Ahh the illusion of nonblocking execution on consumer devices. The moment the os decides to preempt your thread, there no longer exists a non blocking interface. A single thread never gets a guarantee of execution, so waiting for some exact point in time is failed design

A nonblocking te isn’t gonna see 100% of snaps your just being disingenuous with that comparison tryna get some gotcha but snap% abs does tell a lot of the story role is a every snap guy who can get open against man and zone and burden is a wr who gets mostly drawn up plays out

1/ PostgreSQL 19 Beta 1 is not a “new syntax” release. It is more interesting than that. A lot of the work is about turning operational pain into database primitives: AIO workers, parallel autovacuum, online checksums, nonblocking repack, better logical replication, replica read-your-writes. That is where mature infrastructure gets better. #PostgreSQL #Databases 2/ The performance theme is practical, not flashy. Postgres 19 builds on the async I/O subsystem introduced in 18: `io_method=worker` can now scale I/O workers with `io_min_workers` / `io_max_workers`. It also claims up to 2x better insert performance when foreign key checks are present. That matters because FK-heavy schemas are exactly where “just denormalize it” is usually the wrong answer. 3/ The maintenance changes are the part I would watch closest. Autovacuum can use parallel workers. Vacuum has a new scoring system to prioritize tables. Queries can mark pages all-visible, reducing future vacuum work. And the new `REPACK` command, including `CONCURRENTLY`, is aimed at rebuilding tables with less operational overhead. This is the unglamorous side of database engineering: making the thing survive real workloads. #OpenSource 4/ PostgreSQL 19 also keeps pushing replication and migration closer to first-class workflows. Logical replication can replicate sequence values. `CREATE PUBLICATION ... EXCEPT` lets you publish everything except selected tables. `CREATE SUBSCRIPTION ... SERVER` simplifies credential management. And `WAIT FOR LSN` gives applications a cleaner “read your writes from a replica” pattern. That is a big deal for teams trying to scale reads without quietly accepting stale UX. #SoftwareArchitecture 5/ The security/compatibility changes are opinionated too. - MD5 auth now warns after successful login - RADIUS support is removed - SNI support arrives via `pg_hosts.conf` - JIT is disabled by default because costing was unreliable - `default_toast_compression` moves to lz4 - beta includes protocol “grease mode” to catch ecosystem bugs Postgres is doing what good infrastructure projects do: removing footguns slowly, in public, with migration paths. #DevOps 6/ My take: PostgreSQL 19 is a reminder that the best database releases are not always about one killer feature. They are about collapsing a hundred small production chores into safer defaults, sharper observability, and fewer 3 a.m. runbooks. That is how boring infrastructure earns trust. #CloudComputing

As a total Zig newbie, it feels like the exact things you’re considering as potential rust traits (event loop tasks, process exit callbacks, and nonblocking file I/O) are what the Zig team has been actively reworking and improving lately. (Looking at 0.16 and the roadmap). Interesting. Once those APIs mature, I see a parallel universe in which I'll read headlines of Bun being ported from Rust back to Zig 😂 Would be kinda poetic… and probably only take a week.

one of the tricky things about the rust port is layering. it’s currently many dozens of crates, which speeds up compile times but blocks cyclic dependencies. a lot of bun’s zig codebase uses tagged pointers for interfaces, for things like: - event loop tasks - process exit callbacks - nonblocking file i/o it sounds like traits are the idiomatic rust approach for this, but traits are costly - they don’t get devirtualized at compile-time or through LTO. same for fn pointers. in zig we didn’t split up packages so this wasn’t a problem. But I want compile times to be faster and ideally some parts of bun can be a crate for others to use without pulling in the whole thing. can someone with rust experience suggest a better approach that has ~no runtime or memory overhead and is visible to the compiler/linker? what the current rust code does is: - macro that generates extern “Rust” declarations for each type - release build as single codegen unit and enable LTO - no shared libraries, rely on linker’s LTO to do the inlining into direct function calls based on the tag this seems to work but it sounds uncommon/messy? is there a better approach that avoids making it slower?

youtu.be/hFWiUsG13ZA Simple update. NonBlocking hashmap, LRU, resizing. How-to YouTube. Sea-of-Nodes, phases, GCM, x86 ops, exec binary, risc5 & arm. LLVM JIT, C2, as a lib, PyPy; profile-vs-JIT. Simple design. Security risks for analysis in binary. Range types, lattice

