Tsukasa no Hibi

There is not a single day without any attention-grabbing headline about either some cool project made possible through LLMs, companies cutting jobs because of LLMs, new security vulnerabilities found through LLMs, and/or open-source projects getting mad and banning the use of LLMs.

Here is the short version upfront: I think LLMs can be a great tool. And just like with every tool, you can either use them wisely and measured or swing them around, shouting like a madman. The way people react to the results will probably be reflected in your behaviour.

I think the use of LLMs in programming can be an incredible boon. I’ve been one of the testers of the original GitHub Copilot back when it was just a line/block suggestion system. It was incredibly cool to see the suggestion matching a loop or a single conditional.

This is obviously not what people refer to when they talk about “vibe-coding” today. People, even tech-illiterate folks, now interact with a computer program that can basically drive your entire PC. “Build the application, make no mistakes” has become a bit of a meme but essentially hits how a lot of people picture programming today.

And yeah, sure, there are people who no longer even look at the code. Uncle Bob, for instance. But hey, the benchmark for any self-respecting software-developer is to take a look at what Uncle Bob is doing and then doing the polar opposite. We have access to sophisticated systems that are able to regurgitate the entirety of mankind’s knowledge. That includes a ton of shitty GitHub and SourceForge projects.

Trusting the machine and leaving it to dictate the coding style, the architecture and the finer implementation details is a hallmark sign of a failed project – at least in my opinion. A machine cannot and must not be trusted to make these decisions. Not to mention that LLMs cannot make decisions. It is insane to me that a ton of companies trust mission-critical data to a third-party LLM, thus exfiltrating sensitive data – all by design, by process.

However, in my opinion, a true programmer also does not reject a good tool. Give an LLM a little bit of context about errors and let it derive possible causes. A good programmer weighs the results themselves, checks the results themselves. A good programmer already has ideas and merely uses an LLM as a consult. The final decisions always lie with the programmer. Because it is his code. Because it is his job. His responsibility. His liability. When something does not work, management does not call up Dario or Sam Altman – they call up the programmer.

As such I am equally shocked that a ton of open-source projects dogmatically reject code created with LLM assistance. I fully understand that nobody wants a fully LLM-written pull request on their project, something that has not been battle-tested by a human. It sucks having to review fully generated nonsense, the structure is often bad and the formatting nearly unreadable.

But to exclude code contributions because there might be a single line of auto-suggested loops in there is just as bad as the people screaming about programming being a solved problem. It’s pure dogmatism, pure populism. It is nonsense.

The way I see it, you can go one of two routes: You can either be a dogmatist or a pragmatist. A pragmatic programmer (pun intended!) will use a good tool when it presents itself. A pragmatic programmer will weigh the use of that tool not based on propaganda and group-think but by the use they get out of it.

The other option is to be a dogmatist. Someone who outright rejects an idea not because of practical reasons but because of a self-inflicted sense of moral superiority. “My code is better”, they might say. “It is unethical”, they might add.

I hate the “you ngmi” crowd that rolled over from cryptocurrencies and now view LLM-assisted coding as the new gold rush.

I equally hate the pink-haired transgender luddite that calls people nazis for using LLMs.

At the end, LLMs are tools. You decide how to use them. You are responsible for your actions. You are responsible for the results. Do not act like LLMs produce better or worse code than the average dogshit developer. Do not simply take the dogshit code and commit it. Think for yourself. Because that is your responsibility. Your job. Your most important tool is your mind and your most important asset is your ability to make decisions – something the machine cannot and must not ever do for you.

In short: Stop being annoying about vibe-coding.

You have added your non-Steam game to Steam. Customized all the icons, banners, graphics. Everything looks perfect. Almost.

Because whenever you open a game’s details in your library, you are greeted with a basically empty screen. Such wasted space, you could see the latest news for it.

This is where my External Data Sources plugin for Millennium comes in. It allows you to specify an RSS or Atom feed for a non-Steam game in the game’s properties.

The feed will render as a normal game news feed on the game’s details page.

Finally, you can have your Wowhead news straight in your World of Warcraft details page in the Steam client!

Trying to create a desktop shortcut for the World of Warcraft: Forever beta results in an error message. Trying to fiddle with the WoW Launcher also seems to result in issues.

