the stream

Apple's Magic Mouse line uses a touch surface instead of a scroll wheel, which isn't directly supported by the standard MS Windows mouse driver.

Apple provided Windows drivers for these mice as part of their Bootcamp package, last published around 2016. There's tons of articles out there on how to retrieve these drivers with Brigadier (Github Repo), but they won't work with the latest Magic Mouse that has an USB-C charge port.

Apparently this is due to newer Magic Mouse models having a different Bluetooth PID that's not listed in the .inf for Bootcamp drivers.

Luckily, someone went to dig through all this and created an installer that beats AppleWirelessMouse.sys into a shape that's both acceptable by Windows 10 and 11, and solves the BT PID issue:
Apple Magic Mouse Scroll Fix for Windows 11 (Github) (cache)

As of now, the install.ps1 powershell script from that repo looks ok to me, though I didn't use their mouse driver rebuild (the affected machine here is on Win 10), and instead went for BootCamp-041-88430 drivers (use -m iMac17,1 with Brigadier, extract driver from AppleWirelessMouse64.exe in the Apple drivers directory with something like 7zip, copy file to a drivers/ directory next to the script).

It took about 15 seconds after the script was completed for my mouse to come back online, but then its touch scroll function was immediately usable, without a reboot.

I've been pushing back an upgrade for quite some time now, but running on php 7.4 for the old Tiki version as grown unsustainable.

Up to now, I've not seen any major problems after doing the default upgrade steps - theming is off, as always, and I'll probably have to find a new default theme.

Not sure if there's any problems hidden in features I haven't used for ages, but the blogs (and images) seem to be ok, at least for recent posts.

The only actual error was that I wasn't able to update the search index. Even after switching my search provider in Tiki's web interface, I got:

error: Unsupported index type "lucene". Needs to be one of "mysql", "elastic" or "manticore". Try resaving the Search Control Panel
error: Contact the site administrator. The index needs rebuilding.

Search index rebuild failed. Last messages shown above.

Turns out the unified_engine in setting in the tiki_preferences table was not updated, so I did that manually in the DB shell with a
update tiki_preferences set value='mysql' where name='unified_engine';

For some reason this didn't seem activate until I also cleaned up all the orphan preferences from the web admin menu, but now the search index has rebuilt.

Oh, and the editor's preview tab produces a 400 error (but using the Preview button still works).

I recently tried to upgrade the Windows 10 Bluetooth drivers on an ThinkPad T490 with an Intel Wireless-AC 9560 chipset from the version supplied by Lenovo (22) to the latest Intel drivers (23 - BT-23.10.0-64UWD-Win10-Win11.exe at this time).

After installing the driver package, existing Bluetooth devices were inaccessible, and deletion from the Bluetooth settings panel was blocked ("remove failed"), even after removing corresponding devices from Device Manager. Looking at the Registry, it turned out that it was also not possible to manually delete associated registry entries, which now belonged to an unknown SID. (Maybe I could have replaced ownership on those entries, but they have a bunch of subdirectories, each of which would have had to be reclaimed individually...)

My solution was to uninstall the new Intel Bluetooth drivers: Windows then reinstalled the old version, and with those active, it was possible to properly delete all existing BT devices. With those cleaned away, I ran the new driver package installer another time, and re-paired all devices from scratch.

I use a sensor running Tasmota that reads data from my utility meter to graph the energy consumption in Home Assistant. This has been working mostly fine until I experimentally added a solar panel recently: It now has happened twice that the Energy dashboard in Home Assistant has shown an absurd value that's near or around the total consumption, instead of the difference to the previous measurement.

After finding this post on the issue tracker (cache), I'm moderately certain the underlying issue is that my sensor definition is using state_class: total_increasing, which gets reset if the measurement value is ever decreasing. Now in theory the utility meter counter shouldn't allowed to be decreasing, but it's not impossible it has the wrong configuration. I have now changed the setting to state_class: total (which will also make it possible to track if that's actually the problem I'm having).

After the first time this error showed up, I found a post that tells how to fix wrong values in the Home Assistant sqlite database (cache):

• retrieve database ID of the sensor from the statistics_meta table
• look at data for that ID in the statistics table
• update statistics and statistics_short_term with correct data

In my case, I only have one Tasmota sensor, so using the sqlite3 shell a
select * from statistics_meta where statistic_id like '%tasmota%';
gave me the id of the sensor (9).

Then I just ran
select id,state,sum from statistics where metadata_id = 9;
and paged back through the returned values to find any irregularities. Since the error had occurred recently, it was easy to visually identify the point where the sum value jumped up.

