Full Width 
Shortcuts 
Sign Up 
Log In 
Articles 
Reading List More from dthompson
I'm quite a bit late on this one, but Haunt version 0.4.0 was released released back in July. I haven't had much time for blogging, but I'm catching up now! This release contains a small set of improvements and bug fixes since the 0.3.0 release in 2024. About Haunt Haunt is a static site generator that uses the Guile Scheme as its configuration language. It aims to be simple, functional, and extensible. Features include: Easy blog and Atom/RSS feed generation Markdown post support Simple development server for viewing edits before publishing Purely functional build process User extensibility Notable changes Added support for HTML in Markdown documents. This was a long time coming because guile-markdown did not support it and the library was abandoned by the original maintainer. As part of my work at Spritely, we forked it, implemented the relevant portions of the CommonMark specification, and released it. Spritely's guile-commonmark fork is now considered to be the official upstream by Guix and others. A further consequence of this is that guile-lib is now a required dependency for building Haunt as we need the (htmlprag) module to parse Markdown documents with embedded HTML. html->shtml from guile-lib's (htmlprag) module is now used instead of xml->sxml in the HTML reader. It was silly of me to use xml->sxml for this purpose years ago, but at the time I wanted guile-lib to be an optional dependency. Added haunt new subcommand for creating a new site. Added default directory, template, and prefix arguments to flat-pages procedure. Added support for index metadata flag to flat pages for pretty URLs. Flat pages now receive all page metadata, not just the page title. This is a breaking change from 0.3.0. Added .scm as an additional extension for sxml-reader. make-file-extension-matcher now supports multiple extensions. Fixed emission of <script> and <style> elements. Fixed handling of no available reader in flat pages builder. Fixed unreachable error handling clause when a reader is not found for a post. Fixed default blog theme template missing an <html> tag. Fixed overloaded -h option in haunt serve. Deprecated post in Skribe reader in favor of document. Download Haunt 0.4.0 is already available in Guix: guix pull guix install haunt See the Haunt project page for information on how to build from source. Thank you to Camilo Rodrigues, Noé Lopez, jgart, Jakob L. Kreuze, and Daniel Meißner for their contributions to this release! Happy haunting!
I'm happy to announce that guile-bstructs 0.2.0 has been released! About guile-bstructs Guile-bstructs is a library that provides structured read/write access to binary data for Guile. A bstruct (short for “binary structure”) is a data type that encapsulates a bytevector and a byte offset which interprets that bytevector based on a specified layout. See the guile-bstructs project page for more information. Notable changes Added support for anonymous structs and unions. Contrived example: (define-bstruct <location> (struct (id int) (union (struct (x double) (y double) (z double)) (coords (array 3 double))))) Made bstruct-wrap offset optional. Added bstruct-by-value syntax. Added void primitive type. Added support for array elements to bstruct-ref, bstruct-set, etc. Bug fixes Fixed guard clause for bstruct-by-value. Added missing symbolic-match? check in define-bstruct-primitive. Internal type descriptor ids are now stable to allow for clean redefinition at the REPL. Download Starting with this release, I am no longer uploading release tarballs. This is due to Guix moving away from them and for good reason after events like the xz-utils backdoor. Build from the v0.2.0 tag in Git, instead.
I'm happy to announce that guile-websocket 0.3.0 has been released! Guile-websocket is an implementation of the WebSocket protocol, both the client and server sides, for Guile Scheme. Highlights for this release are: New (web socket) module that provides a standard Scheme port interface for WebSockets. WebSocket is a framed protocol, not a simple bytestream, but the new wrappers websocket and wrap-websocket provide the useful illusion of a raw bytestream. The #:max-attempts argument to read-data-frame can now be #f, meaning that the read operation should never give up in the case of a read timeout. This is useful for keeping server-side client connections alive as long as the client hasn't hung up, even if ping frames haven't been sent in awhile. New websocket-upgrade-request?, websocket-upgrade-response?, and make-websocket-upgrade-response procedures in (web socket server). These new procedures allow for integration of WebSockets into custom Guile HTTP servers rather than having to use the overly simplistic built-in one that only handles WebSocket requests. The #:configure-socket argument to open-websocket-for-uri has been deprecated. In practice, user code just needs to set some socket flags, particularly SOCK_NONBLOCK for asynchronous I/O. Use the new #:flags argument instead. source tarball: https://files.dthompson.us/releases/guile-websocket/guile-websocket-0.3.0.tar.gz signature: https://files.dthompson.us/releases/guile-websocket/guile-websocket-0.3.0.tar.gz.asc See the guile-websocket project page for more information. Bug reports, bug fixes, feature requests, and patches are welcomed.