Thankfully, the old launch command via Battle.net.exe works fine:

Battle.net.exe --exec="launch WoWF"

With this command you can easily create a desktop shortcut or add the game to Steam.

However, there is a catch: If you want to use the Steam overlay or track your time, this approach does not work due to the way Battle.net launches processes.

You will need a tool like bnetlauncher to work around this. Except that bnetlauncher also does not work, so I have fixed up a branch with the required changes.

To start World of Warcraft: Forever, you can add bnetlauncher.exe to Steam as a non-Steam game with the following parameters:

bnetlauncher.exe wowclassicbeta -t 15

This should bring up the Battle.net client, switch to the World of Warcraft: Forever page and automatically “click” the “Play” button to launch the game. The Steam overlay should work and you should be displayed as playing a non-Steam game for the entire duration of your session.

As of late there are several reports of 1Password for Windows spawning but not cleaning up their 1Password-BrowserSupport processes.

These processes normally are part of the native messaging between the web-browser and the 1Password client. Usually, there should be one process per browser process (main process, not the children).

As mentioned before, there are several reports of hundred to thousands of these processes spawning on users’ machines. I’ve had the problem myself, it is not fun. The system becomes unresponsive, there is an excessive CPU drain (often through WMI) and trying to plug-in or remove monitors cause a noticeable delay of several seconds where the entire system freezes.

So, how do you fix it?

Option 1: Disable the 1Password browser extension

This is not really what you want, but it does work. Disabling the extension will stop spawning new instances of the process.

Option 2: Add the browser to the “trusted” list in 1Password

As mentioned on the forums, the better way is to add the browser executable as a “trusted” web-browser.

Open the 1Password for Windows settings, go to the “Browser” tab, click “Add Browser” and select the executable for your web-browser.

This can be necessary if you are using more obscure browsers like Wavebox that are not automatically on 1Password’s “trusted” list.

Restart the browser afterwards, the processes should now spawn and be terminated properly.

I have had an odd issue for a while: Kagi Privacy Pass would work perfectly in regular browser windows, however as soon as I launched a private window and tried to do searches, I would receive the Chrome error ERR_BLOCKED_BY_CLIENT for the extension’s chrome-extension://mendokngpagmkejfpmeellpppjgbpdaj/pages/redirector URLs in Vivaldi.

The source of my problem: I added Kagi as a custom search engine with my token in the URL. This token was also being passed as part of the redirector, causing the request to be blocked.

The fix was just as simple: Add another instance of Kagi as a search engine, however without the token parameter. Enable this second instance as a search engine for private windows.

Problem solved.

In my last post I kept reminiscing about the most safe and versatile way to boot a Macrium Reflect rescue media. After doing exactly what I set out to do (install a rescue media onto a USB drive – without Ventoy, Easy2Boot or any other shenanigans), I was presented with the chance to actually try out an ST400 for myself via a loaner device.

The Problem

Inserting and booting Windows 10 or Windows 11-based ISO files into the virtual optical drive of the ST400 worked fine. Only with Linux ISOs like Ubuntu or various Arch-based distros I ran into trouble.

After the bootloader, I got booted into initramfs shell or experienced some form of I/O related error with the emulated optical drive.

Stranger still, when disabling SecureBoot, booting the same ISO files works fine.

Even stranger still, booting the same ISO on another machine worked fine with SecureBoot enabled.

So there is clearly something afoot here.

The Workaround

If disabling SecureBoot is not an option, you can work around the issue. Ignore the online discourse about faulty cables or faulty ISOs – this is some odd synchronicity between your mainboard and the ST400 itself.

On most Linux bootloaders you can press “E” to edit the current entry.

Remove the “quiet” and “splash” arguments, and insert the two arguments “break” and “debug” instead:

setparams 'Try or Install Ubuntu'

   set gfxpayload=keep
   linux  /casper/vmlinuz --- break debug
   initrd /casper/initrd

The exact parameters present in the original command line can vary across distros, however the parameters you need to add/remove stay the same.

Then press F10 to boot. You will still run into the issue.

Now hold the “1” key on the ST400 pressed for 3 seconds. This causes the optical drive to reload.

This will be noted by some debug output on the Linux console.

Now enter “exit” and press Enter, the boot sequence should continue fine.

