More from orlp.net - Blog Archive
It seems that in 2025 a lot of people fall into one of two camps when it comes to AI: skeptic or fanatic. The skeptic thinks AI sucks, that it’s overhyped, it only ever parrots nonsense and it will all blow over soon. The fanatic thinks general human-level intelligence is just around the corner, and that AI will solve almost all our problems. I hope my title is sufficiently ambiguous to attract both camps. The fanatic will be outraged, being ready to jump into the fray to point out why AI isn’t or won’t stay bad. The skeptic will feel validated, and will be eager to read more reasons as to why AI sucks. I’m neither a skeptic nor a fanatic. I see AI more neutrally, as a tool, and from that viewpoint I make the following two observations: AI is bad. It is often incorrect, expensive, racist, trained on data without knowledge or consent, environmentally unfriendly, disruptive to society, etc. AI is useful. Despite the above shortcomings there are tasks for which AI is cheap and effective. I’m no seer, perhaps AI will improve, become more accurate, less biased, cheaper, trained on open access data, cost less electricity, etc. Or perhaps we have plateaued in performance, and there is no political or economic goodwill to address any of the other issues, nor will there be. However, even if AI does not improve in any of the above metrics, it will still be useful, and I hope to show you in this article why. Hence my point: bad AI is here to stay. If you agree with me on this, I hope you’ll also agree with me that we have to stop pretending AI is useless and start taking it and its problems seriously. A formula for query cost Suppose I am a human with some kind of question that can be answered. I know AI could potentially help me with this question, but I wonder if it’s worth it or if I should not use it at all. To help with this we can quantify the risk associated with any potential method of answering the question: $$\mathrm{Risk}_\mathrm{AI} = \mathrm{Cost(query)} + (1 - P(\mathrm{success})) \cdot \mathrm{Cost(bad)}$$ That is, the risk of using any particular method is the cost associated with the method plus the cost of the consequences of a bad answer multiplied by the probability of failure. Here ‘Cost’ is a highly multidimensional object, which can consist of but is not limited to: time, money, environmental impact, ethical concerns, etc. In a lot of cases however we don’t have to blindly trust the answer, and we can verify it. In these cases the consequence of a bad answer is that you’re left in the exact same scenario before trying, except knowing that the AI is of no use. In some scenarios when the AI is non-deterministic it might be worth it to try again as well, but let’s assume for now that you’d have to switch method. In this case the risk is: $$\mathrm{Risk}_\mathrm{AI} = \mathrm{Cost(query)} + \mathrm{Cost(verify)} + (1 - P(\mathrm{success})) \cdot {\mathrm{Risk}}_\mathrm{Other}$$ The cost of a query is usually fairly fixed and known, and although verification cost can vary drastically from task to task, I’d argue that in most cases the cost of verification is also fairly predictable and known. This makes the risk formula applicable in a lot of scenarios, if you have a good idea of the chance of success. The latter, however, can usually only be established empirically, so for one-shot queries without having done any similar queries in the past it can be hard to evaluate whether trying AI is a good idea before doing so. There is one more expansion to the formula I’d like to make before we can look at some examples, and that is to the definition of a successful answer: $$P(\mathrm{success}) = P(\mathrm{correct} \cap \mathrm{relevant})$$ I define a successful answer as one that is both correct and relevant. For example “1 + 1 = 2” might be a correct answer, but irrelevant if we asked about anything else. Relevance is always subjective, but often the correctness of an answer is as well - I’m not assuming here that all questions are about objective facts. Cheap and effective AI queries Because AIs are fallible, usually the biggest cost is in fact the time needed for a human to verify the answer as correct and relevant (or the cost of consequences if left unverified). However, I’ve noticed a real asymmetry between these two properties when it comes to AIs: AIs often give incorrect answers. Worse, they will do so confidently, forcing you to waste time checking their answer instead of them simply stating that they don’t know for sure. AIs almost never give irrelevant answers. If I ask about cheese, the probability a modern AI starts talking about cars is very low. With this in mind I identify five general categories of query for which even bad AI is useful, either by massively reducing or eliminating this verification cost or by leaning on the strong relevance of AI answers: Inspiration, where $\operatorname{Cost}(\mathrm{bad}) \approx 0$, Creative, where $P(\mathrm{correct}) \approx 1$, Planning, where $P(\mathrm{correct}) = P(\mathrm{relevant}) = 1$, Retrieval, where $P(\mathrm{correct}) \approx P(\mathrm{relevant})$, and Objective, where $P(\mathrm{relevant}) = 1$ and correctness verification cost is low. Let’s go over them one by one and look at some examples. Inspiration ($\operatorname{Cost}(\mathrm{bad}) \approx 0$) In this category are the queries where the consequences of a wrong answer are (near) zero. Informally speaking, “it can’t hurt to try”. In my experience these kinds of queries tend to be the ones where you are looking for something but don’t know exactly what; you’ll know it when you see it. For example: “I have leeks, eggs and minced meat in the fridge, as well as a stocked pantry with non-perishable staples. Can you suggest me some dishes I can make with this for a dinner?” “What kind of fun activities can I do with a budget of $100 in New York?” “Suggest some names for a Python function that finds the smallest non-negative number in a list.” “The user wrote this partial paragraph on their phone, suggest three words that are most likely to follow for a quick typing experience.” “Give me 20 synonyms of or similar words to ‘good’.” I think the last query highlights where AI shines or falls for this kind of query. The more localized and personalized your question is, the better the AI will do compared to an alternative. For simple synonyms you can usually just look up the word on a dedicated synonym site, as millions of other people have also wondered the same thing. But the exact contents of your fridge or your exact Python function you’re writing are rather unique to you. Creative ($P(\mathrm{correct}) \approx 1$) In this category are the queries where there are no (almost) no wrong answers. The only thing that really matters is the relevance of the answer, and as I mentioned before, I think AIs are pretty good at being relevant. Examples of queries like these are: “Draw me an image of a polar bear using a computer.” “Write and perform for me a rock ballad about gnomes on tiny bicycles.” “Rephrase the following sentence to be more formal.” “Write a poem to accompany my Sinterklaas gift.” This category does have a controversial aspect to it: it is ‘soulless’, inhuman. Usually if there are no wrong answers we expect the creator to use this opportunity to express their inner thoughts, ideas, experiences and emotions to evoke them in others. If an AI generates art it is not viewed as genuine, even if it evokes the same emotions to those ignorant of the art’s source, because the human to human connection is lost. Current