/CMakeFiles/spice.dir/src/common/dafra.c.o] Terminated

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3691: ext/spice/CMakeFiles/spice.dir/src/common/dafrda_c.c.o] Terminated

gmake[2]: *** [ext/spice/CMakeFiles/spice.dir/build.make:3565: ext/spice/CMakeFiles/spice.dir/src/common/dafgsr_c.c.o] Terminated

gmake[2]: *** [modules/globebrowsing/ext/geos/CMakeFiles/geos.dir/build.make:2514: modules/globebrowsing/ext/geos/CMakeFiles/geos.dir/src/io/StringTokenizer.cpp.o] Terminated

Error: Process completed with exit code 143.

Gemini - via aistudio.google.com free plan - pointed out that 

Exit code 143 indicates that a process was terminated by a SIGTERM signal. In GitHub Actions, this almost always happens when the build runner runs out of memory (OOM) and the operating system or the GitHub runner agent terminates the compiler processes to save the system.

Compiling large C++ projects like OpenSpace with heavy submodules (such as Assimp/Draco, Boost, and Spicelib) using GCC 13 can easily exceed the 7 GB RAM limit of the standard GitHub-hosted runner, especially when building in parallel.

Here are the two most effective ways to resolve this:

1. Limit Build Parallelism

By default, cmake --build . --parallel attempts to use all available virtual CPUs. While GitHub runners have 2 vCPUs, compiling templates in parallel can still exceed 7 GB of RAM.

Limit the concurrent build jobs to 2 etc etc

So I changed 
 cmake --build . --parallel

to
 cmake --build . --parallel 3

like the workflow which was working, and that solved the problem. Told Gemini that, and the reply was,

You are correct. Standard GitHub-hosted runners for public repositories were updated to 4 vCPUs and 16 GB of RAM.

When you run cmake --build . --parallel without specifying a limit, it attempts to spawn 4 compilation jobs. With heavy C++ headers and templates, 4 concurrent GCC processes can still spike past the 16 GB memory limit, triggering the termination.

Limiting the build to --parallel 3 (or --parallel 2 if memory usage remains high) is a solid approach to keep the peak memory consumption within the runner's limits.

Then I asked Gemini,

Same codebase, same runner.
cmake --build . --parallel causes termination.
make -j $(nproc) does not cause termination, why?

Then Gemini changed the answer to specify that parallel without a number means unlimited parallel jobs - 

The difference lies in how GNU Make and CMake interpret the omission of a job

number.

1. cmake --build . --parallel maps to unlimited jobs

When you run cmake --build . --parallel without specifying a number, CMake

translates this to the underlying build tool's default parallel flag.

For GNU Make, this translates to running make -j (with no number). In GNU Make,

omitting the number instructs it to run with infinite/unlimited parallel jobs.

Instead of spawning 4 compiler processes (the number of CPU cores), it will try

to compile dozens of source files simultaneously. This immediately exhausts

the 16 GB of RAM and triggers the OOM killer.

2. make -j $(nproc) strictly limits the jobs

When you run make -j $(nproc):

1.  $(nproc) evaluates to 4.

2.  The command becomes make -j 4.

3.  GNU Make is strictly limited to running a maximum of 4 compiler processes at

    any given time, which usually fits within the 16 GB RAM limit.

Solution

If you want to use the CMake CLI wrapper safely, you should explicitly pass the

CPU count to it, similar to what you did with make:

cmake --build . --parallel $(nproc)

Moodle course order - manual sorting option vanished

One of our Moodle instances had the manual sorting option for courses disappear, at server/course/management.php

ChatGPT gave some hallucinatory suggestions, Gemini via aistudio.google.com gave the first suggestion of removing pagination by using the dropdown to show all courses instead of only the first 20 - the number of courses had become 21. Even after doing this, the sorting option did not return. But once I added a dummy category and some dummy courses and then deleted them, the sorting option returned. Gemini said, "When you created the dummy category and dummy courses, and then deleted them,

you triggered a core Moodle background script called fix_course_sortorder()." So that has fixed it.

Reading up on this function, it looks like for very early versions, this function was also buggy - https://moodle.org/mod/forum/discuss.php?d=126896 - but in current versions, this seems to be rewritten, like in the current main branch.

msmtp instead of mail

Noticed only today that for some scripts on our internal server, the mail command was commented out and replaced by msmtp like 

#mail -s "Automatic Encoded file Statistics for : date" -t $MAIL_COPY $MAIL_DESTINATION < $MAIL_FILE 

{ echo "To: $MAIL_DESTINATION"; echo "Cc: $MAIL_COPY"; echo "Subject: Automatic Encoded file Statistics: $(date)"; echo; cat "$MAIL_FILE"; } | /usr/bin/msmtp -t

ChatGPT explains,

Yes, on many modern Linux systems, using msmtp directly is now preferred over the traditional mail command, especially for scripts and automated systems.

The reason is that the mail command historically depended on a local Mail Transfer Agent (MTA) such as:

Sendmail

Postfix

Exim

Many newer minimal/server/container installations no longer install or configure a full local MTA.

This was mentioned in this previous post, which mentions a replacement of ssmtp with msmtp. 