Something about "still nonblocking" following "instant local WAV bongo" is taking me out

Claude Code CLI 2.1.122 changelog: New features: • Added ANTHROPIC_BEDROCK_SERVICE_TIER environment variable to select a Bedrock service tier (default, flex, or priority), sent as the X-Amzn-Bedrock-Service-Tier header • OpenTelemetry: added claude_code.at_mention log event for @-mention resolution Fixes: • Fixed /branch producing forks that fail with "tool_use ids were found without tool_result blocks" when the source session contained entries from rewound timelines • Fixed /model not showing the Effort option for Bedrock application inference profile ARNs, and those ARNs not receiving output_config.effort • Fixed Vertex AI / Bedrock returning invalid_request_error: output_config: Extra inputs are not permitted on session-title generation and other structured-output queries • Fixed Vertex AI count_tokens endpoint returning 400 errors for users behind proxy gateways • Fixed spinnerTipsOverride.excludeDefault not suppressing the time-based spinner tips • Fixed ToolSearch missing MCP tools that connected after session start in nonblocking mode • Fixed !exit / !quit in bash mode terminating the CLI instead of running as a shell command • Fixed images sent to newer models being resized to 2576px per side instead of the correct 2000px maximum • Fixed remote control session idle status redrawing twice per second, which could flood tmux -CC control pipes and pause the terminal • Fixed assistant messages appearing blank in some sessions due to a stale view preference • Fixed a malformed hooks entry in settings.json no longer invalidating the entire file Security/safety change: • Clarified the /mcp message shown when an MCP server is still unauthorized after the browser sign-in flow Other changes: • Pasting a PR URL into the /resume search box now finds the session that created that PR (GitHub, GitHub Enterprise, GitLab, and Bitbucket) • /mcp now shows claude.ai connectors hidden by a manually-added server with the same URL, with a hint to remove the duplicate • OpenTelemetry: numeric attributes on api_request/api_error log events are now emitted as numbers, not strings • Voice mode: keybindings bound to Caps Lock now show an error since terminals don't deliver Caps Lock as a key event Source: github.com/anthropics/claude…

Top System design concepts for interview in 2026 : BASIC ____________ - Application, Process and Thread - How to Create a Thread - How to Join a Thread - How to Interrupt a Thread - Threads Synchronization - Happens-before Relationship - Locks Granularity - Volatile Variables - Atomic Variables - Wait/Notify and the Producer-Consumer Pattern - ThreadLocal Variables - Race Condition - Deadlock - Starvation - Livelock ADVANCED _____________ - Lock and ReentrantLock - Read-Write Lock - Condition Variables - Semaphore - CyclicBarrier and Parallel Sum -CountDownLatch and Merge-Sort Algorithm - Exchanger - Phaser -CopyOnWrite Collections - NonBlocking Queues - Blocking_Queues - ConcurrentMap - Map-Reduce Algorithm - Executors - Scheduled Tasks - ThreadPoolExecutor and ThreadsFactory - Fork-Join Pool - CompletableFuture - Parallel Streams - Spinlock and Busy Wait - Lockfree and Waitfree Algorithms - Throughput and Latency in Concurrent Applications - Profiling - Microbenchmarks with JMH

MORGAN STANLEY / GOLDMAN SACHS / JO MORGAN interview scheduled ? Go through these 39 Concepts : What's your current score out of 39 ? BASIC ____________ - Application, Process and Thread - How to Create a Thread - How to Join a Thread - How to Interrupt a Thread - Threads Synchronization - Happens-before Relationship - Locks Granularity - Volatile Variables - Atomic Variables - Wait/Notify and the Producer-Consumer Pattern - ThreadLocal Variables - Race Condition - Deadlock - Starvation - Livelock ADVANCED _____________ - Lock and ReentrantLock - Read-Write Lock - Condition Variables - Semaphore - CyclicBarrier and Parallel Sum -CountDownLatch and Merge-Sort Algorithm - Exchanger - Phaser -CopyOnWrite Collections - NonBlocking Queues - Blocking_Queues - ConcurrentMap - Map-Reduce Algorithm - Executors - Scheduled Tasks - ThreadPoolExecutor and ThreadsFactory - Fork-Join Pool - CompletableFuture - Parallel Streams - Spinlock and Busy Wait - Lockfree and Waitfree Algorithms - Throughput and Latency in Concurrent Applications - Profiling - Microbenchmarks with JMH

youtu.be/7QWHRY07W2U Hashmap perf, concurrent vs nonblocking, choosing your metric. Performance antipatterns (presentation), logging, cost of traces. Exceptions, error codes, stack traces in distributed system, deadlines,exceptions from other threads