Observations

The same behaviour is also present if you use VMware Workstation, add the IODD optical drive as a CD-ROM/DVD-ROM drive to a virtual machine, and try to boot.

Using the same ISO file in VMware Workstation directly via the IODD hard disk mode from the ST400 works – so the ISO file is fine.

Just like on “real hardware”, the issue disappears once SecureBoot is disabled on the VMware Workstation host machine.

I am fairly sure that this is related to the way my mainboard’s UEFI handles SecureBoot. Perhaps this is even the intended/correct behaviour, however I just find it somewhat irritating.

The good news is that the Macrium Reflect rescue media is based on Windows, so there would be no need to fenagle with workarounds when in need.

A good backup strategy consists of two things: A way to create backups – and a way to restore them. We often think only of the first step, while assuming that the restore process will “just work”. Until it does not work and we start sitting in hot seats in hours of need.

A recent crash during a VMware Workstation upgrade caused by a page fault error in a Citrix USB passthrough driver left my system in a bad state. Instead of going the painful route of trying to manually repair the damage done, I opted for a rollback to my last backup. Why spend hours wrangling with the registry and obscure driver catalogues when a rollback only takes 10 minutes, right?

As a huge fan of Macrium Reflect Home this could not be easier: Start Macrium Reflect, select the backup I want to restore, select the partitions to restore, reboot and let the restore process run.

In theory. Ideally. If everything goes well.

In practice, Macrium Reflect prepared itself, rebooted the machine, and the WinRE-based rescue environment booted up. Macrium Reflect started loading. And then it was stuck on a “Loading wpeinit” step, seemingly hanging. After 10 minutes of waiting, I hit the Escape key and the application kept loading – however without the required drivers, so a restore was not possible because no network drivers were available.

So what’s up with this?

In Macrium Reflect you have several options when it comes to building a rescue media. You can either create an ISO file, prepare a USB thumb drive, or use a WIM image to create an extra environment that gets added to the Windows boot loader.

I was using the additional boot menu entry with the normal Windows WinRE-based environment as a base image, as that is what Macrium recommends within the application.

This WinRE image seems to be fucked on both machines I have tried booting it.

However, as it turns out, a clean image labeled as “Windows PE 10 (WADK)” you are able to select under the “Advanced” button in the “Create Rescue Media” is a better fit and will work. It is likely that local WinRE images are somehow corrupted, although I did not investigate this any further.

After automatically downloading and building the image, I was able to boot into the refreshed Macrium rescue media and restore my backup without any issues.

The lesson here is to quickly check whether or not you can actually boot the rescue media. Do not trust blindly in the process, actually test it. Repeat these tests at regular intervals. It does not take long and will save your butt when you need the process to work.

Ideally, I would like to use something like Bootimus to offer these environments via PXE on my network, however given the graceless way Bootimus handles Windows-based images, I highly doubt I will ever be able to use or trust it to work reliably. iVentoy would work reliably, however comes with its own set of issues related to non-Windows environments.

Something like a IODD ST400 looks to be a better option, however given the current SSD prices I simply cannot afford a combination of device and (sufficiently large) drive.

A much cheaper option is creating a rescue environment on a USB thumb drive. This is the option I will be pursuing, as a cheap SanDisk thumb drive is only a few bucks and having a physical media to boot from gives me more peace of mind.

As a final thought, I must re-iterate my recommendation for people to do their backups daily. You do not need them often, and with good software like Macrium Reflect Home you can set up the backup once and then just let it run in the background.

I recently had the chance to work with a UniFi UDB-Switch during a network deployment. Long story short: It is an unmanaged switch with a 10 Gbit RJ45 uplink port and 7x 2.5 Gbit RJ45 ports. Basically your normal switch.

Where the UDB-Switch differs from the Flex 2.5 or other UniFi products is the fact that it has an integrated WiFi 7 uplink, meaning that you can deploy it without having to use any of the RJ45 ports to connect it to the rest of the network.

For the deployment I was part of, drilling through the walls or running cables was simply not an option. Thankfully, the distance we had to cover was short, so we contemplated whether or not the UDB-Switch could be the answer. We took the plunge and deployed it – and it worked beautifully for the simple setup of bridging a handful of wired devices across two rooms.

