Dear wanderer:

You are looking for https://support.pkware.com/pkzip/application-note-archives.

This is simply not true. The de facto standard is "the behavior of PKZIP, circa 30 years ago". Conveniently, PKZIP happens to be shareware (at least the versions of interest here). One confounding factor, as documented by Jason Summers's blog, is the difficulty of getting your hands on bonafide PKZIP originals. https://entropymine.wordpress.com/2019/07/27/will-the-real-pkz110-exe-please-stand-up/

The purported ambiguity in the implementor's notes is overstated. With exactly one exception, all of the concrete criticisms I've ever come across (including, unfortunately, some of the specific ones woodruffw echoed in the Astral blog post earlier this month) come down to implementors just flat-out doing the wrong thing (e.g. disregarding the central directory entirely), or, when presented with the choice of:

1. doing the obviously correct thing, versus

2. taking a really, really, really obtuse position that a particularly strained reading justifies an implementor's decision to do certain things (like trying to infer the position of the central directory instead of just using the field that is explicitly labeled for that purpose—bonus points if accompanied by complaints that the field is "redundant")

... they go for the latter, out of a seeming inability to say, "Whoops, how silly of me," and just fixing their damn (mental model and corresponding) implementation.

Not really. The correct behavior is, well, correct. There are just a lot of people who either succumb to taking (frankly bizarre) shortcuts or stubbornly insist on reading the spec the wrong way.

This surely means "multiple End of Central Directory records". To reject archives with multiple headers would mean accepting only archives that contain a single file.

It's here: https://news.ycombinator.com/item?id=44955771

(NB: there is no more detail in the referenced comment than what is stated here.)

If only full-stack developers actually cared about the user who is sitting in front of a Web browser (i.e. end users) as much as they cared about the "users" of their APIs who are sitting in front of a text editor (i.e., programmers).

What until you're motivated to articulate what you thought the value proposition of NPM-style build-time package fetchers actually ever was and then eventually settle into the realization of how little disk space has been saved in the whole scheme, how much bandwidth has been wasted on thousands of redundant package fetches, and the immeasurable amount of human toil and the resulting hit to the world's creative output from overlay version control systems instead of just leaving the version control to the DVCS that you're using.

Of course, this is another example where people say "JavaScript" and mean "NPM".

The last time I looked, around 8 years ago, of the multiple settings that Git offers wrt line endings, there was no combination of any of any of them that would do the most reasonable thing, and all of the advice and documentation around use of these settings were wrong. The best course of action was to use the defaults and just deal with it.

It's strange that EXIF even includes such properties—what's wrong with the image's existing dimension fields?

This section doesn't appear in all revisions of the APPNOTE. Notably, it doesn't appear in revision 1.0, nor does it appear in 6.2.0, which is the basis for the ISO standardized ZIP spec.

Huh? For the reasons above, I would be very, very surprised if this were true and Woodruff isn't just confused. This would make it incompatible with the vast majority (virtually all) ZIP-producing software. (Is there any implementation which actually works this way?)

It would take a very strained reading to end up with a misunderstanding aligning with the latter behavior (and it's very unfortunate that the "relative"/"absolute" terminology was chosen here, as where "relative" appears in the spec, it is used to mean almost the exact opposite of how it's used here—the relativity of the "relative offsets" mentioned in the spec are explicitly described as being relative to the start of the "disk" (which in almost all cases constitutes the only "disk" comprising the archive—i.e. it's the offset from the start of the file).

Here, or in general?

I do think about this a lot. This is a nice, succinct way to put it. (Critique, though: "classism" is not the best way to put it. For better or worse, "privilege" is probably one of the best words we have for this. Separately: Since "privilege" became a staple of common rhetoric, I've mused a lot about trying to convince people to minimize the focus on "privilege" (to avoid the familiar kneejerk reactions from those hearing it who have associated it with overuse), with the intent to be to sway people instead by speaking about privilege without actually using the word "privilege" and speaking exclusively in terms of affordances*.)

See: https://hypothes.is/a/TCB5zClKEeyrIOu9mp-5TA and tag:"privilege vs affordance". (NB: Hypothes.is doesn't linkify the tag in the preceding annotation correctly.)