After calculating the offset to get the next increment from the correct sum, I updated the sum data for all rows that had the wrong value:
update statistics set sum = sum - 8654.0847 where metadata_id == 9 and id >= 143546;

...and then I applied the same offset to the values in statistics_short_term, which were all wrong:
update statistics_short_term set sum = sum - 8654.0847 where metadata_id == 9;

This still leaves me with the same corruption in the cost data that's calculated from the energy usage, but since I don't use that information, I just ignored this detail.

The Vivaldi web browser has an arm64 version that runs on the recently released arm64 port of Raspberry Pi OS. Unfortunately, out of the box, this version doesn't support a lot of media formats, and relies - same as Chromium, which Vivaldi is built on - on an external decoder library for some of them.

Now the Vivaldi distribution comes with a downloader script for that shared library at /opt/vivaldi/update-ffmpeg. As of right now, the script doesn't find a valid file for arm64 though. Looking at the code, it fetches a custom version of a Debian/Ubuntu package for chromium-codecs-ffmpeg-extra from Launchpad. As it turns out, Raspberry Pi OS ships an own version of exactly that deb, which installs a version of the required library to /usr/lib/chromium-browser/libffmpeg.so. This is great, except that it's not a place where Vivaldi looks for the file.

Vivaldi itself is started through an own wrapper script that's installed to /usr/bin/vivaldi. The wrapper has a few places where it tries to LD_PRELOAD a matching libffmpeg.so from, one of which is a custom location below the current user's ~/.local/ directory, involving the checksum of the expected library version. Luckily, the checksum is not actually checked - it's just used as part of the directory name.

So I came up with the following one-liner to create that directory for the current user and drop a symlink to Raspbian's chromium codec file:

grep -A1 arm64 /usr/bin/vivaldi | awk -F'=' '/FFMPEG_SUM_DEB/ { cmd="mkdir -p ~/.local/lib/vivaldi/media-codecs-"$2" ; ln -s /usr/lib/chromium-browser/libffmpeg.so ~/.local/lib/vivaldi/media-codecs-"$2"/libffmpeg.so" ; system(cmd) }'

This fetches a few candidate lines from the Vivaldi startup wrapper, finds the one that has the required checksum, and runs it through awk to create directory and symlink.

I have not yet tried to find out if the expected checksum changes between Vivaldi versions, but unless there's a major change in /usr/bin/vivaldi, I can just run the command again to create a new directory in the required location.

Note this doesn't provide Widevine or other DRM codecs, but at least I can now play back videos that have been reencoded to H.264/AAC by Mastodon.

(I have pretty much the same information over in a Fediverse thread, including a reply from Vivaldi that said the libffmeg download link is usually updated by them in time, and my hack shouldn't be required.)

I noticed that the old Mac Mini I reinstalled with Linux (a longer story involving the rEFInd Boot Manger) was running quite hot as soon as it got a bit busy. After installing and (auto-)configuring lm-sensors it turned out the fan was running at low speed despite the high temperatures.

I had at some point replaced the SSD I had in there (instead of the original SATA disk) with a different model. There's a temperature sensor glued to the front of the disk, which is a feed to the automatic fan speed management. The original disk (and the older SSD) had a metal casing, which acted as a heat transport to the sensor. The new SSD is enclosed in plastic instead, and while I put the sensor to the same position, it never measured the temperatures required to spin up the system fan.

For now, I ended up placing the sensor to the rear on top of the shielding of the Wifi module instead (the chipset heat sink didn't seem like a good idea). This seems to work well enough, and the fan now actually turns up when required:

$ sensors
applesmc-isa-0300
Adapter: ISA adapter
Master :     2424 RPM  (min = 1500 RPM, max = 5500 RPM)
TC0D:         +46.2°C  
TC0H:         +45.8°C  
TC0P:         +47.0°C  
TH0P:         +59.5°C  
TN0D:         +68.8°C  
TN0P:         +57.5°C  
TW0P:         +56.5°C  
Tm0P:        -128.0°C  

nouveau-pci-0200
Adapter: PCI adapter
temp1:        +68.0°C  (high = +95.0°C, hyst =  +3.0°C)
                       (crit = +100.0°C, hyst =  +2.0°C)
                       (emerg = +135.0°C, hyst =  +5.0°C)

coretemp-isa-0000
Adapter: ISA adapter
Core 0:       +44.0°C  (high = +105.0°C, crit = +105.0°C)
Core 1:       +46.0°C  (high = +105.0°C, crit = +105.0°C)

I recently noticed that the Vivaldi browser for Android adds the OS version and the type of the device being used to the user agent string sent to web servers.