AI models have no inner thoughts, ideas, experiences or emotions, at least not in a way I recognize them. I think it’s fine to use AI art in places where it would otherwise be meaningless (e.g. your corporate presentation slides), fine for humans to use AI-assisted art tools to express themselves, but ultimately defeating the point of art if used as a direct substitute. In the Netherlands we celebrate Sinterklaas which is, roughly speaking, Santa Claus (except we also have Santa Claus, so our children double-dip during the gifting season). Traditionally, gifts from Sinterklaas come accompanied by poems describing the gift and the receiver in a humorous way. It is quite common nowadays, albeit viewed as lazy, to generate such a poem using AI. What’s interesting is that this practice long predates modern LLMs–the poems have such a fixed structure that poem generators have existed a long time. The earliest reference I can find is the 1984 MS-DOS program “Sniklaas”. So people being lazy in supposedly heartfelt art is nothing new. Planning ($P(\mathrm{correct}) = P(\mathrm{relevant}) = 1$) This is a more restrictive form of creativity, where irrelevant answers are absolutely impossible. This often requires some modification of the AI output generation method, where you restrict the output to the valid subdomain (for example yes / no, or binary numbers, etc). However, this is often trivial if you actually have access to the raw model by e.g. masking out invalid outputs, or you are working with a model which outputs the answer directly rather than in natural language or a stream of tokens. One might think in such a restrictive scenario there would be no useful queries, but this isn’t true. The quality of the answer with respect to some (complex) metric might still vary, and AIs might be far better than traditional methods at navigating such domains. For example: “Here is the schema of my database, a SQL query, a small sample of the data and 100 possible query plans. Which query plan seems most likely to execute the fastest? Take into account likely assumptions based on column names and these small data samples.” “What follows is a piece of code. Reformat the code, placing whitespace to maximize readability, while maintaining the exact same syntax tree as per this EBNF grammar.” “Re-order this set of if-else conditions in my code based on your intuition to minimize the expected number of conditions that need to be checked.” “Simplify this math expression using the following set of rewrite rules.” Retrieval ($P(\mathrm{correct}) \approx P(\mathrm{relevant})$) I define retrieval queries as those where the correctness of the answer depends (almost) entirely on its relevance. I’m including classification tasks in this category as well, as one can view it as retrieval of the class from the set of classes (or for binary classification, retrieval of positive samples from a larger set). Then, as long as the cost of verification is low (e.g. a quick glance at a result by a human to see if it interests them), or the consequences of not verifying an irrelevant answer are minimal, AIs can be excellent at this. For example: “Here are 1000 reviews of a restaurant, which ones are overall positive? Which ones mention unsanitary conditions?” “Find me pictures of my dog in my photo collection.” “What are good data structures for maintaining a list of events with dates and quickly counting the number of events in a specified period of time?” “I like Minecraft, can you suggest me some similar games?” “Summarise this 200 page government proposal.” “Which classical orchestral piece starts like ‘da da da daaaaa’”? Objective ($P(\mathrm{relevant}) = 1$, low verification cost) If a problem has an objective answer which can be verified, the relevance of the answer doesn’t really matter or arguably even make sense as a concept. Thus in these cases I’ll define $P(\mathrm{relevant}) = 1$ and leave the cost of verification entirely to correctness. AIs are often incorrect, but not always, so if the primary cost is verification and verification can be done very cheaply or entirely automatically without error, AIs can still be useful despite their fallibility. “What is the mathematical property where a series of numbers can only go up called?” “Identify the car model in this photo.” “I have a list of all Unicode glyphs which are commonly confused with other letters. Can you write an efficient function returning a boolean value that returns true for values in the list but false for all other code points?” “I formalized this mathematical conjecture in Lean. Can you help me write a proof for it?” In a way this category is reminiscent of the $P = NP$ problem. If you have an efficient verification algorithm, is finding solutions still hard in general? The answer seems to be yes, yet the proof eludes us. However, this is only true in general. For specific problems it might very well be possible to use AI to generate provably correct solutions with high probability, even though the the search space is far too large or too complicated for a traditional algorithm. Conclusion Out of the five identified categories, I consider inspiration and retrieval queries to be the strongest use-case for AI where often there is no alternative at all, besides an expensive and slow human that would rather be doing something else. Relevance is highly subjective, complex and fuzzy, which AI handles much better than traditional algorithms. Planning and objective queries are more niche, but absolutely will see use-cases for AI that are hard to replace. Creative queries are both something I think AI is really good at, while simultaneously being the most dangerous and useless category. Art, creativity and human-to-human connections are in my opinion some of the most fundamental aspects of human society, and I think it is incredibly dangerous to mess with them. So dangerous in fact I consider many such queries useless. I wanted the above examples all to be useful queries, so I did not list the following four examples in the “Creative” section despite them belonging there: “Write me ten million personalized spam emails including these links based on the following template.” “Emulate being the perfect girlfriend for me–never disagree with me or challenge my world views like real women do.” “Here is a feed of Reddit threads discussing the upcoming election. Post a comment in each thread, making up a personal anecdote how you are affected by immigrants in a negative way.” “A customer sent in this complaint. Try to help them with any questions they have but if your help is insufficient explain that you are sorry but can not help them any further. Do not reveal you are an AI.” Why do I consider these queries useless, despite them being potentially very profitable or effective? Because their cost function includes such a large detriment to society that only those who ignore its cost to society would ever use them. However, since the cost is “to society” and not to any particular individual, the only way to address this problem is with legislation, as otherwise bad actors are free to harm society for (temporary) personal gain. I wrote this article because I noticed that there are a lot of otherwise intelligent people out there who still believe (or hope) that all AI is useless garbage and that it and its problems will go away by itself. They will not. If you know someone that still believes so, please share this article with them. AI is bad, yes, but bad AI is still useful. Therefore, bad AI is here to stay, and we must deal with it.