UTM and the mouse

When switching between machines using a KVM switch, found that Ubuntu 24.04 running on UTM on Mac, full screen, would ask, "Allow UTM to use the mouse" or something like that. We should choose cancel instead of Yes. If we choose Yes, the mouse cursor would not be seen, though mouse-clicks would be registered correctly on Ubuntu 24.04. If we choose cancel, the mouse would work properly.

two or three ways to make LLMs help with Blender scene creation

If we directly ask Gemini to create a Blender file, it says it can't, but it can generate a python script which can be run inside Blender to create a scene. This is useful if we need to generate repetitive visuals or something which can be programmatically generated. Example, CirclesVideos.blend - the prompt for Gemini 3, via the free tier of aistudio.google.com, is below.

Can you now generate a python script to create a blender scene with

1. Fisheye 180 degree equidistant camera in the centre, pointing straight up

2. equidistant circular screens subtending just less than 40 degrees so that they do not touch, having a small gap between them, 9 circular screens nearest the equator, and more rows of similarly sized circular screens, but fewer as we go up the hemisphere since the circumference of higher latitudes would be lower

3. The lowest row of circular screens must be completely above the horizon for the viewport of the 180 degree fisheye equidistant camera

4. The screens should fade in one by one starting from the lowest row, left to right, sequentially, one after the other, with a one second fade in time. So, circle one fades in by the 00:00:01 mark, then circle two fades in and finishes fading in by the 00:00:02 mark, and so on. The total duration of the animation can be one second greater than the time required to fade in all the circles.

5. Each circular screen should have an image texture mapped to a unique file - named seq01.jpg for circle one, seq02.jpg for circle two and so on, with the jpg textures located in a subdirectory of the project directory called seq. The image texture would be square in shape, but should be zoomed in so that the entire circle is filled with the texture.

This resulted in the following:

Here is the complete Python script to generate your exact fulldome scene.

It handles the math for perfectly spacing the rows along the hemisphere without

overlapping, scales the circular cutouts to exactly 38 degrees (leaving

a 2-degree gap so they don't touch), sets up the sequential 1-second fade-ins,

and auto-links your seq/seqXX.jpg files.

How to use it:

1.  Open a new Blender file.

2.  Go to the Scripting tab.

3.  Click New, paste the code below, and click the Run Script (Play) button.

4.  Set your 3D Viewport to Rendered View to see the fade-in math and image

    textures working perfectly.

import bpy

import math

# ==========================================

# 1. CLEANUP PREVIOUS SCENE

# ==========================================

bpy.ops.object.select_all(action='SELECT')

bpy.ops.object.delete(use_global=False)

# ==========================================

# 2. SCENE & ANIMATION SETTINGS

# ==========================================

FPS = 24

TOTAL_SCREENS = 14

FADE_DURATION = FPS * 1 # 1 second fade in per screen

scene = bpy.context.scene

scene.render.engine = 'CYCLES'

scene.render.resolution_x = 2048

scene.render.resolution_y = 2048

scene.render.fps = FPS

# Total duration = 1 second per screen, plus 1 second at the end

TOTAL_FRAMES = (TOTAL_SCREENS * FADE_DURATION) + FADE_DURATION

scene.frame_start = 1

scene.frame_end = TOTAL_FRAMES

# ==========================================

# 3. FULLDOME CAMERA (Equidistant)

# ==========================================

cam_data = bpy.data.cameras.new("Fulldome_Camera")

cam_data.type = 'PANO'

cam_data.panorama_type = 'FISHEYE_EQUIDISTANT'

cam_data.fisheye_fov = math.radians(180)

cam_obj = bpy.data.objects.new("Fulldome_Camera", cam_data)

bpy.context.collection.objects.link(cam_obj)

scene.camera = cam_obj

# Camera at center (0,0,0) pointing straight UP (Zenith)

cam_obj.location = (0, 0, 0)

cam_obj.rotation_euler = (math.radians(180), 0, 0)

# Dark Background

world = scene.world

world.use_nodes = True

bg_node = world.node_tree.nodes.get("Background")

if bg_node:

    bg_node.inputs[0].default_value = (0.01, 0.01, 0.02, 1)

# ==========================================

# 4. MATH: SCREEN PLACEMENTS

# ==========================================

# Dome Radius

R = 10.0

# To subtend exactly 38 degrees (leaving a gap under 40deg),

# the physical radius of the plane at distance R is R * tan(19 degrees)

screen_radius = R * math.tan(math.radians(19))

# Calculate positions (Elevation and Azimuth)

# Row 1: 9 screens at 20 degrees elevation. 

# (Bottom edge is 20 - 19 = 1 degree above horizon. Perfect!)

# Row 2: 5 screens at 60 degrees elevation.

# (Bottom edge is 60 - 19 = 41 degrees. Top of Row 1 is 39. Gap of 2 degrees!)

screen_data = []

# Row 1 (9 Screens, lowest row)

for i in range(9):

    azimuth = i * (360.0 / 9.0)

    screen_data.append({'el': 20.0, 'az': azimuth})

# Row 2 (5 Screens, upper row)

# Offset azimuth slightly so they interlock beautifully

for i in range(5):

    azimuth = i * (360.0 / 5.0) + 36.0 

    screen_data.append({'el': 60.0, 'az': azimuth})

# ==========================================

# 5. MATERIAL GENERATOR

# ==========================================

def create_fade_circle_material(index, start_frame, end_frame):

    mat_name = f"Mat_Screen_{index:02d}"

    mat = bpy.data.materials.new(name=mat_name)

    mat.use_nodes = True

    mat.blend_method = 'CLIP' # Helps viewport transparency

    nodes = mat.node_tree.nodes

    links = mat.node_tree.links

    nodes.clear()