I'm happy to announce that guile-websocket 0.2.1 has been released! Guile-websocket is an implementation of the WebSocket protocol, both the client and server sides, for Guile Scheme. This small patch release contains the following changes: Greatly improved validation of initial handshake request on server. In previous releases, when bots discovered a server and sent random HTTP requests they’d cause needless exceptions to be thrown and clog up server logs. The accept-client hook can now return #t when a new client has not connected but the server loop should continue anyway. source tarball: https://files.dthompson.us/releases/guile-websocket/guile-websocket-0.2.1.tar.gz signature: https://files.dthompson.us/releases/guile-websocket/guile-websocket-0.2.1.tar.gz.asc See the guile-websocket project page for more information. Bug reports, bug fixes, feature requests, and patches are welcomed.
I'm pleased to announce that the very first release of guile-bstructs, version 0.1.0, has been released! This is a library I've been working on for quite some time and after more than one rewrite and many smaller refactors I think it's finally ready to release publicly. Let's hope I'm not wrong about that! About guile-bstructs Guile-bstructs is a library that provides structured read/write access to binary data for Guile. A bstruct (short for “binary structure”) is a data type that encapsulates a bytevector and a byte offset which interprets that bytevector based on a specified layout. Some use cases for bstructs are: manipulating C structs when using the foreign function interface packing GPU vertex buffers when using graphics APIs such as OpenGL implementing data types that benefit from Guile's unboxed math optimizations such as vectors and matrices This library was initially inspired by guile-opengl's define-packed-struct syntax but is heavily based on "Ftypes: Structured foreign types" by Andy Keep and R. Kent Dybvig. The resulting interface is quite similar but the implementation is completely original. This library provides a syntax-heavy interface; nearly all of the public API is syntax. This is done to ensure that bstruct types are static and well-known at compile time resulting in efficient bytecode and minimal runtime overhead. A subset of the interface deals in raw bytevector access for accessing structured data in bytevectors directly without going through an intermediary bstruct wrapper. This low-level interface is useful for certain batch processing situations where the overhead of creating wrapper bstructs would hinder throughput. Example Here is some example code to give you an idea of what it’s like to use guile-bstructs: ;; Struct (define-bstruct <vec2> (struct (x float) (y float))) ;; Get size/alignment (bstruct-sizeof <vec2>) ; => 8 (bstruct-alignof <vec2>) ; => 4 ;; Allocation (define v (bstruct-alloc <vec2> (x 42.0) (y 61.0))) ;; Check type (bstruct? <vec2> v) ; => #t ;; Read (bstruct-ref <vec2> v x) ; => 42.0 (bstruct-ref <vec2> v x y) ; => 42.0, 61.0 ;; Write (bstruct-set! <vec2> v (x 0.0) (y 0.0)) ;; Convert to s-expression (bstruct->sexp <vec2> v) ; => ((struct (x 42.0) (y 61.0))) ;; Convert FFI pointer (pointer->bstruct <vec2> my-pointer) ;; More type examples: ;; Type group with a union (define-bstruct (<mouse-move-event> (struct (type uint8) (x int32) (y int32))) (<mouse-button-event> (struct (type uint8) (button uint8) (state uint8) (x int32) (y int32))) (<event> (union (type uint8) (mouse-move <mouse-move-event>) (mouse-button <mouse-button-event>)))) ;; Array (define-bstruct <matrix4> (array 16 float)) ;; Bit fields (define-bstruct <date> (bits (year 32 s) (month 4 u) (day 5 u))) ;; Pointer (define-bstruct (<item> (struct (type int))) (<chest> (struct (opened? uint8) (item (* <item>))))) ;; Packed struct modifier (define-bstruct <enemy> (packed (struct (type uint8) (health uint32)))) ;; Endianness modifier (define-bstruct <big-float> (endian big float)) ;; Recursive type (define-bstruct <node> (struct (item int) (next (* <node>)))) ;; Mutually recursive type group (define-bstruct (<forest> (struct (children (* <tree>)))) (<tree> (struct (value int) (forest (* <forest>)) (next (* <tree>))))) ;; Opaque type (define-bstruct SDL_GPUTexture) Download Source tarball: guile-bstructs-0.1.0.tar.gz GPG signature: guile-bstructs-0.1.0.tar.gz.asc This release was signed with this GPG key. See the guile-bstructs project page for more information.