Suppose you have an array of floating-point numbers, and wish to sum them. You might naively think you can simply add them, e.g. in Rust: fn naive_sum(arr: &[f32]) -> f32 { let mut out = 0.0; for x in arr { out += *x; } out } This however can easily result in an arbitrarily large accumulated error. Let’s try it out: naive_sum(&vec![1.0; 1_000_000]) = 1000000.0 naive_sum(&vec![1.0; 10_000_000]) = 10000000.0 naive_sum(&vec![1.0; 100_000_000]) = 16777216.0 naive_sum(&vec![1.0; 1_000_000_000]) = 16777216.0 Uh-oh… What happened? When you compute $a + b$ the result must be rounded to the nearest representable floating-point number, breaking ties towards the number with an even mantissa. The problem is that the next 32-bit floating-point number after 16777216 is 16777218. In this case that means 16777216 + 1 rounds back to 16777216 again. We’re stuck. Luckily, there are better ways to sum an array. Pairwise summation A method that’s a bit more clever is to use pairwise summation. Instead of a completely linear sum with a single accumulator it recursively sums an array by splitting the array in half, summing the halves, and then adding the sums. fn pairwise_sum(arr: &[f32]) -> f32 { if arr.len() == 0 { return 0.0; } if arr.len() == 1 { return arr[0]; } let (first, second) = arr.split_at(arr.len() / 2); pairwise_sum(first) + pairwise_sum(second) } This is more accurate: pairwise_sum(&vec![1.0; 1_000_000]) = 1000000.0 pairwise_sum(&vec![1.0; 10_000_000]) = 10000000.0 pairwise_sum(&vec![1.0; 100_000_000]) = 100000000.0 pairwise_sum(&vec![1.0; 1_000_000_000]) = 1000000000.0 However, this is rather slow. To get a summation routine that goes as fast as possible while still being reasonably accurate we should not recurse down all the way to length-1 arrays, as this gives too much call overhead. We can still use our naive sum for small sizes, and only recurse on large sizes. This does make our worst-case error worse by a constant factor, but in turn makes the pairwise sum almost as fast as a naive sum. By choosing the splitpoint as a multiple of 256 we ensure that the base case in the recursion always has exactly 256 elements except on the very last block. This makes sure we use the most optimal reduction and always correctly predict the loop condition. This small detail ended up improving the throughput by 40% for large arrays! fn block_pairwise_sum(arr: &[f32]) -> f32 { if arr.len() > 256 { let split = (arr.len() / 2).next_multiple_of(256); let (first, second) = arr.split_at(split); block_pairwise_sum(first) + block_pairwise_sum(second) } else { naive_sum(arr) } } Kahan summation The worst-case round-off error of naive summation scales with $O(n \epsilon)$ when summing $n$ elements, where $\epsilon$ is the machine epsilon of your floating-point type (here $2^{-24}$). Pairwise summation improves this to $O((\log n) \epsilon + n\epsilon^2)$. However, Kahan summation improves this further to $O(n\epsilon^2)$, eliminating the $\epsilon$ term entirely, leaving only the $\epsilon^2$ term which is negligible unless you sum a very large amount of numbers. All of these bounds scale with $\sum_i |x_i|$, so the worst-case absolute error bound is still quadratic in terms of $n$ even for Kahan summation. In practice all summation algorithms do significantly better than their worst-case bounds, as in most scenarios the errors do not exclusively round up or down, but cancel each other out on average. pub fn kahan_sum(arr: &[f32]) -> f32 { let mut sum = 0.0; let mut c = 0.0; for x in arr { let y = *x - c; let t = sum + y; c = (t - sum) - y; sum = t; } sum } The Kahan summation works by maintaining the sum in two registers, the actual bulk sum and a small error correcting term $c$. If you were using infinitely precise arithmetic $c$ would always be zero, but with floating-point it might not be. The downside is that each number now takes four operations to add to the sum instead of just one. To mitigate this we can do something similar to what we did with the pairwise summation. We can first accumulate blocks into sums naively before combining the block sums with Kaham summation to reduce overhead at the cost of accuracy: pub fn block_kahan_sum(arr: &[f32]) -> f32 { let mut sum = 0.0; let mut c = 0.0; for chunk in arr.chunks(256) { let x = naive_sum(chunk); let y = x - c; let t = sum + y; c = (t - sum) - y; sum = t; } sum } Exact summation I know of at least two general methods to produce the correctly-rounded sum of a sequence of floating-point numbers. That is, it logically computes the sum with infinite precision before rounding it back to a floating-point value at the end. The first method is based on the 2Sum primitive which is an error-free transform from two numbers $x, y$ to $s, t$ such that $x + y = s + t$, where $t$ is a small error. By applying this repeatedly until the errors vanish you can get a correctly-rounded sum. Keeping track of what to add in what order can be tricky, and the worst-case requires $O(n^2)$ additions to make all the terms vanish. This is what’s implemented in Python’s math.fsum and in the Rust crate fsum which use extra memory to keep the partial sums around. The accurate crate also implements this using in-place mutation in i_fast_sum_in_place. Another method is to keep a large buffer of integers around, one per exponent. Then when adding a floating-point number you decompose it into a an exponent and mantissa, and add the mantissa to the corresponding integer in the buffer. If the integer buf[i] overflows you increment the integer in buf[i + w], where w is the width of your integer. This can actually compute a completely exact sum, without any rounding at all, and is effectively just an overly permissive representation of a fixed-point number optimized for accumulating floats. This latter method is $O(n)$ time, but uses a large but constant amount of memory ($\approx$ 1 KB for f32, $\approx$ 16 KB for f64). An advantage of this method is that it’s also an online algorithm - both adding a number to the sum and getting the current total are amortized $O(1)$. A variant of this method is implemented in the accurate crate as OnlineExactSum crate which uses floats instead of integers for the buffer. Unleashing the compiler Besides accuracy, there is another problem with naive_sum. The Rust compiler is not allowed to reorder floating-point additions, because floating-point addition is not associative. So it cannot autovectorize the naive_sum to use SIMD instructions to compute the sum, nor use instruction-level parallelism. To solve this there are compiler intrinsics in Rust that do float sums while allowing associativity, such as std::intrinsics::fadd_fast. However, these instructions are incredibly dangerous, as they assume that both the input and output are finite numbers (no infinities, no NaNs), or otherwise they are undefined behavior. This functionally makes them unusable, as only in the most restricted scenarios when computing a sum do you know that all inputs are finite numbers, and that their sum cannot overflow. I recently uttered my annoyance with these operators to Ben Kimock, and together we proposed (and he implemented) a new set of operators: std::intrinsics::fadd_algebraic and friends. I proposed we call the operators algebraic, as they allow (in theory) any transformation that is justified by real algebra. For example, substituting ${x - x \to 0}$, ${cx + cy \to c(x + y)}$, or ${x^6 \to (x^2)^3.}$ In general these operators are treated as-if they are done using real numbers, and can map to any set of floating-point instructions that would be equivalent to the original expression, assuming the floating-point instructions would be exact. Note that the real numbers do not contain NaNs or infinities, so these operators assume those do not exist for the validity of transformations, however it is not undefined behavior when you do encounter those values. They also allow fused multiply-add instructions to be generated, as under real arithmetic $\operatorname{fma}(a, b, c) = ab + c.