标签：#code

1. Introduction and Problem Statement A good way to learn what a compiler really does when transforming a C source code into a binary is to disassemble the binary and compare it with the C source code. It is especially true for 8 bits microcontrollers like the 8051. In order to test SDCC we are going to use the following C source code. /* ========================================================================== * * Universal Test Corpus - Heterogeneous Architecture Analysis * * ========================================================================== */ #include <stdint.h> // 1. Global variables (testing absolute/relative addressing modes) volatile uint32_t global_var_32 = 0xDEADBEEF ; volatile uint8_t global_var_8 = 0x42 ; const char string_const [] = "TARGET_STRING" ; // 2. Function with parameter passing and local variables (stack / Frame Pointer test) int32_t callee_function ( int16_t a , int16_t b ) { volatile int32_t local_result = 0 ; // Basic and mixed arithmetic operations (8, 16, 32 bits) local_result += ( int32_t )( a * b ); local_result -= ( int32_t )( a / ( b | 1 )); // Avoid division by zero // Shift tests and logical operations (highly variable depending on ISAs) local_result = ( local_result << 2 ) ^ 0x55AA55AA ; local_result = ( local_result >> 1 ) | ( int32_t ) global_var_8 ; return local_result ; } // 3. Main function grouping complex control flows int main ( void ) { volatile int32_t accumulator = 0 ; int16_t i ; // Loop test (Conditional jumps, decrement, comparison tests) for ( i = 0 ; i < 10 ; i ++ ) { if ( i == 5 ) { accumulator += 100 ; } else { accumulator += i ; } } // Multiple branching test (Switch / Jump Table or cascaded if-else) switch ( global_var_8 ) { case 0x10 : accumulator += 10 ; break ; case 0x20 : accumulator += 20 ; break ; default: accumulator -= 5 ; break ; } // Function call (Stack management, save registers Link Register/PC) accumulator += callee_function (( int16_t ) accumulator , 3 ); // Pointer and indirect memory ac

The 10-Minute Tax For many years, as a CDK developer, I'd run cdk synth , then cdk deploy , and then cross my fingers — either it deployed cleanly, or it failed somewhere in the middle of a CloudFormation run that had already been going for ten minutes: ❌ MyStack failed: UPDATE_ROLLBACK_COMPLETE Resource handler returned message: "The runtime parameter of nodejs16.x is no longer supported" (HandlerErrorCode: InvalidRequest) Ten minutes. For something CDK could have told you before it ever talked to CloudFormation. These days I let AI agents write a good chunk of my CDK code, which made this even worse — an agent can't iterate when every failed attempt costs it ten minutes. 🤖 AI Agent development loop: Attempt 1: cdk deploy → ⏱️ 10 min → ❌ deprecated runtime Attempt 2: cdk deploy → ⏱️ 10 min → ❌ invalid memory size Attempt 3: cdk deploy → ⏱️ 10 min → ❌ security group rule conflict Attempt 4: cdk deploy → ⏱️ 10 min → ✅ finally works Total time wasted: 30 minutes on things that were knowable at synth time. And if you're deploying something heavy like an Amazon EKS cluster, the penalty stretches to 25-30 minutes per failed attempt. What if the CDK could catch all of those on cdk synth — in seconds? The CDK Lifecycle: Where Validation Fits Before I show off the new validation, it helps to see where it plugs into the lifecycle every cdk deploy goes through: Stage What Happens Executed By 1. Construction Execute main.ts , call new Stack() , build the construct tree in memory CDK App (local) 2. Synth app.synth() traverses the tree, produces CloudFormation template to cdk.out/ CDK App (local) 3. Template Validation 🆕 Post-synth offline validation — default rule set + registered policy plugins CDK App (aws-cdk-lib, local) 4. Create Change Set 🆕 CloudFormation pre-deployment validation — 6 types of online checks against real account state CloudFormation (AWS) 5. Execute Change Set CloudFormation provisions/updates/deletes actual AWS resources CloudFormation (AWS) The gap was a