    # Base Nodes

    output = nodes.new(type='ShaderNodeOutputMaterial')

    output.location = (1000, 0)

    # Fade Setup (Mix Shader 2)

    fade_mix = nodes.new(type='ShaderNodeMixShader')

    fade_mix.location = (800, 0)

    

    fade_trans = nodes.new(type='ShaderNodeBsdfTransparent')

    fade_trans.location = (600, 100)

    # Cutout Setup (Mix Shader 1)

    cutout_mix = nodes.new(type='ShaderNodeMixShader')

    cutout_mix.location = (600, -100)

    

    cutout_trans = nodes.new(type='ShaderNodeBsdfTransparent')

    cutout_trans.location = (400, 0)

    # Video/Image Setup

    emission = nodes.new(type='ShaderNodeEmission')

    emission.location = (400, -200)

    emission.inputs['Strength'].default_value = 1.5

    img_tex = nodes.new(type='ShaderNodeTexImage')

    img_tex.location = (100, -200)

    # Assuming relative path in a "seq" folder next to your .blend file

    img_path = f"//seq/seq{index:02d}.jpg"

    

    # Try to load the image if it exists, otherwise leave placeholder

    try:

        img = bpy.data.images.load(img_path)

        img_tex.image = img

    except:

        print(f"Warning: Could not load {img_path}. Placeholder created.")

    # Math for the Circle Mask

    tex_coord = nodes.new(type='ShaderNodeTexCoord')

    tex_coord.location = (-300, 0)

    distance = nodes.new(type='ShaderNodeVectorMath')

    distance.operation = 'DISTANCE'

    distance.location = (-100, 0)

    distance.inputs[1].default_value = (0.5, 0.5, 0.0)

    less_than = nodes.new(type='ShaderNodeMath')

    less_than.operation = 'LESS_THAN'

    less_than.location = (100, 0)

    less_than.inputs[1].default_value = 0.5

    # Wire it all up

    links.new(tex_coord.outputs['UV'], distance.inputs[0])

    links.new(distance.outputs['Value'], less_than.inputs[0])

    

    links.new(tex_coord.outputs['UV'], img_tex.inputs['Vector'])

    links.new(img_tex.outputs['Color'], emission.inputs['Color'])

    links.new(less_than.outputs['Value'], cutout_mix.inputs['Fac'])

    links.new(cutout_trans.outputs['BSDF'], cutout_mix.inputs[1])

    links.new(emission.outputs['Emission'], cutout_mix.inputs[2])

    links.new(fade_trans.outputs['BSDF'], fade_mix.inputs[1])

    links.new(cutout_mix.outputs['Shader'], fade_mix.inputs[2])

    

    links.new(fade_mix.outputs['Shader'], output.inputs['Surface'])

    # --- ANIMATE THE FADE ---

    # At start of animation: completely transparent

    fade_mix.inputs['Fac'].default_value = 0.0

    fade_mix.inputs['Fac'].keyframe_insert(data_path="default_value", frame=1)

    

    # Just before its turn: still transparent

    fade_mix.inputs['Fac'].default_value = 0.0

    fade_mix.inputs['Fac'].keyframe_insert(data_path="default_value", frame=start_frame)

    # 1 second later: fully visible

    fade_mix.inputs['Fac'].default_value = 1.0

    fade_mix.inputs['Fac'].keyframe_insert(data_path="default_value", frame=end_frame)

    return mat

# ==========================================

# 6. SPAWN SCREENS

# ==========================================

for i, data in enumerate(screen_data):

    el_rad = math.radians(data['el'])

    az_rad = math.radians(data['az'])

    

    # Convert polar to Cartesian XYZ

    z = R * math.sin(el_rad)

    r_xy = R * math.cos(el_rad)

    x = r_xy * math.cos(az_rad)

    y = r_xy * math.sin(az_rad)

    

    # 1. Create the Plane

    bpy.ops.mesh.primitive_plane_add(size=1.0, location=(x, y, z))

    plane = bpy.context.active_object

    

    # Index is i+1 (so it starts at 1)

    idx = i + 1

    plane.name = f"Screen_{idx:02d}"

    

    # 2. Scale to exact angular size (38 degrees)

    # We multiply by 2 because screen_radius is from the center to edge.

    plane.scale = (screen_radius * 2, screen_radius * 2, 1.0)

    

    # 3. Add Track To Constraint (Always face camera)

    track = plane.constraints.new(type='TRACK_TO')

    track.target = cam_obj

    track.track_axis = 'TRACK_Z'

    track.up_axis = 'UP_Y'

    

    # 4. Create and assign animated material

    start_f = (i * FADE_DURATION) + 1

    end_f = start_f + FADE_DURATION

    

    mat = create_fade_circle_material(idx, start_f, end_f)

    plane.data.materials.append(mat)

print(f"Fulldome scene generated! Total screens: {TOTAL_SCREENS}. Animation frames: {TOTAL_FRAMES}")

---

LLMs can even give us alternate ideas. For example, this conversation, 

Prompt for Gemini,

I'd like to show a montage of videos in Blender to be rendered for fulldome (180

degree fisheye equisolid camera). The script is "We constantly seek happiness in

the form of pleasure, peace, or the satisfaction of curiosity." My current idea

is to depict a pyramid with video textures on the four sides depicting

"pleasure, peace, or the satisfaction of curiosity." Any other ideas?

Which resulted in this reply - 

Designing for a fulldome (180° fisheye) is a completely different beast than

designing for a flat screen.