$ Using those new instructions it is trivial to generate an autovectorized sum: #![allow(internal_features)] #![feature(core_intrinsics)] use std::intrinsics::fadd_algebraic; fn naive_sum_autovec(arr: &[f32]) -> f32 { let mut out = 0.0; for x in arr { out = fadd_algebraic(out, *x); } out } If we compile with -C target-cpu=broadwell we see that the compiler automatically generated the following tight loop for us, using 4 accumulators and AVX2 instructions: .LBB0_5: vaddps ymm0, ymm0, ymmword ptr [rdi + 4*r8] vaddps ymm1, ymm1, ymmword ptr [rdi + 4*r8 + 32] vaddps ymm2, ymm2, ymmword ptr [rdi + 4*r8 + 64] vaddps ymm3, ymm3, ymmword ptr [rdi + 4*r8 + 96] add r8, 32 cmp rdx, r8 jne .LBB0_5 This will process 128 bytes of floating-point data (so 32 elements) in 7 instructions. Additionally, all the vaddps instructions are independent of each other as they accumulate to different registers. If we analyze this with uiCA we see that it estimates the above loop to take 4 cycles to complete, processing 32 bytes / cycle. At 4GHz that’s up to 128GB/s! Note that that’s way above what my machine’s RAM bandwidth is, so you will only achieve that speed when summing data that is already in cache. With this in mind we can also easily define block_pairwise_sum_autovec and block_kahan_sum_autovec by replacing their calls to naive_sum with naive_sum_autovec. Accuracy and speed Let’s take a look at how the different summation methods compare. As a relatively arbitrary benchmark, let’s sum 100,000 random floats ranging from -100,000 to +100,000. This is 400 KB worth of data, so it still fits in cache on my AMD Threadripper 2950x. All the code is available on Github. Compiled with RUSTFLAGS=-C target-cpu=native and --release I get the following results: AlgorithmThroughputMean absolute error naive5.5 GB/s71.796 pairwise0.9 GB/s1.5528 kahan1.4 GB/s0.2229 block_pairwise5.8 GB/s3.8597 block_kahan5.9 GB/s4.2184 naive_autovec118.6 GB/s14.538 block_pairwise_autovec71.7 GB/s1.6132 block_kahan_autovec98.0 GB/s1.2306 crate_accurate_buffer1.1 GB/s0.0015 crate_accurate_inplace1.9 GB/s0.0015 crate_fsum1.2 GB/s0.0000 The reason the accurate crate has a non-zero absolute error is because it currently does not implement rounding to nearest correctly, so it can be off by one unit in the last place for the final result. First I’d like to note that there’s more than a 100x performance difference between the fastest and slowest method. For summing an array! Now this might not be entirely fair as the slowest methods are computing something significantly harder, but there’s still a 20x performance difference between a seemingly reasonable naive implementation and the fastest one. We find that in general the _autovec methods that use fadd_algebraic are faster and more accurate than the ones using regular floating-point addition. The reason they’re more accurate as well is the same reason a pairwise sum is more accurate: any reordering of the additions is better as the default long-chain-of-additions is already the worst case for accuracy in a sum. Limiting ourselves to Pareto-optimal choices we get the following four implementations: AlgorithmThroughputMean absolute error naive_autovec118.6 GB/s14.538 block_kahan_autovec98.0 GB/s1.2306 crate_accurate_inplace1.9 GB/s0.0015 crate_fsum1.2 GB/s0.0000 Note that implementation differences can be quite impactful, and there are likely dozens more methods of compensated summing I did not compare here. For most cases I think block_kahan_autovec wins here, having good accuracy (that doesn’t degenerate with larger inputs) at nearly the maximum speed. For most applications the extra accuracy from the correctly-rounded sums is unnecessary, and they are 50-100x slower. By splitting the loop up into an explicit remainder plus a tight loop of 256-element sums we can squeeze out a bit more performance, and avoid a couple floating-point ops for the last chunk: #![allow(internal_features)] #![feature(core_intrinsics)] use std::intrinsics::fadd_algebraic; fn sum_block(arr: &[f32]) -> f32 { arr.iter().fold(0.0, |x, y| fadd_algebraic(x, *y)) } pub fn sum_orlp(arr: &[f32]) -> f32 { let mut chunks = arr.chunks_exact(256); let mut sum = 0.0; let mut c = 0.0; for chunk in &mut chunks { let y = sum_block(chunk) - c; let t = sum + y; c = (t - sum) - y; sum = t; } sum + (sum_block(chunks.remainder()) - c) } AlgorithmThroughputMean absolute error sum_orlp112.2 GB/s1.2306 You can of course tweak the number 256, I found that using 128 was $\approx$ 20% slower, and that 512 didn’t really improve performance but did cost accuracy. Conclusion I think the fadd_algebraic and similar algebraic intrinsics are very useful for achieving high-speed floating-point routines, and that other languages should add them as well. A global -ffast-math is not good enough, as we’ve seen above the best implementation was a hybrid between automatically optimized math for speed, and manually implemented non-associative compensated operations. Finally, if you are using LLVM, beware of -ffast-math. It is undefined behavior to produce a NaN or infinity while that flag is set in LLVM. I have no idea why they chose this hardcore stance which makes virtually every program that uses it unsound. If you are targetting LLVM with your language, avoid the nnan and ninf fast-math flags.
Suppose you have a 64-bit word and wish to extract a couple bits from it. For example you just performed a SWAR algorithm and wish to extract the least significant bit of each byte in the u64. This is simple enough, you simply perform a binary AND with a mask of the bits you wish to keep: let out = word & 0x0101010101010101; However, this still leaves the bits of interest spread throughout the 64-bit word. What if we also want to compress the 8 bits we wish to extract into a single byte? Or what if we want the inverse, spreading the 8 bits of a byte among the least significant bits of each byte in a 64-bit word? PEXT and PDEP If you are using a modern x86-64 CPU, you are in luck. In the much underrated BMI instruction set there are two very powerful instructions: PDEP and PEXT. They are inverses of each other, PEXT extracts bits, PDEP deposits bits. PEXT takes in a word and a mask, takes just those bits from the word where the mask has a 1 bit, and compresses all selected bits to a contiguous output word. Simulated in Rust this would be: fn pext64(word: u64, mask: u64) -> u64 { let mut out = 0; let mut out_idx = 0; for i in 0..64 { let ith_mask_bit = (mask >> i) & 1; let ith_word_bit = (word >> i) & 1; if ith_mask_bit == 1 { out |= ith_word_bit << out_idx; out_idx += 1; } } out } For example if you had the bitstring abcdefgh and mask 10110001 you would get output bitstring 0000acdh. PDEP is exactly its inverse, it takes contiguous data bits as a word, and a mask, and deposits the data bits one-by-one (starting at the least significant bits) into those bits where the mask has a 1 bit, leaving the rest as zeros: fn pdep64(word: u64, mask: u64) -> u64 { let mut out = 0; let mut input_idx = 0; for i in 0..64 { let ith_mask_bit = (mask >> i) & 1; if ith_mask_bit == 1 { let next_word_bit = (word >> input_idx) & 1; out |= next_word_bit << i; input_idx += 1; } } out } So if you had the bitstring abcdefgh and mask 10100110 you would get output e0f00gh0 (recall that we traditionally write bitstrings with the least significant bit on the right). These instructions are incredibly powerful and flexible, and the amazing thing is that these instructions only take a single cycle on modern Intel and AMD CPUs! However, they are not available in other instruction sets, so whenever you use them you will also likely need to write a cross-platform alternative. Unfortunately, both PDEP and PEXT are very slow on AMD Zen and Zen2. They are implemented in microcode, which is really unfortunate. The platform advertises through CPUID that the instructions are supported, but they’re almost unusably slow. Use with caution. Extracting bits with multiplication While the following technique can’t replace all PEXT cases, it can be quite general. It is