In fact, I was so flabbergasted by the device that I had to grab one for myself to play around with.

Unfortunately, while it is great as a device bridge, it falls short in several other key areas. I was not able to test this during deployment, so the following points are purely based on my personal tests with my own device.

No MLO

Yes, you read it correctly. The UDB-Switch, a device without 2.4 GHz support, lacks MLO. To be fair, it is supposed to come in a future firmware update, however keep in mind that the device has been on the market for about 6 months now – so a key feature that can be pivotal for enhanced performance is just missing from the product you can buy today.

On the 6 GHz band I was able to get around 1.3 Gbit/s in my iperf3 tests across my 10 Gbit network directly from my Gateway Fiber to the UDB-Switch, however keep in mind that these values mean next to nothing.

No LACP

I like the fact that the switch has 2.5 Gbit ports. However, some of my older multi port devices only do 1 Gbit per port; on other switches they benefit greatly from combining 2 or 4 ports into an LACP trunk (at least when multiple devices access them).

The UDB-Switch does not support LACP or bonding. This is not a case of having to uplink it via cable to enable the feature – it just doesn’t support it. Similar to the Flex 2.5 this seems to be fully intended.

Personally, I think this is a shame.

Media Disconnects and Uplink Changes

Switching between a wired uplink and a wireless uplink requires a complete reboot of the switch. This is somewhat baffling, especially if people purchase the device with the intent to have an optional wireless uplink in case the wired one goes down.

I was unable to determine whether this is a bug in the firmware or intended behaviour.

It runs HOT.

Similar to other Ubiquity UniFi devices, the UDB-Switch runs fairly hot, with idle temperatures around 60°C in my case. A bit of sleuthing shows that people report temperatures up to 80°C.

Thoughts

Ultimately a lot of my complaints can be waived away with the statement “that is not what the UDB-Switch is intended for”. And there may be truth to that.

However, the lackluster and outright unfinished firmware suggests a rushed product release. MLO STR is at least on the horizon, with signs of near-future release. However, as for my other points, I doubt any of them will ever get addressed.

The device is great if you need to bridge a gap and can easily replace a dedicated WiFi uplink client. Just be aware of the potential gotchas. My recommendation is to check the specs of the Flex 2.5, as the device seems to be basically the same (minus the WiFi uplink).

For more sophisticated switching requirements, you might fare better with a more potent managed switch and a dedicated WiFi uplink device.

If all you want is reasonable bandwidth and ethernet from, pardon the pun, the aether – then the UDB-Switch is a great choice. The people we deployed the UDB-Switch to are happy, bandwidth is not a problem (even during more congested scenarios).

Most important of all: No cables had to be routed and no holes drilled. And sometimes, that is all that matters. However strange that may sound.

Small oddity about Windows 10: When trying to shut down the computer, it started doing a very quick reboot (without going through the entire BIOS/UEFI/bootloader sequence) back to the login screen. The event log does not help. So what gives?

Turns out, there is a setting called “Fast Startup” in the power options:

Disabling the “Turn on fast startup (recommended)” option fixes Windows’ unwillingness to shut down.

I have no explanation as for why this happens, but at least it enables me to properly shut down my machine instead of choking my power button for 5 seconds to force a power off.

After a good long while, I have updated my local installation of LINQPad, a .net programmer scratchpad application that allows for easy computation and data analysis.

Much to my bemusement the application now features integration of LLMs like OpenAI or Anthropic. However, LINQPad also supports the use of a custom endpoint, meaning you can roll an Ollama setup and waste your own power instead of paying to waste someone else’s.

To configure Ollama, open the AI Settings and add a new AI provider.

When prompted to select the provider type, select “Custom OpenAI”.

Give the endpoint a proper name (i.e. Ollama) and set your endpoint to:

http://localhost:11434/v1/chat/completions/

Uncheck the existing OpenAI and Anthropic models and add your own local ones that you have installed with ollama pull <modelname>.

I found qwen3 to be a somewhat competent test model.

Using this setup, I was able to play around with the LLM integration into LINQPad. Unfortunately, as of version 9.6.6, there is no way to completely hide all traces of the AI stuff. If you are not interested in using these features, be it for monetary or compliance reasons, or out of conviction, you will constantly be reminded that they exist.

