Anders Murphy
@andersmurphy.bsky.social
You know it's going to be a good day as dev when you get a message like this. This is what real engineering looks like:
December 12, 2025 at 7:03 AM
You know it's going to be a good day as dev when you get a message like this. This is what real engineering looks like:
With #brotli over SSE, the window size is everything. If your window is thrashing (too small), you can use 100%+ more CPU.
This also applies to compression. So you can use brotli 3 rather than 4 (recommend normally) have almost the same compression and be 40-50% faster.
This also applies to compression. So you can use brotli 3 rather than 4 (recommend normally) have almost the same compression and be 40-50% faster.
November 23, 2025 at 4:50 PM
With #brotli over SSE, the window size is everything. If your window is thrashing (too small), you can use 100%+ more CPU.
This also applies to compression. So you can use brotli 3 rather than 4 (recommend normally) have almost the same compression and be 40-50% faster.
This also applies to compression. So you can use brotli 3 rather than 4 (recommend normally) have almost the same compression and be 40-50% faster.
Love finding random checkbox art
October 26, 2025 at 12:43 PM
Love finding random checkbox art
immersive arts in action love when people get creative like this!
October 2, 2025 at 5:52 PM
immersive arts in action love when people get creative like this!
`select count(*)` for the win.
August 12, 2025 at 8:52 AM
`select count(*)` for the win.
Today I learnt some QR scanners (google lens) struggle/fail with inverted QR codes and generally a lot of scanners scan non inverted QR codes faster.
Also CSS is awesome:
backdrop-filter: blur(5px)
Also CSS is awesome:
backdrop-filter: blur(5px)
July 27, 2025 at 7:35 PM
Today I learnt some QR scanners (google lens) struggle/fail with inverted QR codes and generally a lot of scanners scan non inverted QR codes faster.
Also CSS is awesome:
backdrop-filter: blur(5px)
Also CSS is awesome:
backdrop-filter: blur(5px)
If you want to checkout Dundee Founder Collective you can find them here:
lu.ma/dll9xxro
lu.ma/dll9xxro
July 27, 2025 at 7:48 AM
If you want to checkout Dundee Founder Collective you can find them here:
lu.ma/dll9xxro
lu.ma/dll9xxro
Here's the code to convert the QR Codes to SVG.
July 27, 2025 at 7:48 AM
Here's the code to convert the QR Codes to SVG.
Had a load of fun at Maker Gonna Make Dundee. Ended up making a location share QR code (after spending a morning messing with quadkeys).
The codes are generated on the server, converted to SVG and then morphed in to DOM via SSE (using #Datastar). They compress nicely too.
The codes are generated on the server, converted to SVG and then morphed in to DOM via SSE (using #Datastar). They compress nicely too.
July 27, 2025 at 7:48 AM
Had a load of fun at Maker Gonna Make Dundee. Ended up making a location share QR code (after spending a morning messing with quadkeys).
The codes are generated on the server, converted to SVG and then morphed in to DOM via SSE (using #Datastar). They compress nicely too.
The codes are generated on the server, converted to SVG and then morphed in to DOM via SSE (using #Datastar). They compress nicely too.
Why TCP slow start makes minification matter. 300ms slower when going from a 10kb to 18kb ( #Datastar unminified), both with brotli level 11. What's worse this is on a good connection if it was 3G with higher packet loss the difference would be even larger.
June 20, 2025 at 12:14 PM
Why TCP slow start makes minification matter. 300ms slower when going from a 10kb to 18kb ( #Datastar unminified), both with brotli level 11. What's worse this is on a good connection if it was 3G with higher packet loss the difference would be even larger.
In the first frame most of the work is done closing the compressor (somethings not quite right, either in user land or the library implementation). In the second frame (a different library) there's no such close overhead. 5X faster reads. Another win for performance profiling.
June 19, 2025 at 6:37 AM
In the first frame most of the work is done closing the compressor (somethings not quite right, either in user land or the library implementation). In the second frame (a different library) there's no such close overhead. 5X faster reads. Another win for performance profiling.
More checkbox art.
June 18, 2025 at 2:32 PM
More checkbox art.
Nothing beats profiling in production especially when you realise you are spending a large chunk of your CPU on something you can inline cache.🔥 #flamegraphs
June 16, 2025 at 10:52 AM
Nothing beats profiling in production especially when you realise you are spending a large chunk of your CPU on something you can inline cache.🔥 #flamegraphs
Compression vs diffing it's not as intuitive as you think.
Sending down 2304+ divs (85041 bytes uncompressed ) over SSE and relying on #brotli compression instead of diffing results in a checkbox change being 13 bytes.
Sending down 2304+ divs (85041 bytes uncompressed ) over SSE and relying on #brotli compression instead of diffing results in a checkbox change being 13 bytes.
June 9, 2025 at 1:18 PM
Compression vs diffing it's not as intuitive as you think.
Sending down 2304+ divs (85041 bytes uncompressed ) over SSE and relying on #brotli compression instead of diffing results in a checkbox change being 13 bytes.
Sending down 2304+ divs (85041 bytes uncompressed ) over SSE and relying on #brotli compression instead of diffing results in a checkbox change being 13 bytes.
And we've been able to do this since 1998.
May 21, 2025 at 10:03 AM
And we've been able to do this since 1998.
#Emacs Tramp + Dired is amazing. Just compiled SQLite on my server and now I can transparently copy it to my local machine. But, as a user it's just like I'm just working on a single machine.
May 21, 2025 at 10:03 AM
#Emacs Tramp + Dired is amazing. Just compiled SQLite on my server and now I can transparently copy it to my local machine. But, as a user it's just like I'm just working on a single machine.