I've tried using a hiking metaphor to describe a similar phenomenon (specifically, and perversely, as a preface when trying to explain second panel syndrome.

It's not. It assumes that if you have your own domain, you're also able to configure/patch the server to be an active participant in the Webfinger protocol—you can't do anything to add Webfinger support to your static site, for example. This is one reason that makes Webfinger a bad protocol.

I explained in a post to Mastodon somewhere that the issue lies in trying to sell, to users, the infrastructure.

For Solid to succeed, it needs to get into the hands of people, and to get into the hands of people, it needs an app that incidentally uses Solid—you're never going to convince the majority of end users to adopt Solid on its merits. They need to experience it—something that TBL should be acutely aware of, since this is what drop adoption of the WWW.

Speaking (or typing, as in the case of transcription) has a substantially negative effect on my ability to process the same amount of information as silent reading.

Later, the authors of this paper state that subjects who were "uncomfortable" with reading out loud had the option to read silently.

I don't doubt the conclusions of the paper, but I suspect that reading aloud actually has a deleterious effect, especially for those who are performing the act of reading without showing signs of having achieved the comprehension desired.

This is at the root of the infamous "Blogging vs. blog setups" comic https://rakhim.org/honestly-undefined/19/.

The fact that this is true is also the entire basis for wikis. It is reasonable to find it irksome that people, perversely, refer to Git repos full of Markdown documentation as "wikis"—which they aren't. They are fully the opposite.

You should. It is.

See also lkesteloot's Java for Everything.

Note that while this isn't true now in 2025, it wasn't exactly true in 2013 when this post was written, either. Eclipse, IntelliJ, and NetBeans are all popular IDEs that were free and (in both senses) at the time this post was written.

This is:

Guo, Phillip. 2013. “The Two Cultures of Computing.” Blog. December. http://pgbovine.net/two-cultures-of-computing.htm.

That's a strange way to put it. Surely this makes lines refer to the Select element—and the properties (such as lines.value below) are just natural, first-class properties of that element?

If not, and there's something else going on here, that's an unforgivable overloading of the "id" attribute, which is already intrinsically significant in XML for orthogonal purposes. This needs to go.

Yeah. Web authoring is really just desktop publishing—except in its current state, foiled by the JavaScript industrial complex.

Great turn of phrase, but it's weakened by the choice to shoehorn React in as a dependency.

These annotations com from the use of https://github.com/dwhly-proj/DailyJournal which sends you an email, waits for your reply, then posts it as a top-level annotation targeting the dummy URL http://my.journal and tagged with DailyJournal.

@dwhly docdrop.org is broken.

Maybe, this is true, but it's not a conclusion we can draw from these tests. In choosing programs to benchmark, he selected notorious behemoths to pit against the IMGUI set—which, it should be noted contains stuff like Dear ImGui and EGUI, which aren't even apps.

And later, in the Windows benchmarks, he shows that the two apps—the only apps—in the IMGUI set actually perform worse than the others. The fact that RAD Debugger has a heavier power draw than clunky Electron apps of all thinks like VSCode should be considered evidence conclusive in the other direction.

@58:18

I'm very anti- this whole, like, NPM thing that happened where all these web people like— not only is their program like not a mathematical object, it's like defined by things not even on their computer—out in the world that might change at some point or go away. That's the worse possible reality, and so I think you want in source control, ideally, everything that you're using.

Kay has certainly spoken about such things (albeit not to my knowledge in any way that illuminates a generalizable approach that solves the problem). I'll elide a reference to anything specific (and the prerequisite side quest to track it down), and instead include a reference to Sussman at Strange Loop: We Really Don't Know How To Compute

NB: Unpleasant restrictions consist of things like, "You cannot read files from the user's local disk outside the file(s)/folder(s) that the user has given the software permission to access, and the person who authored it can't write to arbitrary locations on the user's disk"—both of which are imminently reasonable (and relate to things which you would hope would go one to be addressed (and the quicker, the better) if it were the case that the environment didn't already provide protection against these sorts of things).

Relatedly, Brad Cox quibbles about the inappropriate application of the word "maintenance" to these situations.

