I can’t for the life of me seem to figure out what to do with a clojure program after I have written it. How do I take a .clj file and turn it into a .class?
I don’t want to set up a leingen project or whatever, I don’t want to bundle it as a jar, all I want to do is write a .clj file in a text editor and turn it into a .class so I can run it like java ClassName
Author of the article here. Clojure is more commonly loaded and compiled on the fly, so the easiest way to run a single .clj file is to just do clojure your_file.clj. This uses the official Clojure CLI. Compiling to class files (AOT compilation) is usually reserved for production deployments, where you build an (uber)jar.
That said if you want to just compile a single namespace, you do that with Clojure’s compile function. It’ll look for the namespace on the classpath (defaults to src), and write the resulting classes to classes.
Why do you want it to run like java ClassName in the first place? To me that seems a bit arbitrary, and really doesn’t work unless you add the Clojure runtime on the classpath.
Though I assume you just want to run a .clj file as a command on the command line, and there Babaskha is the de-facto standard tool to that these days.
I still spend inordinate amounts of time playing Heroes 3. It’s been a strong procrastination feeder over the decades, but it also gave me a lot of fun!
I mostly play HotA in HD Mod these days, but I had used VCMI on Android (with SoD). It’s been a bit rough – the client-server thing that it does would have connection issues, the mechanics differed in subtle ways, etc. But I’m also super grateful that it exists.
The projectile-grep(s) are really good. Be sure to ignore entire directories with a .projectile file (like node_modules and similar) to speed things up. Also wgrep is a good package to edit grep results across many files, in one buffer, and then saving all changes at once.
I’ve used the DOS build of OCaml! Back in 2000, in my freshman year at the uni, I was using a 486SX/33 with 4 MB RAM, dual-booting DOS / Monkey Linux. The latter didn’t support OCaml as it was libc5-based, so I had to resort to the DOS build. Used it for some Programming 101 assignments.
Incidentally, I just migrated away from syntax-highlighting code snippets on my blog with highlight.js client-side, although for different reasons than outlined in this article.
I really wanted all my blog posts to not include any JS because, in my view, this makes them documents and not applications (I elaborate on the distinction here). The bespoke static-site generator that I use for my sites already included code to not include highlight.js if the page contains no code snippets. Using highlight.js was just convenient.
Interestingly, I migrated from it to using highlight.js server-side. My SSG is written in a JVM language (Clojure), so it was really easy to add Graal.js as a dependency (I tried Nashorn first, but it barfs on some ES2015 constructs used in highlight.js), call out to highlight.js from Clojure, and plug it into the Markdown parser that I use.
the problem with this is then- how do you address server-side scripting? i can have a server return different pages based on the nature of the request and hence implement paywalls etc. not very document-y, but can be achieved with the limitations you describe.
This isn’t really a quirk. It was usually a given that you should not access stack space that is not yours. Only in “recent times” have red zones been agreed upon.
This and other techniques in the past have been used as anti-debugger tricks.
Exactly. Any interrupt handler (for timer, keyboard, etc) is going to use the stack in the same way as the debugger does. It’s not a “debugging quirk” at all; it’s code that is incorrect and that is failing in a condition that is exacerbated by the use of a debugger.
The stack is arranged as a contiguous array of memory locations. A pointer to the base of the stack is in the %esp register. The base of the stack, 0(%esp), contains the return-point. The return-point is an address in memory where we return after computing the value and therefore should not be modified. We are free to use the memory locations above the return-point (-4(%esp), -8(%esp), -12(%esp), etc.) to hold our intermediate values.
Does that mean that his approach is fundamentally wrong?
Wow. I read that paper years ago. Great paper. However I must’ve skimmed over these parts.
Yes, the approach is fundamentally wrong.
The code used in the paper is 32 bit code. The ABI does not guarantee that the stack cannot be touched.
Now in modern x86-64 code, they have a red zone. The above approach would work totally fine there.
On any protected mode operating system, when an interrupt/syscall occurs it will use the kernel stack instead of your stack. This is why the paper got away with it on a 32 bit OS.
