CPP, DATABASE, INFRASTRUCTURE, JOURNAL, MONTHLY-UPDATE, TOOLS

My Tech Journal - July 2026

My Tech Journal - July 2026 Edition

C++

  • C++20: An (Almost) Complete Overview - Marc Gregoire - CppCon 2020 C++20 introduced four major features: modules, ranges, Coroutines and Concepts.
    1. modules
    2. Ranges
      Range
      • is an object referring to a sequence / range of elements
      • is similar to begin/end iterator pair (but does not replace them)
      • allows range adapters to lazily transform/filter underlying sequences of elements
      • All standard library algorithms that accept ranges also support
        • Projection (a callback that transforms elements before handing over to the algorithm)
        • Views (transform/filter range: lazily evaluated, non-owning, non-mutating)
        • Range factories (construct views to produce values on demand e.g. sequence of integers)
        • Pipelining (views can be chained using pipes ‘|’ just like the Unix pipe feature)
    3. Coroutines
    4. Concepts
      • Concepts are a mechanism for specifying requirements / contract for template parameters.
      • The predicates are evaluated at compile time.
      • This feature is more helpful for library developers.
  • Build an AI Coding Assistant in C++ - Workshop Preview by Jody Hagins
    • This is an excellent presentation.
    • Jody Hagins explains that
      • LLMs are trained to predict the next token. Everything the model outputs is generated 1 token at a time.
      • Models are essentially stateless.
    • 2 things that Make or Break AI generated code:
      1. Context Management
        • AI models have finite “working memory”.
        • As it fills, the model loses track of instructions, file contents, and prior decisions.

        Monitor context usage - when it hits 55-60%, save the state and restart with a clean context.

      2. Deterministic Workflows
        • AI models are probabilistic. The model may skip steps or take unexpected paths.
        • Keep all control flow external - scripts to run tests, linters, builds, and feed results back into the model.

        This is especially important for complex coding tasks.

    • The best advice is
      • If possible, provide sample code. The model will use that for reference and generate code with similar style / patterns.
      • Use distinct code review personas (each with specific responsibilities)

Database

Infrastructure

  • OSDI ‘20 - The CacheLib Caching Engine: Design and Experiences at Scale
    • Caching is pervasive, but the systems defer along several axes - performance goals, system topology (in-process vs remote), WL, domain-specific features …

    • The caches have different goals

      Cache Type Use-Case Special Considerations
      CDN cache at the edge static content prioritize low latency; strict TTL to prevent stale content
      Lookaside key-value session info; user profile high throughput
      SocialGraph   strict consistency requirements
      Storage user generated content large capcity
    • CacheLib: website Repo
      • General-purpose caching engine
      • High capacity caches
      • Rich feature set
      • Aggregates optimizations from across the teams
  • AWS re:Invent 2023 - Dive deep on Amazon S3 (STG314)

Interesting Projects, Code Repositories

Interesting Tools