Your pyramid idea is conceptually strong, but in a fulldome environment, a

single object floating in the center of the screen can sometimes feel a bit

"small," leaving the vast periphery of the dome empty. In fulldome, you want to

surround the viewer or give them a sense of scale and immersion.

Here are a few alternative concepts specifically tailored to make the most of

the 180° fisheye camera in Blender, incorporating your video montages and

script.

1. The Canopy of Spheres (Focus on "Looking Up")

  - The Visual: The camera sits at the base of a massive, glowing, abstract tree

    or DNA-like structure that stretches up to the "zenith" (the dead center of

    your fisheye circle, which is the top of the dome). Instead of leaves,

    giant, slow-moving translucent bubbles or glass orbs hang above the viewer.

  - The Montage: Inside these orbs are your videos.

      - Pleasure videos can be in warm, rapidly pulsing orange/red orbs.

      - Peace videos in slow-spinning, tranquil blue/teal orbs.

      - Curiosity videos in geometric, bright white/purple orbs.

  - Why it works for Fulldome: Audiences in planetariums naturally look up. By

    placing the anchor of the structure at the bottom edge of the fisheye (the

    springline) and having it branch out overhead, you create a massive sense of

    scale and immersion.

2. The Monolith Portals (Surrounding the Viewer)

  - The Visual: Instead of looking at a pyramid from the outside, place the

    camera inside a massive, open space (like a serene desert, a starry void, or

    a calm ocean). Standing out of the ground, surrounding the camera in

    a 360-degree radius (or semi-circle), are three monolithic, towering

    screens/portals.

  - The Montage: As the narrator speaks, the camera slowly rotates.

      - "Pleasure" lights up the monolith to the left, reflecting its video off

        the glossy floor.

      - "Peace" lights up the center monolith.

      - "Curiosity" lights up the right monolith.

  - Why it works for Fulldome: It places the viewer inside the geometry rather

    than watching it from afar. Wide, towering objects that wrap around the

    peripheral vision (the edges of the fisheye circle) look incredibly imposing

    and immersive in a dome.

3. The Tunnel of Seeking (Forward Motion)

  - The Visual: "Constantly seek" implies forward momentum. The camera is moving

    slowly and continuously forward through a massive, abstract, hexagonal or

    circular tunnel.

  - The Montage: The walls of the tunnel are made of floating video panels.

      - As the word "Pleasure" is spoken, panels whiz by showing vibrant,

        sensory clips.

      - For "Peace," the tunnel widens, the movement slows slightly, and the

        panels show calming nature or meditative clips.

      - For "Curiosity," the tunnel breaks apart into a constellation of

        floating screens showing science, exploration, or complex data.

  - Why it works for Fulldome: Forward motion (called vection) is the single

    most powerful optical illusion in a fulldome. A slow, steady forward push

    through a tunnel makes the audience feel like they are physically moving

    through space.

4. The Fluid Kaleidoscope (Abstract Blending)

  - The Visual: Instead of harsh geometric shapes (like a pyramid), map your

    videos onto fluid, undulating mesh surfaces—like ribbons of silk or waves of

    water—that drift right over the camera lens.

  - The Montage:

      - A vibrant ribbon of video (Pleasure) swoops across the dome.

      - It is intercepted by a slow-moving, massive, calm wave of video (Peace)

        that washes over the whole screen.

      - Finally, a sharply focused, fractal-like geometric burst of video

        screens (Curiosity) shatters through the peaceful wave.

  - Why it works for Fulldome: Sharp edges and straight lines (like a pyramid)

    are heavily curved and distorted by fisheye lenses. Organic, fluid shapes

    hide this distortion and feel incredibly natural and dreamlike in a dome.

5. If You Stick With the Pyramid: The Interior Pyramid

If you really love the pyramid motif, flip the perspective. Place the camera

inside a colossal pyramid.

  - The Visual: The camera looks up toward the apex of the pyramid (which sits

    in the center of your fisheye render). The walls of the pyramid surround the

    viewer.

  - The Montage: The three visible interior walls of the pyramid become massive

    projection mapping surfaces. One wall illuminates with "Pleasure," the next

    with "Peace," and the third with "Curiosity." As the script concludes, the

    apex of the pyramid opens up, revealing a starry sky or a bright light

    (representing the overarching "Happiness").

💡 Quick Blender Fulldome Tips:

1.  Camera Setup: Set your camera to Panoramic -> Fisheye Equisolid. Field of

    View: 180 degrees. Render resolution must be perfectly square (e.g., 4096

    x 4096).

2.  The Sweet Spot: In a dome, the audience is usually looking about 30 to 45

    degrees up from the bottom edge. Don't put your most important video exactly

    in the dead center (zenith), as it forces the audience to crane their necks

    straight up.

3.  Avoid Fast Camera Pans: Panning left or right in a dome causes severe motion

    sickness. Keep camera movements to slow forward/backward tracking, or very

    slow rotations. Let the objects (the video panels) move around the camera

    instead.

false alarm emails from github pages

We got three emails from github pages in March and April, saying Please verify your Pages domain (ACTION NEEDED) - "We are unable to find a verification TXT record  .... Please verify your domain within 7 days by following our instructions before it becomes unverified."

Found in each case by going to the github settings page for gh-pages, that the domain was verified OK, and the verification txt record was also fine. 

Since this has only happened for a domain which has DNS operated by Microsoft - Azure DNS - I guess the errors were DNS failures from Azure.

