Andrew Lilley Brinker

@alilleybrinker.com

2154 Followers 430 Following

Memory safety / Rust 🦀, Dodger baseball ⚾️ ...also #jj-vcs, urbanism, and more. Helping you assess OSS risks with @hipcheck.mitre.org. Principal Engineer at MITRE (opinions are mine) blogs: alilleybrinker.com / possiblerust.com

Comment chain 1 nrc (@ncameron.org): I think you wouldn’t have to tweak rust too much to get there. It would be backwards incompatible, but not too much. Plus of course a total rewrite of the compiler and cargo, but with the help of llms, both issues are probably solvable now. Andrew Lilley Brinker (@alilleybrinker.com): I am pretty pessimistic about Rust’s cultural and organizational ability to do that, even if it’s technically feasible. Comment chain 2 huevosabio.bsky.social (@huevosabio.bsky.social): Or another way to see it, would it be possible to make Rust compile much faster? Andrew Lilley Brinker (@alilleybrinker.com): It would require large breaking changes and a degree of coordination, product direction, and follow-through which I don’t believe Rust could organizationally muster.
Economic theory tells us we should be suspicious of correlations found in observational data. In observational data, correlations are almost certainly not reflecting a causal relationship because the variables were endogenously chosen by people who were making decisions they thought were best. In pursuing some goal while facing constraints, they chose certain things that created a spurious correlation with other things. And we see this problem reflected in the potential outcomes model itself: a correlation, in order to be a measure of a causal effect, must be based on a choice that was made independent of the potential outcomes under consideration. Yet if the person is making some choice based on what she thinks is best, then it necessarily is based on potential outcomes, and the correlation does not remotely satisfy the conditions we need in order to say it is causal. To put it as bluntly as I can, economic theory says choices are endogenous, and therefore since they are, the correlations between those choices and outcomes in the aggregate will rarely, if ever, represent a causal effect.
Image of the Revzani Tank, which looks like a Jeep Wrangler that really wants to be a Halo Warthog when it grows up
In the societies of control, on the other hand, what is important is no longer either a signature or a number, but a code: the code is a password, while on the other hand the disciplinary societies are regulated by watchwords (as much from the point of view of integration as from that of resistance). The numerical language of control is made of codes that mark access to information, or reject it. We no longer find ourselves dealing with the mass/individual pair.
Hex wrench set product picture, showing many different sizes of hex wrench in two different scales.
John Oliver staring intensely
GLaDOS' Morality Core from Portal
A map of the United States from "The New Deal Map" site, with an enormous number of dots indicating distinct New Deal-funded works programs and conservation programs. There are particular concentrations in the Bay Area, Los Angeles area, and around New York City, but dots truly cover the entire country.
"‘Get good’ is not a strategy. There is a common reply in conversations about memory safety, coming from the most hardcore skeptics: Programmers should just write better code. They argue, explicitly or implicitly, that programmers who benefit from the guardrails of memory safety are bad programmers, and that real programmers are sufficiently skilled that they do not need a machine double-checking their work. Let’s be clear: This is anti-intellectual nonsense—macho self-aggrandizement masquerading as a serious technical argument. You should not take it seriously and should consider someone advancing this argument as fundamentally unserious and to be ignored."
Diagram from "Student Motivations and Expectations for an Introductory Programming Course in Biology" by Zuckerman, Vo, and Juavinett, explaining the career choice to enroll in "Introduction to Python for Biologists" as being influenced by prior experiences (prior exposure to coding, mathematics experience and gatekeeping), self-efficacy (belief in possible success, strongly informed by support, course description, and prior experiences), and social support (direct connections such as family, secondary connections, and societal figures).
From the TigerBeetle issue tracker's list of problems they have with Zig: "Language features around pinned structs, result location semantics aliasing and implicit pass by reference. We currently avoid pass by value syntax altogether, and always use *const T or *T."
Jarred posting on Hacker News about Andrew's blog post. " Jarred 3 hours ago | next [–] I have learned so much reading Andrew’s code and as I said in the original post: Bun would never have happened without Zig. > The post claims they were fuzzing their Zig code, while during our calls the whole Bun team told us that they were not fuzzing anything. This appears to be an outright fabrication. Fuzzilli integration: https://github.com/oven-sh/bun/pull/24826 Merged PRs fixing issues Fuzzilli found in Bun’s Zig code: - https://github.com/oven-sh/bun/pull/28926 - https://github.com/oven-sh/bun/pull/28934 - https://github.com/oven-sh/bun/pull/29255 - https://github.com/oven-sh/bun/pull/29210 - https://github.com/oven-sh/bun/pull/29199 Searching “Fuzzilli” shows more PRs: https://github.com/search?q=repo%3Aoven-sh%2Fbun+is%3Apr+Fuz..."
