NIGHTMARE ON AUTOPILOT (HELP NEEDED)
The incredible @TeslaBest Lars Strandridder just had his "Best in Tesla" YouTube channel hacked.
Watching years of hard work drive away on autopilot (or worse, teleoperated) is heartbreaking. If any of you have contacts at Google, or have recovered from such a hack, please DM Lars (or me if you prefer).
And everyone: Please subscribe to Lars' new channel right away (link below), and consider watching one of his latest videos to help him hit monetization fast.
We at Tesla Fam always have each other's backs—and honestly, his content is among the absolute best out there. It's time well spent!
I already subbed. How about you?
Also - please share/like this post, or make your own copy. Let's let Lars crush this one, not the bad guys.
https://t.co/bDEjJM7rYK
Yesterday I was in a CSI episode. Today I am in an Agatha Christie novel…
PDB with two identical types `std::__non_rtti_object` that got duplicated.
Same hierarchies, same bases, same members → same hash.
Yet the compiler reports a different sizeof.
"Just keep swimming… just keep swimming…" 🐠
Deduplicate identical types, drop unreferenced ones, and strip compiler-generated junk — the literal millions of types across all PDBs are collapsing by orders of magnitude.
This is letting me scale fast.
PDB files are a crime scene, and I’m the CSI guy.
Today’s subject:
TSizedInlineAllocator<1,32,TSizedDefaultAllocator<32>>
This UE5 type is duplicated… 69 times.
All from the same source file + line.
Why? mini🧵
There are 70 .cpp files (one type is shared between two modules) that include the file and instantiate the template at line 893. Then those generate a .obj that is finally merged into the DLL.
Same story for 20+ other parameter combos (48, 64, 96…).
Hundreds of identical template instantiations are scattered everywhere.
My fix?
I’m bypassing PDB type hashing entirely and using XXH3 to hash the full type content myself: fast, 100 % reliable, zero surprises.
Moral: always try existing tech first… but when it fails you, don’t waste time fixing it or looking for workarounds.
Build (or pick) an alternative you fully control.
End of war story! 🔥
Another day in the paradise of PDB files…
Just hit a wild corner case that breaks “type hashes” in UE5’s UnrealEditor-Core.pdb.
Two different types… same hash, and it is NOT a hash collision. Can you guess why this happens? 👀
Check this out ↓
Here’s the actual LLVM code that computes the type hash (It mimics Visual Studio behaviour).
It’s rather intricate, and the bug triggers because the compiler generated a weird type name that isn’t “<unnamed-tag>”-like.
Result? The algorithm falls back to hashing ONLY the name (hashStringV1). Same name → same hash → chaos when parsing.
Works 99 % of the time… until UE5 hits this exact edge case 🤦♂️
So what now?
Qué hace Europa mientras Israel asesina en Gaza y prepara otra actuación en Eurovisión
El genocidio continua en Gaza y se conocen nuevos casos de detenciones ilegales y torturas.
Texto de @olgarodriguezfr
https://t.co/femnjBXpMC
@xai I believe we’re still at least 10 years away from this AI-native paradigm hitting and scaling globally, though.
So, we’re safe... kind of... for now. 😅
🧵 Tons of debate on AI vs. software engineering lately.
The more I work with AI, the more I'm convinced that to truly replace devs, we need an AI-native paradigm and not our human one.
So, we're safe... kind of... for now.
Let me explain...