It is possible to disable this behaviour in one of the advanced flags though, by opening vivaldi://flags from the address bar, and searching for the freeze-user-agent flag. When this setting is enabled, Vivaldi will always use "Android 9; Unspecified Device" in the browser user agent.

A few years ago, I set up a copy of Shaarli, but ended up never using it. My initial idea was to have it as a central repository for bookmarks, but then Floccus showed up as a way more convenient solution for bookmark sync. I also thought about using it for sharing links with short summaries (similar to the posts on here, or on G+ before), but then the Fediverse came along and soaked up most up my online activity.

I didn't really pick up the link sharing habit over there - wasn't shure about how to blend it in with my other posts. I thought about creating a second account just for that, but keeping multiple accounts is too much of a hassle for me (which also sunk the idea to use language-specific alts or fan out by general topics). Didn't go back up using Tiki either, as I do much of my reading on mobile devices nowadays, and writing Tiki posts from there is kind of annoying. With Shaarli, I can use the Shaarlier app to quickly drop links and a short description.

Not sure how this will end up in the longer run, but it's a try.

My Shaarli is over here, and also has an own separate RSS feed.

After Microsoft had deprecated the Microsoft Message Analyzer (cache) tool last year, there was no good way to work with network traces generated on Windows systems with the netsh trace ... command.

Microsoft has now released ETL2PCAPNG (cache), a small program that converts ETL files to PCAPNG, see this announcement on the Microsoft Core Infrastructure and Security blog (cache), so trace files can now be processed with standard utilities like Wireshark.

On my new Blackberry Key2 LE, I have run into a system notification when certain bluetooth headphones are connected:



Great. What do I care? Let's disable this notification.



Helpfully, the message shows the name of the Android component, com.qualcomm.qtil.aptxui

So I enabled developer mode on the phone, connected from the PC via adb shell, and killed off the whole thing:

bbe100:/ $ pm disable-user --user 0 com.qualcomm.qtil.aptxui
Package com.qualcomm.qtil.aptxui new state: disabled-user

I have not noticed any negative consequences.

But, really, what is this idiocy with hardware vendors injecting ads into the system? Qualcomm has 37 packages on the system, according to a pm list packages | grep qualcomm. I'm not inclined to trust them very much, but unfortunately there's little to no documentation on this kind of vendor crap on Android.

I took the opportunity to disable another obvious one, ''com.qualcomm.qti.haven.telemetry.service".

Last night, my 2.4GHz FRITZ!WLAN Repeater 310 upgraded to the new FritzOS release 7.01. The repeater is configured as FRITZ Mesh device (connected to a FRITZ!Box 7490 as mesh master). I'm currently tinkering with a Blackberry Key2 LE that I bought to replace the aging S4 mini that I used up to now (more on that later, maybe). After upgrading the repeater, the Key2 LE couldn't connect to the wifi network anymore. In an area only covered by the repeater, it didn't even see the SSID anymore. It was still possible to connect directly to the base station though, when moving into range. All the other devices I had available had no problems connecting through the repeater either.

As AVM doesn't currently provide downgrade images for the Repeater 310, I tried a factory reset and re-establishing the mesh connection, to no avail.

I then started playing with wifi network settings, and as it turns out, there's a simple fix: Don't use channel 13 on the repeater.

After excluding channels 12 and 13 from the wifi channel selection on the FRITZ!Box, the system switched down to channel 11, and suddenly the network appeard back in view on the Blackberry device. Also tried manually setting channel 6, which worked too.

Not sure what to make of that. Channel 13 was fine with this device before the FritzOS upgrade (yes, I know using it is not recommended, but this is the first time that created a problem for me).

I was not aware that RFC6762 (cache) reserves the ".local" TLD for exclusive use for Link-Local addresses with Multicast DNS:



Lutz Donnerhacke points to this in a German language article that explains how this leads to problems accessing a CIFS mount in a Windows domain that uses the .local TLD.

Details here: The default OpenSSH key encryption is worse than plaintext (cache)



Ssh keypairs for Ed25519 use a new format to encrypt the passphrase. Since 2013, it's been possible to create RSA keys with new-format passphrase encryption using ssh_keygen -o, but since that's not been the default, I don't assume anyone has ever used that (I haven't).