Blender 5 VSE difference for extending clips

 Blender 5 Video Sequence Editor (VSE) handles extension of clips differently than earlier versions. In earlier versions, if we extended a clip past its end by dragging its boundary, it would create still frames from the last frame of the clip, and if we used a speed control modifier, we could stretch clips both to extend them as well as to make them shorter. With Blender 5, we can't do that directly. Now, we need to go to clip properties, increase duration of the clip there. Just using speed control is no longer enough. Via  https://blender.stackexchange.com/questions/69679/extend-video-strip-in-vse

can't have a show all apps button on old Realme phone

There was no way to install a button to "show all apps" like on my Samsung M34, on the old Realme phone: in settings -> home screen, there is an option to show drawer instead of home screen icons - all the apps are listed in the drawer. Chose that instead.

how to remove sim from moto phone

This video helped me - https://www.youtube.com/watch?v=Ddb-PMm58Oo

We need to use the lever / slider mechanism on the sim tray to slide the sim out, since we can't get a grip on the sim in any other way.

Artemis II launch video

Youtube link to VR360 video in 8K - Experience NASA's Artemis II Moon Launch in 360° -

https://www.youtube.com/watch?v=9LFBFRH51Jo

Archive.org link - https://archive.org/details/experience-nasas-artemis-ii-moon-launch-in-360-9-lfbfrh-51-jo

Wordpress website giving critical error

One of our wordpress websites was giving a critical error, but only when trying to edit a particular page. A few days later, the admin for that site fixed it, probably taking help from some LLM etc - 

I enabled safe mode and also edited the configuration wp-config.php file, 

define('WP_MEMORY_LIMIT', '512M');

define('WP_MAX_MEMORY_LIMIT', '512M');

This seems to have solved the problem for now.

Logged in, saw that Wordpress and some plugins and themes needed to be updated, and updated them too. After updating, when I tried clicking on Edit for the home page, it seemed to work OK, probably because of the fix applied above.

using cloudflared to connect to our internal server

Got this message from PB:

Have done the following to do the cloudflare (ZeroTrust) tunnelling to ourservername. 

Cloudflare Zero Trust + Google Workspace + ourservername (SSH)

Architecture Overview

User (Client Machine)

        │

Cloudflare Access (Zero Trust)

        │

Google Workspace (Authentication + MFA)

        │

Cloudflare Tunnel (cloudflared)

        │

ourservername (SSH)

Authentication Flow Chart

User runs SSH (like ssh  ssh-ourservername.ourdomain.org)

        │

Cloudflare Access intercepts request

        │

Redirect to Google Login

        │

User authenticates + MFA

        │

Cloudflare verifies identity + policy

        │

Cloudflare establishes Tunnel

        │

Access granted to ourservername ssh

On the client machine 

Install cloudflared 

Create a config file in .ssh folder

$ ~/.ssh/config

Host ourservername

  HostName ssh-ourservername.ourdomain.org

  User ourusername

  ProxyCommand cloudflared access ssh --hostname %h

Then you the connection would be like

$ ssh ourservername

Connection was successful with ssh with the configuration as above. Connection was initially not successful with Filezilla. 

ChatGPT told me that filezilla does not understand the proxycommand instruction, hence we can run cloudflared in proxy mode by running this in a terminal:

cloudflared access ssh --hostname ssh-ourservername.ourdomain.org --url localhost:2244

This creates a local listener on localhost:2244 which forwards via Cloudflare to the remote SSH server

With this, Filezilla could connect to localhost 2244 successfully (and I could see the files on ourservername with Filezilla.)

OpenSpace - workarounds for toggle trails causing Pluto trail to appear

 Using the default action by pressing the 'T' key, Pluto trail starts to appear if we disable it before we start. 

One way might be to remove the Pluto Trail asset from openspace before starting the playback?

Another way would be to edit the recording to replace 

openspace.toggleFade("Scene.*Trail.Renderable");    openspace.toggleFade("Scene.*trail.Renderable");

with

openspace.setPropertyValue("Scene.*Trail.Renderable.Fade", 0, 1.0); openspace.setPropertyValue("Scene.*trail.Renderable.Fade", 0, 1.0)

openspace command convention

Quick note to self - when we want to use the built-in fadeIn or fadeOut commands, we can't use the name of the object as seen in the info tab - like Scene.Moon here,

We will get errors like
Error when calling script 'openspace.fadeIn': Could not find property 'Scene.Moon.Enabled'

Checking the script log, we need to append .Renderable to that name, like

openspace.fadeOut("Scene.Moon.Renderable", 5)

Edit - Note: All of the above were with respect to the 0.21.x version. In version 0.22, things may change - for eg this commit.

tumblerd prevents disk from being ejected on Linux

When trying to eject a USB external disk, a message along the lines of "process tumblerd is currently using the disk, eject anyway?"

https://duckduckgo.com/?q=what+is+tumblerd+which+is+keeping+disk+from+being+ejected+on+linux&ia=web

Since I had a very large number of images on that drive, and I had just saved an mp4 file onto the drive, tumblerd was probably creating thumbnails for all of the multimedia files. So, did 

killall -HUP tumblerd

and then the disk could be ejected.

OpenSpace session recording "save frames" crash

On my Lenovo ThinkCentre M93 machine with an NVidia GTX1050 graphics card, OpenSpace session recording playback with "save frames" was crashing after a few hundred frames being saved. Each frame was taking around 1000 to 1500 ms for 4096x4096 frames.

According to Claude.ai, the crashes might be due to CEF "being starved" between updates - and I submitted this fix as an issue on the OpenSpace repo.