I get why people love pi now! Just built out a codex/claude subagents experience in a few hours for my pi, including: - spawning subagents with depth - codex/claude like subagent ui exp - nonblocking subagents - subagent completion notification steer - follow up with subagents - wait on subagents hope you find it useful pi install npm:pi-agents-pool

Labeling Google as CIA is understating what they are dramatically. Its actually the department of defense (war). When I was there, we had military scientists on permanent assignment. Their badges were blue. We also had people like Vint Cerf on staff and the CEO was Eric Schmidt. Google itself, I believe, was the private contract for a military project. They did not invent the algorithms for search, they were actually based on formula first formally published in 1976. Its just that the 1990s was the first time we had the computational power to build it at scale. Each original corporation on the prism slides represent an aspect of computation and computer science. The NSA later built a data center in Utah based on the prototype design Google represented. Look up room 641a, every single major internet backbone is mirrored into the department of defense and Google itself is now part of the core network where we installed PoPs in every major ISP to offload traffic into the Google network which is a clos design. The entire backbone represents an optical processor (the real meaning of prism?) where there is more bandwidth between the network card and the RAM than the RAM has to the CPU. With true nonblocking network between using electro optical modulators, it's able to run an arbitrary precision realtime optical processor across the network backbone itself and I believe it's highly likely this is running an actual AI versus the chat bots we are used to.

INNOVATION: While traditional GPU network topology with 51.2T Spectrum-X switches can only fit 8,192 GB200 NVL72 in 2 tiers, Microsoft & OpenAI will be one of the first to deploy an rail only multi plane design with 512 radix 51.2T Spectrum-X & 51.2T Arista switches that can support up to 512,000 GPUs with only 2 tiers nonblocking fabric. In addition, OpenAI's GB200 & GB300 NVL72 are both using ConnectX-8 which will runs OpenAI’s custom protocol MultiPath Reliable Connection (MRC) instead of running RoCEv2 or UET protocol.

As best I can tell, it's parallel storage attached over nonblocking frontend Ethernet. There's some magic happening inside BlueField-4, and it exposes an interface into the GPU host with which NIXL can interact.

macbooks got so prevalent. those websites are only terrible without buttery rendering and mouse wheel scrolling. also, if you don’t hijack but read scroll pos nonblocking it’s much better

Your run game will get a lot better with this nonblocking tight end gone

🏆 𝗖𝗼𝗻𝘁𝗲𝘀𝘁 𝗕𝘂𝗴 𝗗𝗶𝗴𝗲𝘀𝘁 - 𝗣𝗧𝟬𝟭𝟱𝟴 🏆 𝗖𝗼𝗻𝘁𝗲𝘀𝘁: UXD Protocol | @sherlockdefi 𝗙𝗶𝗻𝗱𝗶𝗻𝗴 𝗜𝗗: H-02 𝗗𝘂𝗽: 0 𝗣𝗮𝘆𝗼𝘂𝘁: 7,677 Bug Summary: An attacker can permanently block LayerZero message channels between chains with different gas limits by sending a transaction with an excessively large _toAddress parameter from a high gas-limit chain to a lower gas-limit chain, causing the destination chain to be unable to process the message due to gas constraints, which breaks the nonblocking mechanism and reverts to blocking behavior. Heuristics: When cross-chain protocols accept arbitrary-length calldata that gets transmitted between chains with different gas limits, verify that payload size is validated before transmission to prevent unprocessable messages on the destination chain that can permanently block message queues. Bug Signature: -Arbitrary-length calldata accepted without size validation -Cross-chain gas limit asymmetry exploitation -Message queue blocking through unprocessable transactions -DoS via resource exhaustion on destination chain -Bypassing nonblocking safeguards through oversized payload injection 𝗥𝗲𝗳 𝟭: github.com/sherlock-audit/20… 𝗥𝗲𝗳 𝟮: audix.ai/checklist