Right now I am in the process of migrating everything except telephony off my FRITZ!Box. After a lot of experimentation, sleepless nights and much frustration, I have come to the following conclusions about FRITZ!’s little all-in-one device.

Running the FRITZ!Box in IP-Client mode is basically impossible due to the sheer number of bugs. In this mode the router becomes just another client device in the network.

In theory this should be exactly what I want. It receives it’s IP address via DHCP and only acts as a SIP gateway. Or so you think. Because it doesn’t do jack shit. Once in IP-Client mode, the box no longer cares about WAN IP changes. It doesn’t even detect pulling all ethernet cables and just goes on being “online”.

Because of this, WAN IP changes do not cause a SIP REGISTER packet to be sent, meaning that you will have to wait the allotted time defined as REGISTER renewal in the original REGISTER response. In the meantime you cannot accept or initiate calls. Good stuff, FRITZ!

Now in the olden times people were able to activate telnet access to their FRITZ!Box and put a simple shell script into the debug.cfg to check for WAN IP changes and to reload the voipd if necessary.

But no worries, this feature has long been deactivated, so that absolutely no means of fixing a 15 year old issue are present anymore. Yes, you can restart the device through TR-064 via a shell script from another system, however I am not a fan of restarting an entire device because one teeny tiny option is missing. So… Good stuff, FRITZ!

And just to be extra annoying the FRITZ!Box decided to enable PVC on its own. Even finding the source of this issue took a good amount of time because why would I ever look at settings that I never changed to begin with? Good stuff, FRITZ!

While we are on the topic of settings: You cannot change the local domain name from fritz.box to something compliant with a .internal domain. Good stuff, FRITZ!, please ensure your fancy domain name keeps getting renewed.

All of these issues point to poor software quality or malicious compliance. Even without the ability to access the internals of the device and just by using the web interface (after several factory resets, to boot!) I was able to produce a ton of bugs and seemingly configuration corruptions when using the device in IP-Client mode.

Why telephony in IP-Client mode doesn’t regularly poll its own WAN IP to re-issue a REGISTER on change is beyond me. Why upstream gateways are not being accepted unless a static IP address is provided is also beyond me. Why this device is the only device on my entire network to behave so catastrophically uncompliant? I wish I knew.

There is no doubt that for regular households a FRITZ!Box is an excellent device. TR-069 automatically pulls the configuration for subscribers, WiFi configuration is dead simple, telephony has all the basic features you need. And all of that is pretty much working great when in router mode. But as soon as you switch into IP-Client mode, even after a full factory reset, the device breaks apart.

Skimming over 10-15 years worth of forum posts in search of solutions to seemingly wide-spread problems leaves me with the following conclusion: Instead of addressing long-standing problems within the networking core of the product, FRITZ! seemingly focuses on FRITZ!nas (because everyone and their dog uses their underpowered router as a NAS), on increasing the filter lists from 10 to 12 or on releasing another model of a FRITZ!Box with 1G ports.

The final verdict is that I will have to use the FRITZ!Box as a modem/router combo just so that I can keep the telephony features working reliably. Unfortunately it needs to sit at the front door because it cannot comprehend what a window is. Because unless it runs as a router, a lot of features of the FRITZ!Box simply stop working reliably or actively break.

Everything (including WiFi) apart from telephony will move behind the UniFi Gateway Fiber, using dedicated hardware.

What a shame, FRITZ! always stood for “not fancy, but reliable and compliant”. I’ve been using their hardware since the late 1990s, starting with the FRITZ! PCI ISDN card. Lately, their hardware offerings have been rather weak (why doesn’t the FRITZ!Box 5690 Pro have 10Gbit WAN ports? Why is there no XGPON/XGSPON support present?) and the reliability of the routers have been middling.

My 5690 Pro keeps crashing (as in: the device has to reboot) when certain WiFi settings are active. Extremely “Pro”. Makes you wonder if it will end up like a 7590 with hardware defects after 5 years due to maliciously misused components. Because we all love planned obsolescence!

While I do have my gripes with UniFi as well, it is at least “the Apple of networking stuff”: Nice user experience for dummies like me who want more control but are not Cisco-certified network gods. And I can change the gosh darn DHCP parameters like network boot and TFTP! Imagine that! And thanks to the SFP+ slot on the Gateway Fiber and a nifty X-GPON SFP module, I can quickly (!) bring up a backup GPON connection should the FRITZ!Box ever shit the bed, since all of the network topology is already behind (or will soon be behind) the gateway. Imagine that!