Claude suggested workaround was in SGCT.

// sgct/src/image.cpp

png_set_compression_level(png, -1);  // changed to:

png_set_compression_level(png, 1);

As the comment just above that line says, -1 sets the compression level to default, which seems to be 6. Setting this to 1 made the frame export only 300 ms per frame (3x faster) and also fixed the crashes - so CEF was not "being starved" of updates.

The increase in file size was not much, 4 MB increased to 5.1 MB for a typical frame.

Another change I made before exporting again, was to close all other windows and apps like the Edge Browser, which I had used earlier for triggering the exports - now OpenSpace has some checkboxes in the export settings which allows us to hide the UI while exporting, so we don't need to trigger the export via a web browser.

Edit: No, the problem is not solved. No crash when saving frames for this recording, ts_003_moon_to_Earth.osrectxt

but crashes after a few hundred frames when saving frames for this recording, ts_002_india_rotate_zoom.osrectxt

So my current workaround was to split the recording into multiple smaller recordings.

Edit - another workaround to get around the crashes, which seem to be caused by CEF, seems to be to disable the module CEFui inside OpenSpace > Settings - issue thread is at https://github.com/OpenSpace/OpenSpace/issues/3954 

uninstalling ollama

https://github.com/ollama/ollama/issues/986

ran all the lines suggested there, got some 15 GB freed.

Copy-pasting from the above,

Here's a general guideline on how to uninstall it:

Delete the Ollama binary:

Use the rm command to remove the Ollama binary. For example:

sudo rm /usr/local/bin/ollama

If the script created a systemd service, disable and remove it:

If the script created a systemd service for Ollama, you should disable and remove it using the following commands:

sudo systemctl stop ollama

sudo systemctl disable ollama

sudo rm /etc/systemd/system/ollama.service

sudo systemctl daemon-reload

Remove any created user and group (if applicable):

The script might have created a user and group named "ollama." You can remove them using the following commands:

sudo userdel ollama

sudo groupdel ollama

-----------------------

More lines from other people, esp if location of ollama and downloaded models are different - 

sudo rm -v /usr/local/bin/ollama

sudo rm -rv /usr/local/lib/ollama

sudo systemctl stop ollama

sudo systemctl disable ollama

sudo rm -v /etc/systemd/system/ollama.service

sudo systemctl daemon-reload

sudo userdel ollama

sudo groupdel ollama

sudo rm -rv /usr/share/ollama

# for downloaded models

rm -rv ~/.ollama

sudo rm -rv /usr/lib/ollama/

sudo rm -rv /usr/local/lib/ollama

domain transfers - godaddy to nettigritty and one account to another

We had a few domains which needed to be transferred. A couple of domains needed to be transferred from one account to another on nettigritty, and a domain needed to be transferred from godaddy to nettigritty.

downloading from a server via http (no SSL) with auth

Downloading some planetarium files from a server in Sweden, which had password protected directories but no SSL (no https) - 

1. Directly downloading via the browser - slow, prompts us to "Keep" or delete the file every time due to being downloaded unsecurely
   
   
2. Using aria2 - 
   
    aria2c -x16 -s16 --http-user=USERNAME --http-passwd=PASSWORD url
   
   Need ProtonVPN, since without it, http downloads are severely throttled to less than 100 kbps.
   
   
3. Better option, running 
   
   aria2c -x16 -s16 --http-user=USERNAME --http-passwd=PASSWORD -i urls.txt
   
   on one of our AWS servers - all the URLs get downloaded in a jiffy, then we can use Filezilla and download to our local machine over sftp without throttling issues.

change of email id in CMS

One of the users in our CMS (google apps script with php/pgsql authentication) changed their email id in the profile, but was then unable to find any of the assets they had been working on.

Now changed the 'username' field in the postgresql database for this user from

old@gmail to new@new.email

(the email id field had already been changed by the user.)

That seems to have solved the issue. This is a duplicate of an earlier post, https://hnsws.blogspot.com/2024/06/unable-to-see-old-assets-after-email.html

OpenSpace - hover over a planet and allow it to rotate under you - OpenSpace seminar

As noted in an earlier post (which I had forgotten), the relevant option is in Settings > Navigation Handler > "Follow Anchor Node Rotation" 

When I asked Gemini via aistudio.google.com, it found a transcript from an OpenSpace seminar, which is a useful link - https://www.ppadomes.org/events/online-seminars/ppsarchive/2025-archive/2025-01-31-chat

github 14 day traffic stats

Since github has a page which shows 14-day stats, like https://github.com/hn-88/OpenSpace-AppImage/graphs/traffic but which is behind the login of the user, I asked Claude to create a stats page for the traffic to all my public repos. The prompt is here, 

https://github.com/hn-88/hn-88.github.io/blob/master/stats/prompt.md

and the github traffic stats page is here

https://hn-88.github.io/stats/

The "clones" number is so high for the https://github.com/hn-88/OpenSpace-AppImage/ repo because of the multiple daily build workflows.

adjusting sound volume on Mac

Since I don't have an Apple keyboard, adjusting the volume on the Mac mini was not very straightforward. An LLM's advice seemed to be a bit sarcastic when taken in the context of how easy it should be versus the method suggested - 

Adjusting the volume using the menu bar is a simple and straightforward method. Here are the steps to follow:

1. Click on the Apple logo in the top-left corner of your screen.
2. Select "System Preferences" from the drop-down menu.
3. In the System Preferences window, click on "Sound."
4. Go to the "Output" tab.
5. Drag the volume slider to adjust the volume level.