On a DOS machine however, any interrupt would’ve corrupted that data.
Your a DOS port just needs to reserve a little space on the stack for each function call. It might also be helpful to just use a common prologue with a frame pointer.
I tried few-shot training GPT-3 by copying 4-5 HN headlines and their n-gate summaries into the prompt. Then I pasted some headlines from HN’s frontpage today:
Microsoft Windows is prohibited at Gitlab
Gitlab (business model: “Uber for Git”) has decided that it will no longer support Microsoft Windows (business model: “Uber for proprietary software”). Hackernews, who have all been using Macs for the past ten years and have never even heard of Windows, are shocked and appalled that anyone would use such a terrible operating system. They spend the next few hours patting themselves on the back for making the right choice, and then get into a heated argument over which text editor is the best.
Bringing back the golden days of Bell Labs
A New York Times article (business model: “Uber for articles about the good old days”) about an attempt to recreate Bell Labs (business model: “Uber for the telephone”) in Silicon Valley. Hackernews, most of whom are too young to remember the good old days, bemoan the fact that everything was better back then and that the world is going to shit. A few Hackernews try to explain to the others that the world has always been going to shit, and that things were never as good as they seem in retrospect.
Alex Jones must pay $4m in damages for Sandy Hook hoax claims
Alex Jones (business model: “Uber for bullshit”) has been ordered to pay $4 million in damages for claiming that the Sandy Hook massacre was a hoax. Hackernews, who have never heard of Alex Jones or Sandy Hook, spend the next few hours trying to figure out what a hoax is.
Is DALL-E 2 ‘gluing things together’ without understanding their relationships?
OpenAI (business model: “Uber for AI”) releases a new version of its DALL-E image-generation algorithm, which is able to generate images from textual descriptions. Hackernews, who are not sure what an algorithm is, spends the next few hours trying to figure out what an image is.
A New York Times article (business model: “Uber for articles about the good old days”)
Hackernews, most of whom are too young to remember the good old days, bemoan the fact that everything was better back then and that the world is going to shit.
Alex Jones (business model: “Uber for bullshit”)
Okay, I think it’s safe to say that GPT-3 just passed the Turing test with flying colours :-D
Seriously though, these systems seem pretty nifty for brainstorming, but the extreme cost of re-training makes it difficult to generate something topical. Even OpenAI’s newest, biggest, most expensive text model (text-davinci-002) is running a year behind on training data.
Tried “Write a dark-humorous one-sentence summary of the following article:\n%s\n\n” on gpt-3 playground (using text-davinci-002) for the first few lines of https://github.com/hackergrrl/art-of-readme, and I got this:
A dark-humorous one-sentence summary of the article would be: “The README file is a long-standing tradition of developers trying to be helpful, but really just annoying users.”
I can’t for the life of me seem to figure out what to do with a clojure program after I have written it. How do I take a
.cljfile and turn it into a.class?I don’t want to set up a leingen project or whatever, I don’t want to bundle it as a jar, all I want to do is write a .clj file in a text editor and turn it into a .class so I can run it like
java ClassNameAuthor of the article here. Clojure is more commonly loaded and compiled on the fly, so the easiest way to run a single
.cljfile is to just doclojure your_file.clj. This uses the official Clojure CLI. Compiling to class files (AOT compilation) is usually reserved for production deployments, where you build an (uber)jar.That said if you want to just compile a single namespace, you do that with Clojure’s
compilefunction. It’ll look for the namespace on the classpath (defaults tosrc), and write the resulting classes toclasses.Then from the shell:
You’ll need the Clojure and Spec jars to be able to run this:
Why a
.class? I use this trick to make my.cljfile executable:https://gist.github.com/yogsototh/dc16e3213115c8305720f8c74d4a8297
If you’re using
deps.ednto build your clojure project you probably wanttools.buildto produce an uberjar. https://clojure.org/guides/tools_build#_compiled_uberjar_application_buildWhy do you want it to run like
java ClassNamein the first place? To me that seems a bit arbitrary, and really doesn’t work unless you add the Clojure runtime on the classpath.Though I assume you just want to run a .clj file as a command on the command line, and there Babaskha is the de-facto standard tool to that these days.
