Having spent four decades as a programmer in various industries and situations, I know that modern software development processes are far more stressful than when I started. It's not simply that developing software today is more complex than it was back in 1981. In that early decade, none

In previous articles, we saw how to use “real” UART, and looked into the trick used by Arduino to automatically reset boards when uploading firmware. Today, we’ll look into how Espressif does something similar, using even more tricks. “Real” UART on the Saola As usual, let’s first simply connect the UART adapter. Again, we connect … Continue reading Espressif’s Automatic Reset → The post Espressif’s Automatic Reset appeared first on Quentin Santos.

Lessons for AI prompting and retrieval

Hey peoples! Tonight, some meta-words. As you know I am fascinated by compilers and language implementations, and I just want to know all the things and implement all the fun stuff: intermediate representations, flow-sensitive source-to-source optimization passes, register allocation, instruction selection, garbage collection, all of that. It started long ago with a combination of curiosity and a hubris to satisfy that curiosity. The usual way to slake such a thirst is structured higher education followed by industry apprenticeship, but for whatever reason my path sent me through a nuclear engineering bachelor’s program instead of computer science, and continuing that path was so distasteful that I noped out all the way to rural Namibia for a couple years. Fast-forward, after 20 years in the programming industry, and having picked up some language implementation experience, a few years ago I returned to garbage collection. I have a good level of language implementation chops but never wrote a memory manager, and Guile’s performance was limited by its use of the Boehm collector. I had been on the lookout for something that could help, and when I learned of it seemed to me that the only thing missing was an appropriate implementation for Guile, and hey I could do that!Immix I started with the idea of an -style interface to a memory manager that was abstract enough to be implemented by a variety of different collection algorithms. This kind of abstraction is important, because in this domain it’s easy to convince oneself that a given algorithm is amazing, just based on vibes; to stay grounded, I find I always need to compare what I am doing to some fixed point of reference. This GC implementation effort grew into , but as it did so a funny thing happened: the as a direct replacement for the Boehm collector maintained mark bits in a side table, which I realized was a suitable substrate for Immix-inspired bump-pointer allocation into holes. I ended up building on that to develop an Immix collector, but without lines: instead each granule of allocation (16 bytes for a 64-bit system) is its own line.MMTkWhippetmark-sweep collector that I prototyped The is funny, because it defines itself as a new class of collector, fundamentally different from the three other fundamental algorithms (mark-sweep, mark-compact, and evacuation). Immix’s are blocks (64kB coarse-grained heap divisions) and lines (128B “fine-grained” divisions); the innovation (for me) is the discipline by which one can potentially defragment a block without a second pass over the heap, while also allowing for bump-pointer allocation. See the papers for the deets!Immix papermark-regionregionsoptimistic evacuation However what, really, are the regions referred to by ? If they are blocks, then the concept is trivial: everyone has a block-structured heap these days. If they are spans of lines, well, how does one choose a line size? As I understand it, Immix’s choice of 128 bytes was to be fine-grained enough to not lose too much space to fragmentation, while also being coarse enough to be eagerly swept during the GC pause.mark-region This constraint was odd, to me; all of the mark-sweep systems I have ever dealt with have had lazy or concurrent sweeping, so the lower bound on the line size to me had little meaning. Indeed, as one reads papers in this domain, it is hard to know the real from the rhetorical; the review process prizes novelty over nuance. Anyway. What if we cranked the precision dial to 16 instead, and had a line per granule? That was the process that led me to Nofl. It is a space in a collector that came from mark-sweep with a side table, but instead uses the side table for bump-pointer allocation. Or you could see it as an Immix whose line size is 16 bytes; it’s certainly easier to explain it that way, and that’s the tack I took in a .recent paper submission to ISMM’25 Wait what! I have a fine job in industry and a blog, why write a paper? Gosh I have meditated on this for a long time and the answers are very silly. Firstly, one of my language communities is Scheme, which was a research hotbed some 20-25 years ago, which means many practitioners—people I would be pleased to call peers—came up through the PhD factories and published many interesting results in academic venues. These are the folks I like to hang out with! This is also what academic conferences are, chances to shoot the shit with far-flung fellows. In Scheme this is fine, my work on Guile is enough to pay the intellectual cover charge, but I need more, and in the field of GC I am not a proven player. So I did an atypical thing, which is to cosplay at being an independent researcher without having first been a dependent researcher, and just solo-submit a paper. Kids: if you see yourself here, just go get a doctorate. It is not easy but I can only think it is a much more direct path to goal. And the result? Well, friends, it is this blog post :) I got the usual assortment of review feedback, from the very sympathetic to the less so, but ultimately people were confused by leading with a comparison to Immix but ending without an evaluation against Immix. This is fair and the paper does not mention that, you know, I don’t have an Immix lying around. To my eyes it was a good paper, an , but, you know, just a try. I’ll try again sometime.80% paper In the meantime, I am driving towards getting Whippet into Guile. I am hoping that sometime next week I will have excised all the uses of the BDW (Boehm GC) API in Guile, which will finally allow for testing Nofl in more than a laboratory environment. Onwards and upwards! whippet regions? paper??!?