On my machine, it was "System Settings" and not Preferences. And, there seemed to be a slightly faster way, which is clicking on the "Control Centre" in the menu bar - screenshot below - the "edit controls" button allows us to drag the icon into the menu bar, too.

interesting - No-AI chatbot

Interesting to see how chatbots can be programmed manually, without any AI - https://p5js.org/tutorials/criticalai4-no-ai-chatbot/

Via https://p5js.org/

Via https://github.com/hedbergj/sims-dev/blob/gh-pages/expanding-universe-2/index.html

using Blender to mask bright scenes

Using the same technique as in this previous post, but running on Blender 5 on Mac Mini, 3840x2160 video for Unseen Earth was being rendered at around 15 fps using the ProRES LT codec. Apparently Blender still doesn't have GPU accelerated output codec support, and this is one of the fastest codecs as suggested by ChatGPT. The ProRES LT can then be re-encoded as HEVC using the py-ffmpeg-warp presets. 

Some more tweaks used were:

• Multiple masks one atop another to darken extremely bright areas,
• Using Shift-D to duplicate and then K to cut the mask
• Using an adjustment layer to increase contrast for the base layer. 

Edit: April 2026 - I now see that Blender 5 has "Curves" as one of the Adjustment layer modifiers - so, we can easily clamp bright regions using a "Curves" filter. And it looks like this has been there in older Blender versions as well.

close all Finder windows on Mac

It looks like Option+Cmd+w is the keyboard shortcut to close all windows of an app. Also, Finder has this method of Option+click on the File menu, which then shows a "Close all" menu item. 

A bit clunky in my opinion. And apparently many people moving from Linux/Windows to Mac have this problem, due to selecting a large number of files and hitting "Get Info..." like I did - MacOS then proceeds to open that many info boxes instead of showing a single window with cumulative info as on Linux/Windows.

tutorials on claude code - ollama - on Google Colab

 https://github.com/musistudio/claude-code-router - "...route Claude Code requests to different models and customize any request."

running locally - https://www.youtube.com/watch?v=nYDUdnMVDdU

claude code + ollama, another tutorial - https://www.youtube.com/watch?v=3x2q6-5XbQ8

ollama on Google Colab - https://www.youtube.com/watch?v=Y4R-DyvAux8

Moodle custom reports not being emailed

One of our Moodle instances had this issue being reported, that some newly created custom reports were not being emailed as scheduled. 

The issue was that notifications - including email notifications - were disabled for "Custom report builder schedules" at https://ourserver.org/admin/message.php

After enabling email notifications now for this, "Custom report builder schedules" - the issue was reportedly resolved.

DNS resolving glitch with lovable

One of our colleagues wanted a domain verified by "lovable the AI agent" - https://lovable.dev/

After setting up the verification txt record and the provided A record, he still got an error 1000 when trying to browse the website on the newly added subdomain - DNS points to prohibited IP.

Lovable suggested removing the proxy toggle on cloudflare (but that was already set to DNS-only).

So, I suggested that he maybe needs to click on some "verify domain ownership" button somewhere in "lovable" control panel if any ... 

He replied that he had initially used bigrock setup, now he "removed the domain in project settings of loveable and added again, it is now working."

google workspace storage limits

At some point of time, Google Workspace has added a feature to admin.google.com where we can set individual users' storage limits on or off, or change the value on an organizational unit basis. Our current Google workspace now has this feature. One possible catch is that the on/off setting is inherited from the base organizational unit. And it also seems to allow certain users to use 146% of their quota!

pdfs - make smaller - split or join

There was a requirement to 
1. Make some pdfs smaller than 200 kB to enable uploading on a govt. website
2. Split some pages of the pdf into separate documents.

I used pdfsam - https://pdfsam.org/download-pdfsam-basic/ - to do the separation and merging, 

and the online tool

https://www.adobe.com/acrobat/online/compress-pdf.html

to compress the pdf size.

troubleshooting dot net server crash on AWS

First, checked if the AWS dashboard has any troubleshooting info for me. But "This user does not have permissions to view AWS Health events" - so I wouldn't get much help from it. Used my earlier post on investigating another Linux server crash as well as ChatGPT for going step by step.

ChatGPT suggested these,

dmesg -T | grep -i oom

journalctl -k | grep -i oom

which had no results. No out-of-memory events seen with those commands.

du -sh /var/log showed 4 GB of journal

journalctl --disk-usage

also showed nearly 4 GB of current and archived logs.

But the system had enough hard disk space left, 

df -h

showed around 30% remaining for the OS disk. 

Checking authentication log with

tail /var/log/auth.log -n4000 | more

found an entry

2026-01-25T14:38:58.186744+05:30 ip-10-0-0-73 systemd-logind[600]: New seat seat0.

But ChatGPT said that's normal, this is apparently done for each login.

Then, copy-pasting the lines just before the crash from syslog, using T13 since the crash was around 13:30

sudo cat  syslog | grep T13 > /home/theadminaccount/syslogsnippet.txt

ChatGPT immediately found the issue. 

ourtestapi.service: Scheduled restart job, restart counter is at 101203

and the reason the service was failing was

The application '/home/thepath/www/ourapi/ourtest.dll' does not exist.