Here’s a raw way with cli-tools:
Obviously, if you want an easier way of setting up the classpath, then the tooling is there for you. But if you don’t want to use it…
I still spend inordinate amounts of time playing Heroes 3. It’s been a strong procrastination feeder over the decades, but it also gave me a lot of fun!
I mostly play HotA in HD Mod these days, but I had used VCMI on Android (with SoD). It’s been a bit rough – the client-server thing that it does would have connection issues, the mechanics differed in subtle ways, etc. But I’m also super grateful that it exists.
Given the title, I find it amusing that the code is written in Go. :)
The first lobsters tag being “go” makes it funny to read too
Ubuntu ubuntu ubuntu ubuntu bonaaanzaaaaa!!!
There’s a lot. My daily drivers come to mind:
And many, many more!
I’m using projectile-ripgrep (fits in neatly into Projectile).
The projectile-grep(s) are really good. Be sure to ignore entire directories with a .projectile file (like node_modules and similar) to speed things up. Also wgrep is a good package to edit grep results across many files, in one buffer, and then saving all changes at once.
Another neat one:
[[[[[[[[[[FF]]]]RC]]]R]]]. This one is symmetric; moving the C around produces differently-coloured versions.Nice! Here’s my little creation:
[[[[[[[[[F]]R]RCC]RR]]R]]]I’ve used the DOS build of OCaml! Back in 2000, in my freshman year at the uni, I was using a 486SX/33 with 4 MB RAM, dual-booting DOS / Monkey Linux. The latter didn’t support OCaml as it was libc5-based, so I had to resort to the DOS build. Used it for some Programming 101 assignments.
The other day, I found myself in need of visualising a tree of paths, like this:
I ended up converting it to a JSON with a one-off Python script, and then using fx for the viewing. It worked very well! Thank you for the tool!
(Incidentally, I’ve been writing my own viewer to satisfy my original need more straightforwardly, but that’s still in very early stages.)
See also gron.
I’ve been learning Rust by attempting to write a TUI tree viewer.
Playing in a Polish Scrabble tournament. I’m not a top-notch player, but I’m reasonably strong, so we’ll see. :)
Incidentally, I just migrated away from syntax-highlighting code snippets on my blog with highlight.js client-side, although for different reasons than outlined in this article.
I really wanted all my blog posts to not include any JS because, in my view, this makes them documents and not applications (I elaborate on the distinction here). The bespoke static-site generator that I use for my sites already included code to not include highlight.js if the page contains no code snippets. Using highlight.js was just convenient.
Interestingly, I migrated from it to using highlight.js server-side. My SSG is written in a JVM language (Clojure), so it was really easy to add Graal.js as a dependency (I tried Nashorn first, but it barfs on some ES2015 constructs used in highlight.js), call out to highlight.js from Clojure, and plug it into the Markdown parser that I use.
Since I saw https://github.com/harfbuzz/harfbuzz/blob/main/docs/wasm-shaper.md, I am re-evaluating what a document even is.
I was thinking how to make the web smaller as well, and came up with a different proposal: https://blog.danieljanus.pl/2019/10/07/web-of-documents/
I wrote that before I became aware of Gemini, though.
the problem with this is then- how do you address server-side scripting? i can have a server return different pages based on the nature of the request and hence implement paywalls etc. not very document-y, but can be achieved with the limitations you describe.
You can, but the lack of cookies forces you to put all state into the URL. It’s doable, but unwieldy.
How about this C? It does have a main(), but with empty body!
That’s not allowed per C89 (no variable length arrays). It may be legal under C99 (at least, GCC and clang accept it as C99) but … ew.
Is this related to remacs in any way, or is this a completely independent project?
It seems it’s a different project as they say:
This isn’t really a quirk. It was usually a given that you should not access stack space that is not yours. Only in “recent times” have red zones been agreed upon. This and other techniques in the past have been used as anti-debugger tricks.