I purchased a Luleey LL-XS2510 a while ago. This SFP module acts as a fiber modem in GPON-based networks and has some creature comforts like configurable GPON serial numbers, vendor ID or MAC – essentially allowing you to “clone” your existing fiber modem to have a failover backup device ready, should your primary modem malfunction.

I wanted to use the module in a UniFi Gateway Fiber, simply to see whether or not the hype around Ubiquity UniFi was justified or not.

However, after configuring the Luleey with a media converter and pushing it into its own subnet, I came across a silly issue: I could not longer access the SFP module’s built-in web UI.

So, what is the situation?

• The Luleey SFP module has a static IP address of 192.168.10.1
• The default network of the UniFi Gateway Fiber is 192.168.1.0/24

Clearly, we need some routing here. The following instructions should get you up and running.

Note: This guide assumes that the port used for the Luleey SFP module is initially set to a LAN port. Please make sure it is not set to WAN!

Step 1: Create a new network

Go to Settings – Networks and create a new Virtual Network.

Set “Gateway Fiber” as the router and the zone to “Internal”.

Leave “Auto Internet Source IP / NAT” ticked but remove the checkbox for “Auto-Scale Network”.

Set the IPv4 address to 192.168.10.254 and the netmask to 24.

Set the VLAN ID to an unused VLAN.

Leave all checkboxes except mDNS unticked and set DHCP Mode to “None”.

Since the Luleey module does not seem to have IPv6 support, set the IPv6 Interface Type to “None” as well.

Save the network.

Step 2: Assign the network

Now open your port overview, select the port with the SFP module and set the native VLAN / Network to the network created during step 1.

Step 3: Create Policy Engine rules

Open System – Policy Engine – Policy Tables and create a new NAT Policy named “Translate Traffic to Luleey”.

Set the Type to “Masquerade” and the Interface/VPN Tunnel to the network created during step 1.

Keep the Source and Port to “Any”.

Set Destination to “Network” and select the network created during step 1. Keep the Port to “Any”.

Save the policy.

Step 4: Test it

You should now be able to open the Luleey web interface by opening http://192.168.10.1.

Caveats

As mentioned before starting, this only works for LAN ports. As soon as you assign the port to a WAN interface, the policy will no longer work.

However, unassigning a WAN interface from the port immediately restores access.

This is quite handy if you ran tests and want to check PON parameters or statistics about HEC errors. Since the module does not reset, the existing values from your prior tests will still be visible – giving you some indications about the quality of the connection.

Of course, console access works as well.

UniFi Caveats

There is one big UniFi-specific caveat: Just plugging in the SFP module will not bring it online in UniFi Gateway. Instead, the port will report failure due to signal problems.

To solve this, please connect a live fiber connection to the module. UniFi should stop throwing fits afterwards and allow operation of the module.

When trying to backup a Synology NAS, one has a few options. Synology’s Hyper Backup is great if you have relatively small shares or want to keep your NAS’ configuration safe. However, with larger volume or share sizes Hyper Backup becomes pretty unwieldy because it needs to finish a backup in one go, with no option to resume (i.e. after a power outage). This is bad because all of a sudden Hyper Backup not only has to roll back its already backed-up data, it starts from the very beginning again.

rsync has the ability to resume file transfers, however it is single threaded and relatively slow with an approximate 30-35 MB/s. This sort of transfer speed does not work for larger volumes or shares.

The solution I ultimately landed on was a two-step approach:

1. Mount your large share(s) via NFS on the destination NAS.
2. Install Rclone from the SynoCommunity community packages.
3. Do a normal local folder-to-folder sync.

rclone sync nfs_source_folder local_destination_folder --metadata --transfers 4 --checkers 16 --progress

With rclone, I get a nice and constant 98 MB/s with the ability to cancel and resume at any time.

You might be tempted to use SFTP instead, however this approach did not give me performance numbers even close to NFS.

Hyper Backups deduplication is nice, however I prefer having my backups finish within this decade over saving a few gigabyte.