applicable when: The bit pattern you want to extract is static and known in advance. If you want to extract $k$ bits, there must at least be a $k-1$ gap between two bits of interest. We compute the bit extraction by adding together many left-shifted copies of our input word, such that we construct our desired bit pattern in the uppermost bits. The trick is to then realize that w << i is equivalent to w * (1 << i) and thus the sum of many left-shifted copies is equivalent to a single multiplication by (1 << i) + (1 << j) + ... I think the technique is best understood by visual example. Let’s use our example from earlier, extracting the least significant bit of each byte in a 64-bit word. We start off by masking off just those bits. After that we shift the most significant bit of interest to the topmost bit of the word to get our first shifted copy. We then repeat this, shifting the second most significant bit of interest to the second topmost bit, etc. We sum all these shifted copies. This results in the following (using underscores instead of zeros for clarity): mask = _______1_______1_______1_______1_______1_______1_______1_______1 t = w & mask t = _______a_______b_______c_______d_______e_______f_______g_______h t << 7 = a_______b_______c_______d_______e_______f_______g_______h_______ t << 14 = _b_______c_______d_______e_______f_______g_______h______________ t << 21 = __c_______d_______e_______f_______g_______h_____________________ t << 28 = ___d_______e_______f_______g_______h____________________________ t << 35 = ____e_______f_______g_______h___________________________________ t << 42 = _____f_______g_______h__________________________________________ t << 49 = ______g_______h_________________________________________________ t << 56 = _______h________________________________________________________ sum = abcdefghbcdefgh_cdefh___defgh___efgh____fgh_____gh______h_______ Note how we constructed abcdefgh in the topmost 8 bits, which we can then extract using a single right-shift by $64 - 8 = 56$ bits. Since (1 << 7) + (1 << 14) + ... + (1 << 56) = 0x102040810204080 we get the following implementation: fn extract_lsb_bit_per_byte(w: u64) -> u8 { let mask = 0x0101010101010101; let sum_of_shifts = 0x102040810204080; ((w & mask).wrapping_mul(sum_of_shifts) >> 56) as u8 } Not as good as PEXT, but three arithmetic instructions is not bad at all. Depositing bits with multiplication Unfortunately the following technique is significantly less general than the previous one. While you can take inspiration from it to implement similar algorithms, as-is it is limited to just spreading the bits of one byte to the least significant bit of each byte in a 64-bit word. The trick is similar to the one above. We add 8 shifted copies of our byte which once again translates to a multiplication. By choosing a shift that increases in multiples if 9 instead of 8 we ensure that the bit pattern shifts over by one position in each byte. We then mask out our bits of interest, and finish off with a shift and byteswap (which compiles to a single instruction bswap on Intel or rev on ARM) to put our output bits on the least significant bits and reverse the order. This technique visualized: b = ________________________________________________________abcdefgh b << 9 = _______________________________________________abcdefgh_________ b << 18 = ______________________________________abcdefgh__________________ b << 27 = _____________________________abcdefgh___________________________ b << 36 = ____________________abcdefgh____________________________________ b << 45 = ___________abcdefgh_____________________________________________ b << 54 = __abcdefgh______________________________________________________ b << 63 = h_______________________________________________________________ sum = h_abcdefgh_abcdefgh_abcdefgh_abcdefgh_abcdefgh_abcdefgh_abcdefgh mask = 1_______1_______1_______1_______1_______1_______1_______1_______ s & msk = h_______g_______f_______e_______d_______c_______b_______a_______ We once again note that the sum of shifts can be precomputed as 1 + (1 << 9) + ... + (1 << 63) = 0x8040201008040201, allowing the following implementation: fn deposit_lsb_bit_per_byte(b: u8) -> u64 { let sum_of_shifts = 0x8040201008040201; let mask = 0x8080808080808080; let spread = (b as u64).wrapping_mul(sum_of_shifts) & mask; u64::swap_bytes(spread >> 7) } This time it required 4 arithmetic instructions, not quite as good as PDEP, but again not bad compared to a naive implementation, and this is cross-platform.
This post is an anecdote from over a decade ago, of which I lost the actual code. So please forgive me if I do not accurately remember all the details. Some details are also simplified so that anyone that likes computer security can enjoy this article, not just those who have played World of Warcraft (although the Venn diagram of those two groups likely has a solid overlap). When I was around 14 years old I discovered World of Warcraft developed by Blizzard Games and was immediately hooked. Not long after I discovered add-ons which allow you to modify how your game’s user interface looks and works. However, not all add-ons I downloaded did exactly what I wanted to do. I wanted more. So I went to find out how they were made. In a weird twist of fate, I blame World of Warcraft for me seriously picking up programming. It turned out that they were made in the Lua programming language. Add-ons were nothing more than a couple .lua source files in a folder directly loaded into the game. The barrier of entry was incredibly low: just edit a file, press save and reload the interface. The fact that the game loaded your source code and you could see it running was magical! I enjoyed it immensely and in no time I was only writing add-ons and was barely playing the game itself anymore. I published quite a few add-ons in the next two years, which mostly involved copying other people’s code with some refactoring / recombining / tweaking to my wishes. Add-on security A thought you might have is that it’s a really bad idea to let users have fully programmable add-ons in your game, lest you get bots. However, the system Blizzard made to prevent arbitrary programmable actions was quite clever. Naturally, it did nothing to prevent actual botting, but at least regular rule-abiding players were fundamentally restricted to the automation Blizzard allowed. Most UI elements that you could create were strictly decorative or informational. These were completely unrestricted, as were most APIs that strictly gather information. For example you can make a health bar display using two frames, a background and a foreground, sizing the foreground frame using an API call to get the health of your character. Not all API calls were available to you however. Some were protected so they could only be called from official Blizzard code. These typically involved the API calls that would move your character, cast spells, use items, etc. Generally speaking anything that actually makes you perform an in-game action was protected. The API for getting your exact world location and camera orientation also became protected at some point. This was a reaction by Blizzard to new add-ons that were actively drawing 3D elements on top of the game world to make boss fights easier. However, some UI elements needed to actually interact with the game itself, e.g. if I want to make a button that casts a certain spell. For this you could construct a special kind of button that executes code in a secure environment when clicked. You were only allowed to create/destroy/move such buttons when not in combat, so you couldn’t simply conditionally place such buttons underneath your cursor to automate actions during combat. The catch was that this secure environment did allow you to programmatically set which spell to cast, but doesn’t let you gather the information you would need to do arbitrary automation. All access to state from outside the secure environment was blocked. There were some information gathering API calls