Exactly. Any interrupt handler (for timer, keyboard, etc) is going to use the stack in the same way as the debugger does. It’s not a “debugging quirk” at all; it’s code that is incorrect and that is failing in a condition that is exacerbated by the use of a debugger.
Original author here. I encountered this while working on a port of Ghuloum’s Scheme compiler (http://scheme2006.cs.uchicago.edu/11-ghuloum.pdf). Ghuloum says (in section 3.4):
Does that mean that his approach is fundamentally wrong?
Wow. I read that paper years ago. Great paper. However I must’ve skimmed over these parts.
Yes, the approach is fundamentally wrong.
The code used in the paper is 32 bit code. The ABI does not guarantee that the stack cannot be touched.
Now in modern x86-64 code, they have a red zone. The above approach would work totally fine there.
On any protected mode operating system, when an interrupt/syscall occurs it will use the kernel stack instead of your stack. This is why the paper got away with it on a 32 bit OS.
On a DOS machine however, any interrupt would’ve corrupted that data.
Your a DOS port just needs to reserve a little space on the stack for each function call. It might also be helpful to just use a common prologue with a frame pointer.
Can we have a snarky version, to replace n-gate.com?
I tried few-shot training GPT-3 by copying 4-5 HN headlines and their n-gate summaries into the prompt. Then I pasted some headlines from HN’s frontpage today:
Microsoft Windows is prohibited at Gitlab
Gitlab (business model: “Uber for Git”) has decided that it will no longer support Microsoft Windows (business model: “Uber for proprietary software”). Hackernews, who have all been using Macs for the past ten years and have never even heard of Windows, are shocked and appalled that anyone would use such a terrible operating system. They spend the next few hours patting themselves on the back for making the right choice, and then get into a heated argument over which text editor is the best.
Bringing back the golden days of Bell Labs
A New York Times article (business model: “Uber for articles about the good old days”) about an attempt to recreate Bell Labs (business model: “Uber for the telephone”) in Silicon Valley. Hackernews, most of whom are too young to remember the good old days, bemoan the fact that everything was better back then and that the world is going to shit. A few Hackernews try to explain to the others that the world has always been going to shit, and that things were never as good as they seem in retrospect.
Alex Jones must pay $4m in damages for Sandy Hook hoax claims
Alex Jones (business model: “Uber for bullshit”) has been ordered to pay $4 million in damages for claiming that the Sandy Hook massacre was a hoax. Hackernews, who have never heard of Alex Jones or Sandy Hook, spend the next few hours trying to figure out what a hoax is.
Is DALL-E 2 ‘gluing things together’ without understanding their relationships?
OpenAI (business model: “Uber for AI”) releases a new version of its DALL-E image-generation algorithm, which is able to generate images from textual descriptions. Hackernews, who are not sure what an algorithm is, spends the next few hours trying to figure out what an image is.
Okay, I think it’s safe to say that GPT-3 just passed the Turing test with flying colours :-D
This is pretty much perfect!
This would be the only weekly newsletter I would subscribe to, awesome.
These are so good that I’m almost tempted to ask for evidence that you actually ran these through GPT-3 instead of writing them yourself.
I wonder how far are we from a Netflix comedy special that’s entirely performed live by an AI…
We’re already there!
Seriously though, these systems seem pretty nifty for brainstorming, but the extreme cost of re-training makes it difficult to generate something topical. Even OpenAI’s newest, biggest, most expensive text model (text-davinci-002) is running a year behind on training data.
I tried replacing the GPT-3 prompt with:
Here is the result. Not particularly snarky. Maybe GPT-3 needs to be shown an example of the tone in order to maintain it.
Tried “
Write a dark-humorous one-sentence summary of the following article:\n%s\n\n” on gpt-3 playground (using text-davinci-002) for the first few lines of https://github.com/hackergrrl/art-of-readme, and I got this:Please @nathell provide different toned variants like this!
Fun fact: it was RMS who coined the name POSIX in the first place.
https://stallman.org/articles/posix.html
You might be interested in Movitz; it’s a freestanding Common Lisp system for x86 PCs.
I know! It was one of the inspirations.