@itanium_guy@XenoPanther I wish these kind of videos were still made. Even about tiny changes. I'd watch ~10 minute talks on... from how and why was the legacy USER32 desktop color, pattern and wallpaper support removed, to some implementation details of RIO sockets.
Yeah, it was bad.
Back in the day one of our customers run dual NetBurst Xeon (so 2×2 cores IIRC) with a whooping 8 GB RAM on 32-bit 2003 R2 and it was PAIN maintaining the SW for it. You added a field to structure somewhere and the whole thing slowed to crawl because it somewhere straggled cache line.
You might find interesting this video, where Arun Kishan (but I could've sworn it was @itanium_guy) explains how they rewrote it:
https://t.co/fU2yXIXnPh
@PetrHonzejk Pro svobodomyslnou společnost fungující na ideálech rovnosti a tolerance by mělo být neakceptovatelné importovat negramotné divochy, kterým jsou tyto ideály pro smích a považují je za známku slabosti hodnou akorát využití.
IDK about networking. We'd obviously want the best performance, but over my career the most performant way to network (in Windows) changed several times.
Nonblocking sockets -> WSAEventSelect -> overlapped -> I/O completion ports -> and finally RIO sockets.
The API will be modelled after one approach and then inevitably new, better and incompatible approach will be invented.
@phantomofearth I'm curious if the whole underlaying issue was fixed, or if it's just this one dialog doing some mitigation, and if it is that, then what is it doing precisely, so I can replicate it in my apps.