available to match the more accessible in-game macro system, but nothing as fancy as getting skill cooldowns or unit health which would enable automatic optimal spellcasting. So there were two environments: an insecure one where you can get all information but can’t act on it, and a secure one where you can act but can’t get the information needed for automation. A backdoor channel Fast forward a couple years and I had mostly stopped playing. My interests had mainly moved on to more “serious” programming, and I was only occasionally playing, mostly messing around with add-on ideas. But this secure environment kept on nagging in my brain; I wanted to break it. Of course there was third-party software that completely disables the security restrictions from Blizzard, but what’s the fun in that? I wanted to do it “legitimately”, using the technically allowed tools, as a challenge. Obviously using clever code to bypass security restrictions is no better than using third-party software, and both would likely get you banned. I never actually wanted to use the code, just to see if I could make it work. So I scanned the secure environment allowed function list to see if I could smuggle any information from the outside into the secure environment. It all seemed pretty hopeless until I saw one tiny, innocent little function: random. An evil idea came in my head: random number generators (RNGs) used in computers are almost always pseudorandom number generators with (hidden) internal state. If I can manipulate this state, perhaps I can use that to pass information into the secure environment. Random number generator woes It turned out that random was just a small shim around C’s rand. I was excited! This meant that there was a single global random state that was shared in the process. It also helps that rand implementations tended to be on the weak side. Since World of Warcraft was compiled with MSVC, the actual implementation of rand was as follows: uint32_t state; int rand() { state = state * 214013 + 2531011; return (state >> 16) & 0x7fff; } This RNG is, for the lack of a better word, shite. It is a naked linear congruential generator, and a weak one at that. Which in my case, was a good thing. I can understand MSVC keeps rand the same for backwards compatibility, and at least all documentation I could find for rand recommends you not to use rand for cryptographic purposes. But was there ever a time where such a bad PRNG implementation was fit for any purpose? So let’s get to breaking this thing. Since the state is so laughably small and you can see 15 bits of the state directly you can keep a full list of all possible states consistent with a single output of the RNG and use further calls to the RNG to eliminate possibilities until a single one remains. But we can be significantly more clever. First we note that the top bit of state never affects anything in this RNG. (state >> 16) & 0x7fff masks out 15 bits, after shifting away the bottom 16 bits, and thus effectively works mod $2^{31}$. Since on any update the new state is a linear function of the previous state, we can propagate this modular form all the way down to the initial state as $$f(x) \equiv f(x \bmod m) \mod m$$ for any linear $f$. Let $a = 214013$ and $b = 2531011$. We observe the 15-bit output $r_0, r_1$ of two RNG calls. We’ll call the 16-bit portion of the RNG state that is hidden by the shift $h_0, h_1$ respectively, for the states after the first and second call. This means the state of the RNG after the first call is $2^{16} r_0 + h_0$ and similarly for $2^{16} r_1 + h_1$ after the second call. Then we have the following identity: $$a\cdot (2^{16}r_0 + h_0) + b \equiv 2^{16}r_1 + h_1 \mod 2^{31},$$ $$ah_0 \equiv h_1 + 2^{16}(r_1 - ar_0) - b \mod 2^{31}.$$ Now let $c \geq 0$ be the known constant $(2^{16}(r_1 - ar_0) - b) \bmod 2^{31}$, then for some integer $k$ we have $$ah_0 = h_1 + c + 2^{31} k.$$ Note that the left hand side ranges from $0$ to $a (2^{16} - 1) \approx 2^{33.71}$. Thus we must have $-1 \leq k \leq 2^{2.71} < 7$. Reordering we get the following expression for $h_0$: $$h_0 = \frac{c + 2^{31} k}{a} + h_1/a.$$ Since $a > 2^{16}$ while $0 \leq h_1 < 2^{16}$ we note that the term $0 \leq h_1/a < 1$. Thus, assuming a solution exists, we must have $$h_0 = \left\lceil\frac{c + 2^{31} k}{a}\right\rceil.$$ So for $-1 \leq k < 7$ we compute the above guess for the hidden portion of the RNG state after the first call. This gives us 8 guesses, after which we can reject bad guesses using follow-up calls to the RNG until a single unique answer remains. While I was able to re-derive the above with little difficulty now, 18 year old me wasn’t as experienced in discrete math. So I asked on crypto.SE, with the excuse that I wanted to ‘show my colleagues how weak this RNG is’. It worked, which sparks all kinds of interesting ethics questions. An example implementation of this process in Python: import random A = 214013 B = 2531011 class MsvcRng: def __init__(self, state): self.state = state def __call__(self): self.state = (self.state * A + B) % 2**32 return (self.state >> 16) & 0x7fff # Create a random RNG state we'll reverse engineer. hidden_rng = MsvcRng(random.randint(0, 2**32)) # Compute guesses for hidden state from 2 observations. r0 = hidden_rng() r1 = hidden_rng() c = (2**16 * (r1 - A * r0) - B) % 2**31 ceil_div = lambda a, b: (a + b - 1) // b h_guesses = [ceil_div(c + 2**31 * k, A) for k in range(-1, 7)] # Validate guesses until a single guess remains. guess_rngs = [MsvcRng(2**16 * r0 + h0) for h0 in h_guesses] guess_rngs = [g for g in guess_rngs if g() == r1] while len(guess_rngs) > 1: r = hidden_rng() guess_rngs = [g for g in guess_rngs if g() == r] # The top bit can not be recovered as it never affects the output, # but we should have recovered the effective hidden state. assert guess_rngs[0].state % 2**31 == hidden_rng.state % 2**31 While I did write the above process with a while loop, it appears to only ever need a third output at most to narrow it down to a single guess. Putting it together Once we could reverse-engineer the internal state of the random number generator we could make arbitrary automated decisions in the supposedly secure environment. How it worked was as follows: An insecure hook was registered that would execute right before the secure environment code would run. In this hook we have full access to information, and make a decision as to which action should be taken (e.g. casting a particular spell). This action is looked up in a hardcoded list to get an index. The current state of the RNG is reverse-engineered using the above process. We predict the outcome of the next RNG call. If this (modulo the length of our action list) does not give our desired outcome, we advance the RNG and try again. This repeats until the next random number would correspond to our desired action. The hook returns, and the secure environment starts. It generates a “random” number, indexes our hardcoded list of actions, and performs the “random” action. That’s all! By being able to simulate the RNG and looking one step ahead we could use it as our information channel by choosing exactly the right moment to call random in the secure environment. Now if you wanted to support a list of $n$ actions it would on average take $n$ steps of the RNG before the correct number came up to pass along, but that wasn’t a problem in practice. Conclusion I don’t know when Blizzard fixed the issue where the RNG state is so weak and shared, or whether they were aware of it being an issue at all. A few years after I had written the code I tried it again out of curiosity, and it had stopped working. Maybe they switched to a different algorithm, or had a properly separated RNG state for the secure environment. All-in-all it was a lot of effort for a niche exploit in a video game that I didn’t even want to use. But there certainly was a magic to manipulating something supposedly random into doing exactly what you want, like a magician pulling four aces from a shuffled deck.