Spawning a subagent in Claude Code feels free. It isn't. We measured it across a real review pipeline, and the number that matters is one almost nobody talks about: each subagent costs roughly 436,000 tokens in fixed overhead before it does any useful work. This post explains where that number comes from, how to reproduce the measurement on your own setup, and what it changes about how you should split work between agents. The experiment We run a weekly review pipeline over a catalog of digital products (Markdown-heavy repos: rules files, skills, templates). The pipeline embeds each product's full content into a reviewer prompt and asks for structured findings. We ran the same product, same full content, two ways: Arm A: three subagents , one per review perspective (buyer value, niche accuracy, compliance). Total prompt size: ~314k characters. Arm B: one subagent covering all three perspectives in sequence. Total prompt size: ~105k characters. Billed token totals, from the session transcript: Arm A (3 agents) Arm B (1 agent) Total tokens 2,150,310 809,070 Distinct defect classes found 20 11 Primary-source fetches performed 0 2 Arm B cost 37.6% of Arm A. The naive expectation — "three agents read the same content, so about 3x" — roughly holds, but the reason is not the content. Where the tokens actually go Breaking the transcript down per turn, each agent carried about 436k tokens of overhead that had nothing to do with the review itself : the initial context load at spin-up plus the cache write on its final turn. The embedded product content — the thing we assumed dominated cost — was only about 46k tokens per agent. That's a 9.5:1 ratio of fixed cost to payload. Two consequences fall out immediately: Embedding full content is cheap. We had been truncating embedded files to save tokens, which quietly excluded the files that carried the product's actual value from review. Full-content embedding turned out to cost almost nothing relative to what we were already paying

Ask any AI coding assistant directly whether using array index as a React key is a good idea, and it will tell you no. It will explain why. Reordering, insertion, and deletion of list items can cause React to misidentify which DOM node corresponds to which data, leading to state bugs and unnecessary re-renders. This is not obscure knowledge. It is one of the most commonly repeated pieces of React advice that exists, and every model has clearly seen it thousands of times during training. And yet, if you look through a codebase where the AI generated a meaningful portion of the list rendering, you will very likely find at least one instance of exactly this pattern. A map over an array, using the index as the key prop, sitting quietly in a component that otherwise looks perfectly reasonable. This is a strange thing to observe once you notice it. The AI is not confused about the rule. Ask it directly and it recites the correct answer immediately and confidently. But somewhere between knowing the rule in the abstract and applying it consistently during generation, something gets lost. Why knowing a rule and applying it are different things There is a meaningful difference between an AI model having encountered information during training and that information reliably surfacing during every relevant generation task. When you ask directly whether index as key is a good idea, you are prompting the model to retrieve and state a fact it has strong, well reinforced associations with. This is a different cognitive task than generating a list rendering component from scratch while simultaneously handling several other decisions about structure, naming, data shape, and styling. During active generation, the model is not running through a checklist of best practices for every line it writes. It is producing output token by token based on patterns, and in the moment of writing a map function, the path of least resistance is often exactly the pattern that gets flagged as wrong when

Today, I’m talking with Lila Snyder, who is the CEO of Bose. You certainly know Bose — it’s one of the most famous brands in all of consumer tech. The company started 60 years ago selling speakers to consumers, and its focus on research and development has led it to be a leader in both […]

As AI commoditizes code output, system comprehension silently decays, creating cognitive debt that threatens safe architectural evolution. This article explores why human understanding must be treated as an essential architectural characteristic, offering actionable strategies, socio-technical metrics, and design checkpoints to preserve intent across modern engineering teams. By Jacobus Meintjes, Narayana Rengaswamy, Paul Katsande, Sureshbabu Bikki

GitHub Code Quality is now generally available on GitHub Enterprise Cloud and GitHub Team. The service combines CodeQL analysis with AI-assisted detection of maintainability and reliability problems, then uses Copilot Autofix to suggest changes for review in pull requests, according to an announcement from GitHub. By Matt Saunders

The AI policies across the Linux ecosystem are very heterogeneous, ranging from the GCC’s restrictiveness, the Linux kernel’s pragmatism, to the more open disclosure-based utility model of Kubernetes' landscape. From core infrastructure to high-level orchestration, these distinct approaches highlight a shared commitment: ensuring the human maintainer remains the indispensable guardian of the code. By Olimpiu Pop