Stopped the service, disabled it, and deleted it from /etc/systemd/system

sudo systemctl stop ourtestapi

sudo systemctl disable ourtestapi

sudo rm /etc/systemd/system/ourtestapi.service

Additionally, as ChatGPT suggested, set up some limits for journalctl's logging by editing

/etc/systemd/journald.conf

and uncommenting these lines and adding values for them - under

[Journal]

SystemMaxUse=900M

SystemKeepFree=1G

RuntimeMaxUse=100M

MaxRetentionSec=30day

and restarted,

sudo systemctl restart systemd-journald

Also vacuumed (removed archived logs) with

sudo journalctl --vacuum-size=3000M

(0 bytes removed - archives were smaller than that)

sudo journalctl --vacuum-size=300M

(500M of archives were removed.)

see who has signed an app on MacOS

Via https://www.youtube.com/watch?v=1UDUtO7LY0k

 Apple Menu > About this Mac > More info button > (Search for System Report on left-hand panel)

In System Report, click the System Report button, then Software > Applications on left-hand panel.

Scroll down to the app in question.

Our self-signed OpenSpace builds show up as "Unknown". Sheepit Mac client also shows "Unknown", so there is a chance that I can edit the manifest of the app to run a script on double-click in order to allow the Mac build to run with a double-click on the app file.

to record a stereo track in Reaper

The default tracks are mono in Reaper. So, to record from a stereo source, as the LLM-powered search says, "create a new track and set it to record from a stereo input by selecting "Input Stereo/Input 1 + Input 2" in the track settings. Make sure to arm the track for recording and then start your audio source to capture the stereo sound."

Azure Storage minimum TLS notification

We got an email "Action required: Update the minimum TLS version of your Azure Storage account to 1.2 or later by 3 February 2026".

Updated the affected storage accounts to TLS 1.2 as required below - they were just backups/files from 2018 - but then, got the message that the old backups can be deleted.

So I have deleted the three storage accounts and their contents now - last modified in 2018 (wordpress files), 2020 and 2021 (server logs). The wordpress files are being served from github pages instead now.

allowing a non-root user to restart a service on Ubuntu Linux server

As described by ChatGPT, the preferred method was to create sudoers rules for the relevant service(s).

sudo visudo -f /etc/sudoers.d/ourapi

and inside that file, write

deployuser ALL=NOPASSWD: /usr/bin/systemctl restart ourapi.service, \

                         /usr/bin/systemctl start ourapi.service, \

                         /usr/bin/systemctl stop ourapi.service, \

                         /usr/bin/systemctl status ourapi.service

where ourapi.service is the service and deployuser is the non-root user.

Here, we double-checked the location of systemctl using 

which systemctl

After doing this, on logging on as deployuser, we can do things like

sudo systemctl start ourapi.service

where it will prompt for deployuser's password and then carry out the command.

But a caveat - we need to type exactly the same command as mentioned in the custom sudoers file - even equivalent commands like

sudo systemctl start ourapi

will not work unless we add that to the sudoers file also.

Azure supported VM sizes

Some Azure VM sizes become unavailable in some regions - case in point, after upgrading to 16 GB RAM for a day, I could not go back to the B-series 8 GB RAM VM in that Central India location (which had shown quota limited before I stopped it). Went back to a slightly more expensive D2asv4.

Edit - I had noted on 7th Jan that the size was Standard B2as v2 - vCPUs = 2, RAM = 8 GiB - and this was "insufficient quota" in Central India at the moment. It is Rs. 2998 per month. 

We then temporarily changed to D4as_v4 for a few hours, which is 7.5k per month. Stop the VM, change the size and then start the VM. And after the work was done, B2as v2 was not available, so we went to D2asv4 which is Rs. 3,778 per month.

some quirks of my KVM setup and Apple keyboard layout

Currently using this product, 

HDMI KVM Switch 4 Port - 

there were some quirks which I had to work around. 

1. The Dell desktop running Linux Mint needed the USB keyboard to be mapped to it while booting, or else it would boot into Windows instead, even though the default boot sequence was set to Linux. Probably because the "fail-safe" boot option is different, when it beeps to indicate "No keyboard detected."
2. The KVM switch defaults to number 4 on power up, not number 1. So, the Dell desktop needs to be connected to 4 and not 1, for booting up without having to hit the KVM switch.
3. The Mac mini can boot OK without having the KVM mapped to it, but if we leave it on and not logged in, the login screen blanks and we can't get the screen to come back to life with the keyboard or mouse - we need to short-press the power button.
4. The Mac mini default keyboard mapping was showing the Indian rupee symbol ₹ instead of the backtick symbol ` for the key to the left of 1 (perhaps after the upgrade to MacOS Tahoe). Since I need ` for markdown on github etc, changed the keyboard layout to "US" from "ABC India" which was the default. For this, following this,
   chose Apple menu > System Settings, then searched for Keyboard in the sidebar. There, Input sources > Edit button, the small + button at the bottom left to add Input sources. Quite easy to miss.

OpenSpace stopped working with undefined symbol: sqlite3_column_table_name

After some packages were updated / upgraded / autoremoved, openspace, which used to work fine, stopped working (on Linux Mint, using the Ubuntu 24.04 deb) with the error message

./OpenSpace: symbol lookup error: /lib/x86_64-linux-gnu/libgdal.so.34: undefined symbol: sqlite3_column_table_name

As suggested by ChatGPT, a reinstall of gdal and sqlite fixed it

sudo apt update

sudo apt install --reinstall libsqlite3-0 libsqlite3-dev libgdal34

https://github.com/hn-88/OpenSpace-AppImage/issues/86 has the full answer given by ChatGPT