The 14th of October 2025 is the day when Microsoft declares Windows 10 dead. The company expects regular home users to have migrated to Windows 11 by then.

Many users, myself included, are reluctant to switch over. The many horrifying news articles, often speaking of bugs so severe that only a restore from a backup will help, are not exactly inspiring confidence in Microsoft’s operating system.

Others will point to Microsoft’s statement that Windows 10 was supposed to be the last Windows version (according to Microsoft).

Personally, I simply cannot get used to the dumbed-down user interface of Windows 11. I need the backwards-compatibility of Windows Explorer, especially in regards to the Windows taskbar.

Windows 11’s new interface has been rewritten and redesigned, disregarding 30 years of backwards compatibility. I have no doubts that “you are just backwards” or “this is the future, old man” will be the comments. Unfortunately, these comments do not address the underlying issue.

The new user interface is not only breaking backwards-compatibility, it is also slow and impractical for power users, missing some basic quality-of-life features like being able to reposition the taskbar to another side of the screen.

Again, the Microsoft sycophants will crawl out of the woodwork and proclaim that all of these issues can be addressed with enough registry tweaking or third-party tools – missing the point once again.

I do not want to use a third-party shell to be able to use a program that I’ve been using for 25 years. I do not want to use a third-party hack (that breaks regularly after Windows updates) to restore functionality that never should have been cut to begin with.

Yes, ExplorerPatcher is a great tool. I’ve given it an extensive go in a virtual machine to see whether or not it makes Windows 11 bearable. It does, because it effectively turns Windows 11’s shell back into Windows 10’s shell. The more pressing questions are: Why do I need to use a tool to begin with and how long until Microsoft renders some of the tweaks to the start menu and taskbar (that I consider vital) completely unusable.

Personally, I do not foresee myself upgrading to Windows 11 anytime soon. No matter how great the under-the-hood changes of Windows 11 are, the user interface makes it a complete no-go for me.

I had hopes that Microsoft would address these long-standing issues over the years. And to be fair, some of them have been addressed – at least a teeny tiny bit. But power-users with the need to use some legacy software are trapped on Windows 10.

And what did Microsoft achieve, I ask you. A worse version of the same user interface we have had for the last 30 years. A taskbar that now needs to load for a while before even appearing. A start menu that is impractical to use. A new file manager that… well, at least Windows Explorer has tabs now. Good on you, Microsoft, you did something good. But you screwed up the context-menu, so you still won’t get a pat on the back from me.

This will become a thorn in Microsoft’s side over time. If you melt away many of the conveniences and third-party applications that anchored people into your ecosystem, while simultaneously antagonizing your old userbase with advertisements, dumbing down the operating system, an overall bad performance when it comes to the quality of updates and unwanted user metrics/AI integration, there is a real risk that people will just stay on outdated systems.

For me personally, switching to Linux is not an option. I came from Linux after wrestling with the open-source desktop for close to a decade. I have had nothing but issues with Wayland during my brief tests every few months, and last time I checked, two or three additional layers of audio subsystem abstraction have been added. I do not have the strength to migrate back and fight the same battles I’ve had in the early 2000s.

I also prefer my software to be written by people that are at least accountable – not by an organization that hires a shaman or a group that gets Trump Derangement Syndrome and purges people they deem “politically evil”.

Why can’t we just have the same desktop we had for 30 years? I cannot comprehend how that is too much of an ask.

For a while I have had an annoying issue with Citrix Workspace 2503.2. In the past, the Receiver application would hold and refresh my token, allowing for prolonged operation. I was able to show the main window over the course of the day and launch multiple applications.

At some point, I started receiving the error message “Unable to complete request” in the main window. The tray context menu switched from the standard menu with Sign In/Out, Connection Center etc. to a simple Open and Exit menu.

The Event Log showed that Receiver.exe was crashing due to this .net exception: The process was terminated due to an internal error in the .NET Runtime at IP 0x7395633A (0x73950000) with exit code 0x80131506.

So, what is going on? Simply put: The current version of .net Framework 8 (at the time of writing: 8.0.16) is causing some sort of regression that ultimately crashes background processes. This leads to the degraded user experience and service interruptions.

Uninstalling .net Framework 8.0.16 and re-installing 8.0.14 solves this issue. Be sure to install the x86 runtime for Citrix Workspace.