I have three files in this repo that shell out to something over the network or a subprocess and can fail in interesting ways: publish_devto.py , server.py , and git_commit.py . Two of them have --selftest blocks that stub the risky call and exercise the actual failure branches. One doesn't, and I only noticed because I went looking for a reason to be suspicious of my own test coverage after seeing a trending post about counting assertions in a test suite and not liking what you find. git_commit.py reads a staged diff and calls claude -p to turn it into a commit message. It has five distinct exit paths, all guarding real failure modes I've hit before in this project: try : diff = subprocess . check_output ([ " git " , " diff " , " --staged " ], text = True , timeout = 20 ) except subprocess . TimeoutExpired : print ( " git diff --staged timed out after 20s " , file = sys . stderr ) raise SystemExit ( 1 ) if not diff . strip (): print ( " Nothing staged. Run `git add` first. " ) raise SystemExit ( 1 ) try : raw = subprocess . check_output ( [ " claude " , " -p " , " --safe-mode " , SYSTEM + " \n\n " + diff ], text = True , timeout = 20 , stderr = subprocess . PIPE , ). strip () except subprocess . TimeoutExpired : print ( " claude -p timed out after 20s " , file = sys . stderr ) raise SystemExit ( 1 ) except subprocess . CalledProcessError as e : print ( f " claude -p exited { e . returncode } : { ( e . stderr or '' ). strip ()[ : 200 ] } " , file = sys . stderr ) raise SystemExit ( 1 ) except FileNotFoundError : print ( " claude CLI not found on PATH " , file = sys . stderr ) raise SystemExit ( 1 ) That's a held index lock hanging git diff , an empty staging area, a claude -p call that times out, one that exits non-zero, and one where the claude binary isn't even on PATH . Real scenarios — the timeout on this exact git diff --staged call was itself a bug I'd already found and fixed once ( docs/project_notes/bugs.md , 2026-08-06: a prior fix claimed to add a timeout

Programming with Claude Code will soon require even less human oversight.

Coder à la voix avec ChatGPT, ça marche. Le hic tient en une ligne : chaque mot que tu dictes part chez OpenAI. Depuis le 23 juillet 2026, Codex se pilote à la voix — il ouvre une pull request, cherche l'origine d'un bug, tout ça dans une phrase. Pratique pour un side-project. Rédhibitoire quand le code appartient à un client. On voulait le même confort sans la fuite. Le résultat est un pipeline 100 % local : faster-whisper pour la transcription, Claude Code et sa commande /voice pour l'agent. Rien ne sort de la machine — ni la voix, ni le contexte, ni le code. Voici la config exacte, la latence qu'on mesure sur un M2, et les deux bugs qui nous ont coûté une demi-journée. Pourquoi pas simplement Codex vocal ? Parce que « coder à la voix » cache deux choses qu'on confond tout le temps. Le mode vocal de ChatGPT est fait pour converser : il répond, il temporise, il reformule. Dicter du code, c'est l'inverse — tu veux une transcription fidèle et muette, qui ne discute pas, ne reformule pas et n'ajoute rien à ce que tu dis. Deux gestes opposés. Le vrai stack n'est donc jamais « ChatGPT vocal seul ». C'est un outil de dictée précis d'un côté, un agent de code de l'autre. Codex vocal fait les deux dans le cloud pour 20 €/mois ; un setup local sépare les deux briques et garde tout sur ta machine. Le tour d'horizon complet — prix, outils, cas d'usage — est dans le guide de référence ; ici, on reste sur le terrain technique. Le chemin le plus court : /voice Depuis mars 2026, Claude Code embarque un mode vocal. Tu tapes /voice dans le terminal, tu tiens la barre d'espace, tu parles, tu relâches. La transcription passe par un Whisper local, pas par une API distante. > /voice [hold space to talk · release to send] Pour 90 % des cas, ça suffit. Tu dictes une intention, l'agent écrit le code, tu relis. Si tu veux garder la main sur le modèle, la langue et le vocabulaire technique, il faut descendre d'un cran et brancher ta propre transcription. Le pipeline DIY, brique par brique T

I read back six months of my own work journal and found three different note-taking systems, only one of which I remember deciding to build. This is what the volume of information actually did to my notes, what got better, what I lost, and how I capture things now.