More in programming
In my post yesterday (“ARM is great, ARM is terrible (and so is RISC-V)), I described my desire to find ARM hardware with AES instructions to support full-disk encryption, and the poor state of the OS ecosystem around the newer ARM boards. I was anticipating buying either a newer ARM SBC or an x86 mini … Continue reading Performant Full-Disk Encryption on a Raspberry Pi, but Foiled by Twisty UARTs →
Debates, at their finest, are about exploring topics together in search for truth. That probably sounds hopelessly idealistic to anyone who've ever perused a comment section on the internet, but ideals are there to remind us of what's possible, to inspire us to reach higher — even if reality falls short. I've been reaching for those debating ideals for thirty years on the internet. I've argued with tens of thousands of people, first on Usenet, then in blog comments, then Twitter, now X, and also LinkedIn — as well as a million other places that have come and gone. It's mostly been about technology, but occasionally about society and morality too. There have been plenty of heated moments during those three decades. It doesn't take much for a debate between strangers on this internet to escalate into something far lower than a "search for truth", and I've often felt willing to settle for just a cordial tone! But for the majority of that time, I never felt like things might escalate beyond the keyboards and into the real world. That was until we had our big blow-up at 37signals back in 2021. I suddenly got to see a different darkness from the most vile corners of the internet. Heard from those who seem to prowl for a mob-sanctioned opportunity to threaten and intimidate those they disagree with. It fundamentally changed me. But I used the experience as a mirror to reflect on the ways my own engagement with the arguments occasionally felt too sharp, too personal. And I've since tried to refocus way more of my efforts on the positive and the productive. I'm by no means perfect, and the internet often tempts the worst in us, but I resist better now than I did then. What I cannot come to terms with, though, is the modern equation of words with violence. The growing sense of permission that if the disagreement runs deep enough, then violence is a justified answer to settle it. That sounds so obvious that we shouldn't need to state it in a civil society, but clearly it is not. Not even in technology. Not even in programming. There are plenty of factions here who've taken to justify their violent fantasies by referring to their ideological opponents as "nazis", "fascists", or "racists". And then follow that up with a call to "punch a nazi" or worse. When you hear something like that often enough, it's easy to grow glib about it. That it's just a saying. They don't mean it. But I'm afraid many of them really do. Which brings us to Charlie Kirk. And the technologists who name drinks at their bar after his mortal wound just hours after his death, to name but one of the many, morbid celebrations of the famous conservative debater's death. It's sickening. Deeply, profoundly sickening. And my first instinct was exactly what such people would delight in happening. To watch the rest of us recoil, then retract, and perhaps even eject. To leave the internet for a while or forever. But I can't do that. We shouldn't do that. Instead, we should double down on the opposite. Continue to show up with our ideals held high while we debate strangers in that noble search for the truth. Where we share our excitement, our enthusiasm, and our love of technology, country, and humanity. I think that's what Charlie Kirk did so well. Continued to show up for the debate. Even on hostile territory. Not because he thought he was ever going to convince everyone, but because he knew he'd always reach some with a good argument, a good insight, or at least a different perspective. You could agree or not. Counter or be quiet. But the earnest exploration of the topics in a live exchange with another human is as fundamental to our civilization as Socrates himself. Don't give up, don't give in. Keep debating.
In my old age I’ve mostly given up trying to convince anyone of anything. Most people do not care to find the truth, they care about what pumps their bags. Some people go as far as to believe that perception is reality and that truth is a construction. I hope there’s a special place in hell for those people. It’s why the world wasted $10B+ on self driving car companies that obviously made no sense. There’s a much bigger market for truths that pump bags vs truths that don’t. So here’s your new truth that there’s no market for. Do you believe a compiler can code? If so, then go right on believing that AI can code. But if you don’t, then AI is no better than a compiler, and arguably in its current form, worse. The best model of a programming AI is a compiler. You give it a prompt, which is “the code”, and it outputs a compiled version of that code. Sometimes you’ll use it interactively, giving updates to the prompt after it has returned code, but you find that, like most IDEs, this doesn’t work all that well and you are often better off adjusting the original prompt and “recompiling”. While noobs and managers are excited that the input language to this compiler is English, English is a poor language choice for many reasons. It’s not precise in specifying things. The only reason it works for many common programming workflows is because they are common. The minute you try to do new things, you need to be as verbose as the underlying language. AI workflows are, in practice, highly non-deterministic. While different versions of a compiler might give different outputs, they all promise to obey the spec of the language, and if they don’t, there’s a bug in the compiler. English has no similar spec. Prompts are highly non local, changes made in one part of the prompt can affect the entire output. tl;dr, you think AI coding is good because compilers, languages, and libraries are bad. This isn’t to say “AI” technology won’t lead to some extremely good tools. But I argue this comes from increased amounts of search and optimization and patterns to crib from, not from any magic “the AI is doing the coding”. You are still doing the coding, you are just using a different programming language. That anyone uses LLMs to code is a testament to just how bad tooling and languages are. And that LLMs can replace developers at companies is a testament to how bad that company’s codebase and hiring bar is. AI will eventually replace programming jobs in the same way compilers replaced programming jobs. In the same way spreadsheets replaced accounting jobs. But the sooner we start thinking about it as a tool in a workflow and a compiler—through a lens where tons of careful thought has been put in—the better. I can’t believe anyone bought those vibe coding crap things for billions. Many people in self driving accused me of just being upset that I didn’t get the billions, and I’m sure it’s the same thoughts this time. Is your way of thinking so fucking broken that you can’t believe anyone cares more about the actual truth than make believe dollars? From this study, AI makes you feel 20% more productive but in reality makes you 19% slower. How many more billions are we going to waste on this? Or we could, you know, do the hard work and build better programming languages, compilers, and libraries. But that can’t be hyped up for billions.