More in programming
Google told K-12 schools they could turn off AI, but only if they gave up search too. That isn’t integration. It’s holding education for ransom.
Once a year at my church, I usually give one sermon. I was asked to tie the theme to a young adult retreat we had taken to the Sunshine Coast. My first experience giving a sermon was many years ago at my old home church in El Monte, California. It was a Vietnamese-American church, but…
Practical tools for design engineers to explore possible futures through speculative design
As with my last post on AI, let me open with a statement: Just because most people hold AI a certain way, doesn’t mean that we always must, either now or in the future. As I noted in that piece, and my shades of gray post, I am well aware — and agree with — … Continue reading Must we always use LLMs this way? →
The CSS Working Group resolved to use cursor: pointer for buttons and other clickable form controls in the new appearance: base styles. That happened on October 21, but I posted about it yesterday and, uh, people have feelings about cursors. I wrote about buttons and cursors back in 2022, when a post titled "Buttons shouldn't have a hand cursor" was making the rounds. I admitted I didn't have a strong opinion either way, then argued that not using the hand cursor had become the new anti-pattern anyway. The web isn't an operating system, and people expect clickable things to show a hand. The issue is Cursors For Base Appearance in CSS Forms Level 1. The group adopted the rules from this comment, and the part that's triggering half the Web is: select, option, input[type=checkbox], input[type=radio], button, input[type=button], input[type=submit], input[type=reset], input[type=color], input[type=image], ::file-selector-button { cursor: pointer; } It seems like we have a tabs vs. spaces sort of debate. Again. Nicole Sullivan pulled cursor data from a bunch of design systems and, in September, 27 of the 33 systems that set a cursor used pointer. After she added more systems, it was closer to a 50/50 split. Elika Etemad made the usability case in the meeting. Per the minutes, she said the hand "makes it clear, as soon as you're hovering an interactive control, you're supposed to click on it," and called it a "usability benefit even if not historically done on some platforms." "But the hand means it's a link" Haz had an interesting reaction: The debate is over, and evil has won. His point is that he can't tell whether ⌘-clicking something will open it in a new window anymore. He feels more confident clicking a button when he knows it won't change the URL, like when it just opens a popover or a panel. He's not wrong about that.* I replied that I've been on team evil since 2022 (albeit without a strong opinion, just following what users seem to expect). Touch screens don't have a cursor at all, and on desktop, browsers show a link's URL in the corner when you hover it. That's some type of hint, at least. Haz pushed back that he'd have to watch the bottom-left corner with his peripheral vision, which is fair. He also said he never cared which cursor buttons got, he just wanted links to have their own. I can appreciate that. "But desktop apps don't do it!" Jacob Paris liked the standard cursor on buttons because it makes web apps feel more like desktop apps than websites. My opinion hasn't changed, though. People have already learned that a hand cursor means clickable on the Web. It's by no means mandatory, but it's familiar and satisfying, so it stuck. Either way, nothing changes today. The resolution only covers appearance: base, which browsers haven't shipped yet. Buttons and selects still use appearance: auto, and their cursor is still default. If your CSS reset, design system, or web components already set cursor: pointer on buttons, keep it. When appearance: base ships, controls that opt in will get the same cursors by default. But remember, these are only defaults! You can still add whatever cursor you want to whatever element you want. Nobody's gonna stop you. In 2022, I said I'd keep changing the cursor on buttons until there's a good reason to stop. In 2026, I still haven't found one…and neither has the CSS Working Group. *Thinking about this more, if 50% of design systems opted to add cursor: pointer to buttons already, isn't this already a moot argument? **In general, I think the Web should be more personal and expressive so who even cares about a default? This is especially true for personal and fun sites. Just don't go too crazy, especially when it affects accessibility and the overall user experience.