Incident management platform provider Rootly has published an account of its decision to drop its long-standing small pull request rule, arguing that the practice no longer serves its purpose now that AI agents generate most of its code. The company describes a shift from measuring PR size to assessing blast radius, with feature flags and rollback capability taking precedence over line counts. By Matt Saunders

Okay, story time. If you've worked on a full stack app where the backend is Java/Spring Boot and the frontend is React, you know the drill. Someone changes something on one side of a contract and nobody tells the other side. Weeks later you're playing detective across five files trying to figure out who calls what. And it's not just REST endpoints. It's the scheduled job that quietly writes to the same table your API touches. It's the service that calls another service, which calls another service. It's the Kafka event your controller publishes that some completely unrelated listener is consuming three modules away. All of that is "the contract" too, it's just invisible unless you go looking for it. Now add AI coding agents into that picture. They're great at writing code in the file they're looking at. They're not great at knowing that the component they're editing calls an endpoint, which hits a controller, which calls a service, which calls a repository, which is also written to by a scheduled job at 2am, which also fires an event three other services are listening for. Agents see one file at a time. So they'll happily rename a field or change a return shape on one side and leave everything downstream of it completely unaware anything changed. I got burned by this enough times that I decided to build the map myself. That's how Contour happened, and then, once I realized AI agents needed to query that map directly instead of just reading it off my screen, Contour MCP happened right after. Let's get into it. The actual problem Working across a UI, a REST API, a service layer, a repository layer, a database, plus schedulers and events sitting on top of all of it, two things go wrong constantly. Agents (and honestly, humans too) edit one side of a flow without knowing the other side exists. People burn real time reconstructing a call chain by hand, jumping through five or six files just to make a change that should be simple. Both come from the same root cause. Nobod

Today, I’m talking with Gaby Del Valle, a policy reporter here at The Verge, about the growing backlash against AI data centers. Gaby recently reported a fantastic piece about Hernando County, Florida, where last month the county commission unanimously approved a yearlong moratorium on data center construction. She attended a protest there organized by a […]

If your automations are simple and low-volume, Zapier's per-task billing is fine and the cheapest thing about it is your time. The moment a single workflow fans out into many steps, or you start running thousands of runs a month, the pricing model — not the sticker price — is what decides your bill. Make charges per module execution, which is finer-grained than a Zapier task; n8n charges per workflow execution regardless of how many steps that workflow has, and it can be self-hosted for infrastructure cost only. The switch point is almost always about billing units, not features. I've run all three in production for internal automations, and the migrations I've done were never triggered by a missing feature. They were triggered by a monthly invoice that grew faster than the value of the work being automated. This post is about spotting that inflection before the invoice does. How does each tool actually count usage? The three tools use three different meters, and conflating them is where most cost surprises come from. Zapier bills per task. A task is one action step that successfully runs. The trigger that starts a Zap does not count; every action after it does. So a Zap that watches a form and does one thing costs one task per submission. A Zap that watches a form, looks up a record, formats a value, and writes to two places costs four tasks per submission. Filters and paths that stop early generally don't consume a task, which matters more than people expect. Make bills per operation. An operation is a single module doing a single unit of work. It's conceptually similar to a Zapier task, but Make's modules are more granular and the included volumes on comparable tiers are typically much higher, so the effective cost per unit of work tends to be lower. The catch is that iterators, aggregators, and array-processing modules can multiply operations fast — a scenario that loops over 50 items can spend 50+ operations in one run. n8n bills per execution. One workflow run

JioHotstar explains the distributed architecture behind its real-time ad request workflow, covering ad decisioning, waterfall tiering, pacing algorithms, latency optimization, and service coordination required to select and deliver personalized advertisements during streaming playback at scale. By Leela Kumili