Might be worth replacing all RSA keypairs for pubkey authentication (and remove the corresponding public key from any authorized_keys files on all destination systems) - and all Ed25519 keys that use the same passphrase. Unless you're absolutely certain no one ever had access to the private key, in which case just upgrading the passphrase encryption using ssh-keygen -p -o -f <PRIVATEKEY> might be good enough.

I've been pulling in some posts from my Mastodon instance to the blog, using iframes to directly show the embed view provided my Mastodon.

There's two problems with that:

One is, that someone who is using a content blocker like uBlock Origin or uMatrix likely won't see anything in these posts when they're blocking third party content by default. The other is - what happens when my Mastodon instance is down?

So I've now added direct links to each of the embedded posts, serving a dual purpose: There's first party content (the link), and also the Tiki engine creates a cache entry that can be accessed when the Mastodon server is unreachable.

I've also tagged all affected posts as mastodon embed, so I can easily find them and replace the content should I ever plan to shut down my Mastodon system.

...just delete all files from DCIM/.thumbnails/ :


(from https://mastodon.infra.de/@galaxis/100335009279539395 (cache))

(from https://mastodon.infra.de/@galaxis/100323013414993622 (cache))

So far, I've run into two major annoyances with Windows 10 1803...

First is that the mobile hotspot function (sharing a cellular data connection to other devices via Wifi) has been gimped, and trying to activate it just results in an "To share your connection, you need to add this feature to your cellular data plan first." - message:


Information around this is very scarce at this time - just a handful of search engine hits. From a post on the italian Microsoft forums, it seems you now need a Store app published by your mobile phone provider to use this feature. WTF, really? I've been tinkering for quite some time to find out where Windows looks up if using the hotspot is allowed or not, to no avail. Also see over here on Mastodon.

The other problem is that powersaving for the first gen switchable graphics in my old notebook doesn't seem to work anymore. In previous versions of W10, the integrated Intel graphics would be used on battery, using a lot less power than the AMD graphics chip.

As for minor niggles, it's not possible anymore to unconditionally set a network connection as "metered" to restrict Windows updates and background data - you'll have to set a daily or monthly data limit for the network, and then tell the system to always restrict background data (instead of within 10% of reaching the limit).


It really seems that W10 1709 was the sweet spot for this old notebook, but after I didn't see any immediate problems with the new version in the week directly after upgrading, I already deleted to rollback version...

...we haven't been able to sleep in the bedroom with a new PAX wardrobe for a week.


(from https://mastodon.infra.de/@galaxis/99633409151790924 (cache))

Today I learned (cache) that USRobotics (cache) still exists (as a division of another company), and they still make a version of the classic Courier 56k modem (cache). Not that I need one of those. But I'm surprised it's still possible to buy them (and they look just the same as 20 years ago).

...is a major pain, since animated GIFs don't have a fixed frame rate: The animation can define a variable pause after each individual frame.

jzw has updated his resize.pl script (cache) with a new function to get around that limitation:



(via tedu)

...cleaning up those when the referenced Mastodon instanced go offline one day will be a major pain... Maybe I should just use screenshots instead?

(A couple of days ago.)

Was completely painless again - I'll still need to change my rspamd configuration to not use the external rmilter anymore, and there's a minor problem with syspatch:


(from https://mastodon.infra.de/@galaxis/98810053668102425 (cache))


(from https://mastodon.social/@phessler/98821962569631460 (cache))

(from https://mastodon.infra.de/@galaxis/346946 (cache))

I'm still using the same Acer Aspire 3820G laptop that I bought over six years ago, though I've switched to Windows 10 during the "free upgrade" time. Mostly everything works, despite missing vendor drivers. The function keys (sleep, sound volume, etc.) have basic support, but I recently noticed that the wireless switch only toggles through a handful of states, none of which have both Wifi and Cellular modem enabled.

For Windows 7, Acer provided Dritek Launch Manager to enable or disable wireless functions (Wifi, Bluetooth, Cellular), but version 4.0.5 doesn't install successfully on Windows 10 - and newer releases don't support the 3820G hardware.

After some experimentation it turns out that just starting the Launch Manager Setup.exe in compatibility mode for Windows 7 (right-click, select "Troubleshoot compatibility") surprisingly does the trick - installer completes, and after a reboot the wireless function key actually starts Launch Manager instead of driving the Win10 builtin toggle.


...so now I can use the Cellular modem and provide a Wifi hotspot at the same time.

...originating from a challenge to build a working game of Tetris in Conway's Game of Life (cache) on the Code Golf stackexchange. Somewhat similar to all the Minecraft CPUs I guess?


(from https://mastodon.xyz/@garrett/8374324 (cache))