A Breville Barista Express, with the portafilter in, pulling an espresso shot.
Menu from Microsoft Teams, where "Tag mentions" is listed twice.
BDG, sunglasses on indoors, dressed entirely in gold, immediately after his win.
Slide deck from the same presentation saying “market capitalization equals egg value plus goose value, and also equals JPY 74t plus alpha”
Quote from Charity's post: "Some technologies are more damaging than others — knives are less damaging than guns, Facebook for colleges was less damaging than Facebook for Myanmar — but we always discover risks before we know how to govern them.2 There is always a gap while we try to catch up. That gap is not proof that AI is evil. It is proof that we have work to do. The fact that we have not solved the problems yet is not an argument to disconnect. It is an argument to engage, especially if you work in technology and already have an arsenal of relevant skills. You do not learn to govern a tool by refusing to touch it. You learn by using it and understanding well enough to critique it, shape it, contribute to it, and set boundaries around it. You learn how to make it boring. “Learn AI so you can complain about AI better.” I said it and I meant it. I still do."
Nintendo GameCube
Table from the linked CISA article showing timelines for remediation based on whether a vulnerability is 1) publicly exposed, 2) in the Known Exploited Vulnerabilities (KEV) list, 3) automatable by an adversary, and offers total or partial control to an attacker. In some cases the remediation timeline is 3 days with a mandatory forensic triage.
"The outrage selected on a single tail event and narrativized it" A Mastodon user noticed a regression in v3.4.3, saw Claude commits, and concluded causation. But v3.4.3 at 3.29 sev/10c is at the 77th percentile — elevated but not extreme. 8 historical releases scored higher. The correlation is noise.
"If you buy into vibe coding you shouldn't be given the same grace; like somebody who can't make straight rip cuts on a table saw."
Lisa Simpson talking to Bart, as they watch an Itchy and Scratchy film at Itchy and Scratchy Land, saying "I wonder if this kind of violence really does desensitize us."
Homer Simpson, talking to Marge and Lisa, saying "But what can I do? I'm just... one... one man." Homer had to count up to 1 with his fingers.
"Because iddqd is such a foundational data structure at Oxide, we go through great lengths to validate its correctness, including the soundness of its abstractions. There is a great deal of analytical reasoning performed by experienced Rust authors and reviewers, as described in the section above. But we also empirically validate iddqd along several different dimensions. None of the layers is sufficient by itself, but by rigorously doing all of them we can be more confident that iddqd is correct. The layers of validation we apply are: (The following is in a table) Technique: Analytical; Catches: reasoning Soundness errors Technique: Example-based tests (including doctests); Catches: Regressions, broken examples Technique: Pathological tests with Miri; Catches: Adversarial trait impls causing UB Technique: Model-based tests (PBT with NaiveMap oracle); Catches: Divergence from oracle over random operation sequences Technique: Panic safety fault injection; Catches: Invariant violations on mid-operation panics LLM adversarial review; Catches: Blind spots in adversarial user code
dril tweet: "drunk driving may kill a lot of people, but it also helps a lot of people get to work on time, so, it;s impossible to say if its bad or not,"
"Depending on the context, different programs will have different priorities, and for some it’s the legitimately correct choice to prefer performance over minimizing risk of misbehavior, in which case turning asserts into optimization opportunities makes perfect sense." — Loris Cro
From the CACM website: "The BLOG@CACM section accepts first-person submissions from an international community of bloggers on issues of importance to the computing community. Please recommend a blogger or blog topic or volunteer as a contributor at blog@cacm.acm.org."
Image from Sim City 2000 with the city planner yelling "YOU CAN'T CUT BACK ON FUNDING! YOU WILL REGRET THIS!"
Loam, an NPC from Hollow Knight: Silksong. Loam is trapped in the Underworks of the Citadel of Song, where he toils on a treadmill endlessly as part of the brutal system of oppression of the faithful to produce power for the rest of the Citadel.
Slide from the talk with the title: "tl;dr." The body reads: "Free software is a political movement rooted in the fundamental nature of software as an inexhaustible resource, dedicated to ensuring all people have the freedom to run, copy, distribute, study, change and improve software for their own ends. Open Source takes Free Software values, de-emphasizes its political nature, and highlights the incredible business impact inherent in harnessing the inexhaustible nature of software as a new development model."
Post from "Vicky ACAB" (@vickyacab.bsky.social) that reads: "Every single municipal budget in the country is unironically the Dril candles tweet but the candles are police"
Comment from the Cargo source code which reads "This typically means some filesystem modifications happened or something transitive was odd. In general we should strive to provide a better error message than this, so if you see this message a lot it likely means this method needs to be updated!"