I started writing this early last week but Real Life Stuff happened and now you're getting the first-draft late this week. Warning, unedited thoughts ahead! New Logic for Programmers release! v0.9 is out! This is a big release, with a new cover design, several rewritten chapters, online code samples and much more. See the full release notes at the changelog page, and get the book here! Write the cleverest code you possibly can There are millions of articles online about how programmers should not write "clever" code, and instead write simple, maintainable code that everybody understands. Sometimes the example of "clever" code looks like this (src): # Python p=n=1 exec("p*=n*n;n+=1;"*~-int(input())) print(p%n) This is code-golfing, the sport of writing the most concise code possible. Obviously you shouldn't run this in production for the same reason you shouldn't eat dinner off a Rembrandt. Other times the example looks like this: def is_prime(x): if x == 1: return True return all([x%n != 0 for n in range(2, x)] This is "clever" because it uses a single list comprehension, as opposed to a "simple" for loop. Yes, "list comprehensions are too clever" is something I've read in one of these articles. I've also talked to people who think that datatypes besides lists and hashmaps are too clever to use, that most optimizations are too clever to bother with, and even that functions and classes are too clever and code should be a linear script.1. Clever code is anything using features or domain concepts we don't understand. Something that seems unbearably clever to me might be utterly mundane for you, and vice versa. How do we make something utterly mundane? By using it and working at the boundaries of our skills. Almost everything I'm "good at" comes from banging my head against it more than is healthy. That suggests a really good reason to write clever code: it's an excellent form of purposeful practice. Writing clever code forces us to code outside of our comfort zone, developing our skills as software engineers. Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you [will get excellent debugging practice at exactly the right level required to push your skills as a software engineer] — Brian Kernighan, probably There are other benefits, too, but first let's kill the elephant in the room:2 Don't commit clever code I am proposing writing clever code as a means of practice. Being at work is a job with coworkers who will not appreciate if your code is too clever. Similarly, don't use too many innovative technologies. Don't put anything in production you are uncomfortable with. We can still responsibly write clever code at work, though: Solve a problem in both a simple and a clever way, and then only commit the simple way. This works well for small scale problems where trying the "clever way" only takes a few minutes. Write our personal tools cleverly. I'm a big believer of the idea that most programmers would benefit from writing more scripts and support code customized to their particular work environment. This is a great place to practice new techniques, languages, etc. If clever code is absolutely the best way to solve a problem, then commit it with extensive documentation explaining how it works and why it's preferable to simpler solutions. Bonus: this potentially helps the whole team upskill. Writing clever code... ...teaches simple solutions Usually, code that's called too clever composes several powerful features together — the "not a single list comprehension or function" people are the exception. Josh Comeau's "don't write clever code" article gives this example of "too clever": const extractDataFromResponse = (response) => { const [Component, props] = response; const resultsEntries = Object.entries({ Component, props }); const assignIfValueTruthy = (o, [k, v]) => (v ? { ...o, [k]: v } : o ); return resultsEntries.reduce(assignIfValueTruthy, {}); } What makes this "clever"? I count eight language features composed together: entries, argument unpacking, implicit objects, splats, ternaries, higher-order functions, and reductions. Would code that used only one or two of these features still be "clever"? I don't think so. These features exist for a reason, and oftentimes they make code simpler than not using them. We can, of course, learn these features one at a time. Writing the clever version (but not committing it) gives us practice with all eight at once and also with how they compose together. That knowledge comes in handy when we want to apply a single one of the ideas. I've recently had to do a bit of pandas for a project. Whenever I have to do a new analysis, I try to write it as a single chain of transformations, and then as a more balanced set of updates. ...helps us master concepts Even if the composite parts of a "clever" solution aren't by themselves useful, it still makes us better at the overall language, and that's inherently valuable. A few years ago I wrote Crimes with Python's Pattern Matching. It involves writing horrible code like this: from abc import ABC class NotIterable(ABC): @classmethod def __subclasshook__(cls, C): return not hasattr(C, "__iter__") def f(x): match x: case NotIterable(): print(f"{x} is not iterable") case _: print(f"{x} is iterable") if __name__ == "__main__": f(10) f("string") f([1, 2, 3]) This composes Python match statements, which are broadly useful, and abstract base classes, which are incredibly niche. But even if I never use ABCs in real production code, it helped me understand Python's match semantics and Method Resolution Order better. ...prepares us for necessity Sometimes the clever way is the only way. Maybe we need something faster than the simplest solution. Maybe we are working with constrained tools or frameworks that demand cleverness. Peter Norvig argued that design patterns compensate for missing language features. I'd argue that cleverness is another means of compensating: if our tools don't have an easy way to do something, we need to find a clever way. You see this a lot in formal methods like TLA+. Need to check a hyperproperty? Cast your state space to a directed graph. Need to compose ten specifications together? Combine refinements with state machines. Most difficult problems have a "clever" solution. The real problem is that clever solutions have a skill floor. If normal use of the tool is at difficult 3 out of 10, then basic clever solutions are at 5 out of 10, and it's hard to jump those two steps in the moment you need the cleverness. But if you've practiced with writing overly clever code, you're used to working at a 7 out of 10 level in short bursts, and then you can "drop down" to 5/10. I don't know if that makes too much sense, but I see it happen a lot in practice. ...builds comradery On a few occasions, after getting a pull request merged, I pulled the reviewer over and said "check out this horrible way of doing the same thing". I find that as long as people know they're not going to be subjected to a clever solution in production, they enjoy seeing it! Next week's newsletter will probably also be late, after that we should be back to a regular schedule for the rest of the summer. Mostly grad students outside of CS who have to write scripts to do research. And in more than one data scientist. I think it's correlated with using Jupyter. ↩ If I don't put this at the beginning, I'll get a bajillion responses like "your team will hate you" ↩