LaTeX is a document preparation system for high-quality typesetting, perfect for academic papers and technical docs. Many people turn to Overleaf as their go-to online editor for LaTeX, but it comes with its own frustrations. If you are tired of Overleaf being costly and always hitting the compile timed out error, this guide is for you! The full MacTeX install weighs in at a massive ~6.4GB, most of which you'll never actually use. Setting up a minimalist LaTeX environment on macOS using BasicTeX and VSCode is a much better alternative that makes your setup ~8 times smaller. It saves storage and makes it much easier to collaborate with your teammates using GitHub as a combo. Install LaTeX via Homebrew We'll use Homebrew to keep things manageable. If you don't have it, grab it at brew.sh . 1. Install LaTeX BasicTeX is the "lean" version of MacTeX. It's only ~140MB initially. brew install --cask basictex 2. Refresh your path and verify Make the TeX binaries available in your current terminal session: eval " $( /usr/libexec/path_helper ) " The default LaTeX compiler pdflatex should be available now. Verify it's working: which pdflatex pdflatex --version 3. Update tlmgr and packages tlmgr is the TeX Live Manager. To update tlmgr and all packages, run the following commands: sudo tlmgr update --self sudo tlmgr update --all 4. Install latexmk (build manager) latexmk is the "build manager" that handles multiple runs of the compiler (necessary for bibliographies and tables of contents). sudo tlmgr install latexmk Verify latexmk version: which latexmk latexmk --version 5. Install essential package collections BasicTeX is too bare-bones for real projects. Since we went minimalist, we need to grab only the packages we actually use. These three collections will cover 90% of your needs while keeping storage down. sudo tlmgr install collection-latexrecommended sudo tlmgr install collection-fontsrecommended sudo tlmgr install collection-latexextra Note: If a build fails due to a mi

Part of "AI, engineering and what survives production", a series on the parts of building with AI that hold up once real traffic hits them. There is a claim going round that you have probably absorbed by now: AI-assisted development is making codebases worse. Refactoring is down, duplication is up, we are all writing more and revising less. The numbers behind it are real, the samples are enormous, and I found I had started repeating the conclusion in conversation without ever having checked it. Then it occurred to me that those figures are averages taken across hundreds of millions of changes from thousands of organisations, not one of which is mine. So what is the rate in your repository? Nobody has told you, and on current evidence nobody is going to. I set out to find mine, assumed it would take an afternoon, and spent three days discovering that the answer is far harder to get at than the confident version suggests. So this is not a piece about what AI does to code. It is about how to ask that question of your own repository without arriving at a wrong answer, which turned out to be the genuinely difficult part. The tool I built to do it is git-habits : free, local, and it reads no source code whatsoever. What git can actually tell you Git history is a surprisingly rich behavioural record. Not of quality, about which it knows nothing at all, but of habits: how often you commit, how large those commits are, whether you go back and change what you wrote last month, and whether anybody still touches the old code. That is a narrower thing than quality and it is the thing the industry claims has changed, so it is the thing worth measuring. Four signals are computable from commit metadata alone, without opening a single source file: Moved lines. The share of changed lines sitting in files git detected as renamed or copied. It is the closest thing history offers to "somebody went back and reorganised this." Legacy touch. The share of changes landing on files nobody has

Empirical research from ReWeaver AI. 42 identical prompts, across 5 tools and 8 production dimensions, compared to human baseline. One metric that changes how you see drift. Everyone knows AI-generated code has quality issues. What’s less understood is that the way most teams measure those issues — by how often they occur — systematically understates the risk. We ran a controlled study to find out how badly. The answer surprised us, particularly in one dimension. What We Did We gave five leading AI coding tools (Cursor, Claude Code, Lovable, Figma Make, and VS Code with Copilot) 42 identical prompts: realistic single-component builds — buttons, forms, dashboards, navs, modals, auth surfaces. We scanned every output with ReWeaver, our deterministic drift-detection engine, across eight production readiness dimensions: User Experience Security & Privacy Accessibility Design Consistency Reliability Maintainability Architecture Testability We also scanned six human-authored open-source repositories as a reference baseline. For each dimension, we calculated two things: Drift frequency — the percentage of lines containing at least one drift occurrence. Counts what went wrong. Production Drift Ratio (PDR) . The PDR is a metric that weights frequency by estimated remediation cost on a 0–1 scale. A PDR of 0.30 is roughly 45 minutes of cleanup per component; 0.70 is about 2.5 hours. The Finding That Stopped Us In Security & Privacy , AI tools produced 3× the human drift frequency . That looks manageable — a meaningful gap, but not alarming. The PDR was 22× the human reference . Not 22% more. 22 times more costly to fix. The frequency gap makes Security & Privacy drift look like a minor concern. The PDR reveals it’s the most expensive problem in the dataset. AI-generated security drift (client-side authorization gates bypassable in DevTools, raw PII and credentials passed through props without tokenization) is syntactically identical to safe code. It passes review, but the fixe