"The Homer" car from The Simpsons, where Homer has been given the opportunity to design a new car himself, without constraints, and he's designed the worst car imaginable.
Ryan Cohen, in an interview with Andrew Ross Sorkin on CNBC, responding to questions about how GameStop could possibly afford to purchase eBay, a company worth substantially more than GameStop is, by saying "Half cash, half stock."
Section title: AI, Rigor, and Respect When using AI on <PROJECT> it's important to assess your use against two key values: rigor and respect. For rigor: we care about delivering a system we understand; one which carefully handles the possible states it may enter and inputs it may receive, and which is graceful under load and precise when reporting or handling errors. We are not building slop, and should not accept slop in the construction of our software. Insofar as we use AI, we should use it consistent with this goal to create a system which is resilient, maintainable, debuggable, and performant. For respect: we are a team, and we care about and support each other. Everyone on the team has expertise and a perspective worth hearing, and we should endeavor to bring our collective intelligence to bear in solving problems. When we collaborate, we work person-to-person, and do not substitute AI chatbots for ourselves. If we do want to incorporate the output of an AI system in a discussion with coworkers, we should clearly signpost that we're sharing the results of an AI prompt, so that we as a team can apply our collective sensibilities to assess the accuracy and validity of its output. We should never mislead each other about what material is or is not AI-generated.
Photo of the Apple Magic Mouse
Four Instagram comments. 1. Whoever shorted Ferrari stock before this release is the biggest winner here. 2. No one asked for this ⚡️ atrocity. 3. Enzo Ferrari rolling in his grave. 4. Is this a Ferrari and Swatch collab?
A screenshot of a small portion of the updated codegen unit graph, showing dropshot and camino build nodes, each with a bunch of metadata for how that build node is configured.
Extremely messy graph of codegen units, looking more like a conspiracy board than anything usable. Nodes and edges overlap in a completely illegible disaster.
The Flair 49 Pro expresso machine. This is a lever-based machine, where the user operates a lever. It's "simple" to use because the only mechanism is a lever, but "complex" because the user is entirely on their own for understanding pre-infusion, target pressure, extraction period, water ratio, etc.
Snapshot of my JJ configuration showing that `jj bookmark advance` has been configured to advance to the "closest pushable" commit, and that "closest pushable" has been defined as a commit that is a closest ancestor revision that is mutable, has a non-empty description, and is either not empty or a merge. This is done with these specific configuration settings: 1. `revsets.bookmark-advance-to` is set to `closest_pushable(@)`. 2. `revset-aliases.'closest_pushable(to)'` is set to `heads(::to & mutable() & ~description(exact:"") & (~empty() | merges()))`.
Tom Steyer is somehow running the most progressive campaign. Despite being a billionaire, he supports taxing the rich and supports the Billionaire Tax currently on California’s ballot, which CA DSA has endorsed. He has done an about-face from his previous position to now supporting state-level Medicare for All. He has called ICE a “violent extremist group” and outlined how he, as governor, would prosecute ICE agents. He has also been endorsed by a number of major labor unions, including the California Teachers Association, the California Federation of Teachers, AFSCME 3299, Unite HERE, the California Nurses Association, and the California Labor Federation, as well as progressive groups such as Our Revolution and Courage California, and former gubernatorial candidate Betty Yee. Notably, Steyer is one of two Democrats in the race, alongside Tony Thurmond, to explicitly support California’s law protecting trans girls’ participation in girls’ sports; Becerra, on the other hand, glibly stated that “there’s nothing in the Constitution that says that you are entitled to play a sport”, and other Democrats, including Porter, agreed. However, while Steyer has disavowed AIPAC’s influence on Democratic primaries (although, notably, AIPAC does not involve itself in non-federal races) and acknowledged that Israel is committing war crimes, he has refused to call Israel’s actions in Palestine a genocide and claimed that he “honestly does not know what genocide means.” We find this rhetorical trick to be pulled from the same Zionist playbook that motivates genocide denial across the political spectrum—playing dumb about the daily horrors and atrocities Israel commits is not “progressive” in any way. Additionally, Steyer is a billionaire. Even if he glibly considers himself a “class traitor”, his wealth was earned through the exploitation of the working class. Much of his wealth was also invested in private prisons and coal mining, accumulated by the same things he now decries.
Header: "Discussion and Conclusions" "From an attacker's point of view, package repositories represent a reliable and scalable malware distribution channel."
Linux Foundation 2024 financial disclosures, showing compensation for James "Jim" Zemlin, Executive Director. Jim received $952,166 in compensation, plus $451,018 listed as "Other."