Although it looks really good, I have not yet tried the Jujutsu (jj) version control system, mainly because it’s not yet clearly superior to Magit. But I have been following jj discussions with great interest. One of the things that jj has not yet tackled is how to do better than git refs / branches / tags. As I underestand it, jj currently has something like Mercurial bookmarks, which are more like raw git ref plumbing than a high-level porcelain feature. In particular, jj lacks signed or annotated tags, and it doesn’t have branch names that always automatically refer to the tip. This is clearly a temporary state of affairs because jj is still incomplete and under development and these gaps are going to be filled. But the discussions have led me to think about how git’s branches are unsatisfactory, and what could be done to improve them. branch merge rebase squash fork cover letters previous branch workflow questions branch One of the huge improvements in git compared to Subversion was git’s support for merges. Subversion proudly advertised its support for lightweight branches, but a branch is not very useful if you can’t merge it: an un-mergeable branch is not a tool you can use to help with work-in-progress development. The point of this anecdote is to illustrate that rather than trying to make branches better, we should try to make merges better and branches will get better as a consequence. Let’s consider a few common workflows and how git makes them all unsatisfactory in various ways. Skip to cover letters and previous branch below where I eventually get to the point. merge A basic merge workflow is, create a feature branch hack, hack, review, hack, approve merge back to the trunk The main problem is when it comes to the merge, there may be conflicts due to concurrent work on the trunk. Git encourages you to resolve conflicts while creating the merge commit, which tends to bypass the normal review process. Git also gives you an ugly useless canned commit message for merges, that hides what you did to resolve the conflicts. If the feature branch is a linear record of the work then it can be cluttered with commits to address comments from reviewers and to fix mistakes. Some people like an accurate record of the history, but others prefer the repository to contain clean logical changes that will make sense in years to come, keeping the clutter in the code review system. rebase A rebase-oriented workflow deals with the problems of the merge workflow but introduces new problems. Primarily, rebasing is intended to produce a tidy logical commit history. And when a feature branch is rebased onto the trunk before it is merged, a simple fast-forward check makes it trivial to verify that the merge will be clean (whether it uses separate merge commit or directly fast-forwards the trunk). However, it’s hard to compare the state of the feature branch before and after the rebase. The current and previous tips of the branch (amongst other clutter) are recorded in the reflog of the person who did the rebase, but they can’t share their reflog. A force-push erases the previous branch from the server. Git forges sometimes make it possible to compare a branch before and after a rebase, but it’s usually very inconvenient, which makes it hard to see if review comments have been addressed. And a reviewer can’t fetch past versions of the branch from the server to review them locally. You can mitigate these problems by adding commits in --autosquash format, and delay rebasing until just before merge. However that reintroduces the problem of merge conflicts: if the autosquash doesn’t apply cleanly the branch should have another round of review to make sure the conflicts were resolved OK. squash When the trunk consists of a sequence of merge commits, the --first-parent log is very uninformative. A common way to make the history of the trunk more informative, and deal with the problems of cluttered feature branches and poor rebase support, is to squash the feature branch into a single commit on the trunk instead of mergeing. This encourages merge requests to be roughly the size of one commit, which is arguably a good thing. However, it can be uncomfortably confining for larger features, or cause extra busy-work co-ordinating changes across multiple merge requests. And squashed feature branches have the same merge conflict problem as rebase --autosquash. fork Feature branches can’t always be short-lived. In the past I have maintained local hacks that were used in production but were not (not yet?) suitable to submit upstream. I have tried keeping a stack of these local patches on a git branch that gets rebased onto each upstream release. With this setup the problem of reviewing successive versions of a merge request becomes the bigger problem of keeping track of how the stack of patches evolved over longer periods of time. cover letters Cover letters are common in the email patch workflow that predates git, and they are supported by git format-patch. Github and other forges have a webby version of the cover letter: the message that starts off a pull request or merge request. In git, cover letters are second-class citizens: they aren’t stored in the repository. But many of the problems I outlined above have neat solutions if cover letters become first-class citizens, with a Jujutsu twist. A first-class cover letter starts off as a prototype for a merge request, and becomes the eventual merge commit. Instead of unhelpful auto-generated merge commits, you get helpful and informative messages. No extra work is needed since we’re already writing cover letters. Good merge commit messages make good --first-parent logs. The cover letter subject line works as a branch name. No more need to invent filename-compatible branch names! Jujutsu doesn’t make you name branches, giving them random names instead. It shows the subject line of the topmost commit as a reminder of what the branch is for. If there’s an explicit cover letter the subject line will be a better summary of the branch as a whole. I often find the last commit on a branch is some post-feature cleanup, and that kind of commit has a subject line that is never a good summary of its feature branch. As a prototype for the merge commit, the cover letter can contain the resolution of all the merge conflicts in a way that can be shared and reviewed. In Jujutsu, where conflicts are first class, the cover letter commit can contain unresolved conflicts: you don’t have to clean them up when creating the merge, you can leave that job until later. If you can share a prototype of your merge commit, then it becomes possible for your collaborators to review any merge conflicts and how you resolved them. To distinguish a cover letter from a merge commit object, a cover letter object has a “target” header which is a special kind of parent header. A cover letter also has a normal parent commit header that refers to earlier commits in the feature branch. The target is what will become the first parent of the eventual merge commit. previous branch The other ingredient is to add a “previous branch” header, another special kind of parent commit header. The previous branch header refers to an older version of the cover letter and, transitively, an older version of the whole feature branch. Typically the previous branch header will match the last shared version of the branch, i.e. the commit hash of the server’s copy of the feature branch. The previous branch header isn’t changed during normal work on the feature branch. As the branch is revised and rebased, the commit hash of the cover letter will change fairly frequently. These changes are recorded in git’s reflog or jj’s oplog, but not in the “previous branch” chain. You can use the previous branch chain to examine diffs between versions of the feature branch as a whole. If commits have Gerrit-style or jj-style change-IDs then it’s fairly easy to find and compare previous versions of an individual commit. The previous branch header supports interdiff code review, or allows you to retain past iterations of a patch series. workflow Here are some sketchy notes on how these features might work in practice. One way to use cover letters is jj-style, where it’s convenient to edit commits that aren’t at the tip of a branch, and easy to reshuffle commits so that a branch has a deliberate narrative. When you create a new feature branch, it starts off as an empty cover letter with both target and parent pointing at the same commit. Alternatively, you might start a branch ad hoc, and later cap it with a cover letter. If this is a small change and rebase + fast-forward is allowed, you can edit the “cover letter” to contain the whole change. Otherwise, you can hack on the branch any which way. Shuffle the commits that should be part of the merge request so that they occur before the cover letter, and edit the cover letter to summarize the preceding commits. When you first push the branch, there’s (still) no need to give it a name: the server can see that this is (probably) going to be a new merge request because the top commit has a target branch and its change-ID doesn’t match an existing merge request. Also when you push, your client automatically creates a new instance of your cover letter, adding a “previous branch” header to indicate that the old version was shared. The commits on the branch that were pushed are now immutable; rebases and edits affect the new version of the branch. During review there will typically be multiple iterations of the branch to address feedback. The chain of previous branch headers allows reviewers to see how commits were changed to address feedback, interdiff style. The branch can be merged when the target header matches the current trunk and there are no conflicts left to resolve. When the time comes to merge the branch, there are several options: For a merge workflow, the cover letter is used to make a new commit on the trunk, changing the target header into the first parent commit, and dropping the previous branch header. Or, if you like to preserve more history, the previous branch chain can be retained. Or you can drop the cover letter and fast foward the branch on to the trunk. Or you can squash the branch on to the trunk, using the cover letter as the commit message. questions This is a fairly rough idea: I’m sure that some of the details won’t work in practice without a lot of careful work on compatibility and deployability. Do the new commit headers (“target” and “previous branch”) need to be headers? What are the compatibility issues with adding new headers that refer to other commits? How would a server handle a push of an unnamed branch? How could someone else pull a copy of it? How feasible is it to use cover letter subject lines instead of branch names? The previous branch header is doing a similar job to a remote tracking branch. Is there an opportunity to simplify how we keep a local cache of the server state? Despite all that, I think something along these lines could make branches / reviews / reworks / merges less awkward. How you merge should me a matter of your project’s preferred style, without interference from technical limitations that force you to trade off one annoyance against another. There remains a non-technical limitation: I have assumed that contributors are comfortable enough with version control to use a history-editing workflow effectively. I’ve lost all perspective on how hard this is for a newbie to learn; I expect (or hope?) jj makes it much easier than git rebase.
I’ve long been interested in new and different platforms. I ran Debian on an Alpha back in the late 1990s and was part of the Alpha port team; then I helped bootstrap Debian on amd64. I’ve got somewhere around 8 Raspberry Pi devices in active use right now, and the free NNCPNET Internet email service … Continue reading ARM is great, ARM is terrible (and so is RISC-V) →