Title: Terminology First, quickly, let's lay out some definitions: - AI: General term for a system based on a Large Language Model (LLM). - Large Language Model (LLM): Type of machine learning model trained on large amounts of text. - AI Model (or just "Model"): The particular LLM being employed. For example: Claude Opus 4.6 (by Anthropic), GPT-5.5 (by OpenAI), Llama 3.1 (by Meta, formerly Facebook). - Harness: Specialized software built around an AI model that turns it into a usable tool for some set of tasks. - AI Agent: An AI model plus a harness that enables the AI to perform tasks autonomously as directed by the user (writing code, gathering and summarizing materials, etc.). - Slop: LLM-generated material, created and used without care for quality. - LLM Sycophancy: The tendency of LLMs to behave sycophantically toward their users, effusively praising them or being dangerously uncritical.
The type signature for the FromResidual trait (part of the Try v2 design), with the type parameter R which is set to `<Self as Try>::Residual`, except `Try` isn't even a parent trait! Most Rust users probably don't even know you can do this!
"Budget and schedule constraints and the desire for efficiency are part of what motivates the creation of software in the first place. Once that software is in place, it’s frequently mission critical, having replaced the knowledge and labor of people who would have previously done the jobs the software now performs. Instead of accountants, a company may have accounting software, with a smaller number of accountants who know how to interact with the software and use it to perform their own jobs built on the knowledge the software provides with its data and embedded business logic. Rewrites to critical software systems are risky precisely because the software itself is so important. Rewriting a software system, whether as an in-place rewrite where components are swapped out piece by piece, or as a wholesale rewrite with a cutover date, risks the proper functioning of the business if the rewrite fails. Complex, long-running software systems can face other severe constraints as well. They may be unable to be turned off—because, having become so business-critical and time-sensitive, any attempt to bring them offline for maintenance or replacement is unacceptable. They may also have become lost artifacts, where the expertise of the individuals who created them or previously worked on them is lost because it wasn’t transferred to newer engineers, resulting in a current team that does not understand the system or feel comfortable making changes to it."
Higher-res version of the same image. Seth Milchick from Severance reading his performance review, with a page titled "The Principal Achievements and Failures of Seth Milchick."
From Severance, Seth Milchick reading his performance report, with the title "The Principal Achievements and Failures of Seth Milchick."
"Creedthoughts" from The Office, where Creed believes he's posting to his personal website, but he is in fact typing into a Microsoft Word document with the title "www.creedthoughts.gov.www/creedthoughts."
Mabosstiff, a dog PokĂŠmon.
CSS color that's a light yellowish-green with the hex code #daddad.
"[S]ure would be cool if github gave us proper conditionals... so here's a doubly-nested ternary-via-truthiness to try to provide the best possible functionality based on whether this is a pull_request, and whether it's from a fork. (PRs run on the *source* but secrets are usually on the *target* -- that's *good* but also really annoying to build CI around when it needs secrets to work right.)"
A light green color with the hex code "#badbad"
Apu from The Simpsons, while Homer opens thousands of candy bars in the store without buying them: Apu: I have asked you nicely not to mangle my merchandise. You leave me to choice but to ask you nicely again.
Comment from Jarred on Hacker News: I work on Bun and this is my branch This whole thread is an overreaction. 302 comments about code that does not work. We haven’t committed to rewriting. There’s a very high chance all this code gets thrown out completely. I’m curious to see what a working version of this looks, what it feels like, how it performs and if/how hard it’d be to get it to pass Bun’s test suite and be maintainable. I’d like to be able to compare a viable Rust version and a Zig version side by side.
The coach from Mean Girls (2004) saying "Don't have sex, because you will get pregnant and die."
Comment from the approver of the change on Hacker News. In it, he explains that the intent of the change had been to match customer expectations of disclosing when Copilot was used, but acknowledges they should have added checks to ensure it was not active if AI features were disabled or changes were not done with the help of AI. He offers his email as a direct outreach point if people have constructive feedback on the topic.
Comment on Hacker News from user “dmitriv” which reads: I am the person who approved this PR and would like to acknowledge and apologize for the mistake of turning this feature on by default without sufficient upfront validation. There was no ill intent by evil corporation, but rather a desire to support functionality that some customers expect of VS Code w.r.t. AI-generated code. As folks mentioned here - many similar tools do this as well. Obviously, it should not be on when disableAIFeatures is on and it should not be reporting changes that were not done by AI. I'll work on fixing those and meanwhile revert default to off in 1.119 update. I am open to any (constructive) comments/suggestions - please feel free to reach me directly (my alias @microsoft.com) or open an issue on GitHub. Happy to answer anything here as well.