See https://hypothes.is/a/jH9HWDgREfCabIfIFCE5Ow

Third option: acknowledge that you chose the wrong foundations to build upon, and just because the present structures have proven to be (what I'd have argued was a predictable) failure, that that doesn't mean that there is no foundation upon which these things can be built while delivering the sort of stability desired.

Assuming the "common platform" is something reasonable (i.e. depends only a runtime that can be expected to be present on all machines) this is as straightforward as zip -r ./package.zip research/.

(The problem isn't figuring out how to do it. It's getting people to stop sleepwalking along with all the "best practices" that are outright inimical to the reproducibility/replicability goals. Almost everyone—including to an extent the author of this post—is unwilling to cast aside their attachments.)

Even before class declaration statements appeared in ES6, this was the wrong way to define methods.

Sure it does.

Yeah. It's a major social problem. Not so much a technical one.

Imagine if there were a readily accessible World Wide Wruntime that anyone with any commodity computing device could use to run arbitrary code (and its corresponding documentation) written by other people...

Proposal for a much more interesting piece of writing: "The Sacrifices We Choose Not To Make"; cf:

1. The Handwriting on the Wall
2. New Christianity

Terrible pedagogy.

It's trivial for someone who has only a single device to get their hands on a Linux image and run it in a lightweight VM if they're coming from Mac or Windows. The preceding sentence doesn't hold true for any other permutation of { Linux, Mac, Windows }.

Yeesh.

We have a word "pathologize" for this, but the author doesn't use it anywhere in this piece. The general thrust behind the word's existence was an implicit understanding that to pathologize X is generally the wrong approach to trying to explain or address X. Because it doesn't really explain anything—you just sort of bottom out (or hit a wall, pick your metaphor); it operates on the same principle behind the degenerate behavior that Feynman observed and criticized about how people think that knowing the name for something is a substitute for understanding that thing, or, when asked to explain magnets, he asked the questioner what they thought they were asking and explained that there's really not an answer (of the sort they thought they wanted) for anything else that they think they understand better than magnets but didn't ask about.

It's notable that the author left out the canonical implementation (PKZIP).

I'd also expect them to have at least tried to test the ZIP support in Mac OS Finder and the Windows shell.

Not true—they're not redundant; it is not correct to assume that this relation holds.

The only correct way to compute the offset is to use the field with information about the offset. It is not correct to do otherwise.

The URL for this page (https://www.beet.tv/2009/10/webs-inventor-sir-tim-bernerslee-double-backslashes-were-unnecessary.html) betrays the author's (editor's?) inability to grasp what TBL says even while they are the ones trying to piggyback off his comments for attention.

[…] People on the radio are calling it "backslash backslash"[…]

Apart from the formatting characters, this should be easy to do—the simplest OCR to ever implement.

Interesting outcome. Rarely (basically never) do you see "it's more verbose to write in $LANG2 what would be written in fewer lines if you were using $LANG1" touted as a positive.

missing comma here

Ibid.

"Web address" is so awkward as a term. Better to just make it concrete, anyway, by providing a familiar example like "The URLs that appear in the address bar in a Web browser […]".

The syntax highlighting here is very strange. It seems to follow the rule that the first atom inside a set of angle brackets is bold, and the subsequent atoms (if any) are not.

There's probably a better way to put this.

The natural way to phrase this would be "the subject of one triple and the object of two". (There should probably be a "both" thrown in there, too, to make the sentence structure even more predictable.)

This is an underrated (actually undervalued) affordance in any codebase.

Do they emit the same code?

* 1.5504

What are overlay SCMs supposed to actually achieve again?

This is a better, but more verbose way to state what I summarized as "evenly distributing yesterday's future".

This belies the inevitable would-be justifications for the technology choices behind these projects.

Why should the app even know let alone care about which version of the serialization format is in play?

I'm confused about why this namechecks a proprietary clone of Greasemonkey instead of just, you know, mentioning Greasemonkey.

This is a superfluous clarification.

(Maybe if it appeared nearer to the earlier discussion of situated software?)

"Malleable software" implies the opposite, right on its face.

Is "creator" the right word here? There's lots of software that falls outside of what is the subject of this paper that enables creators. Indeed, it has been a common refrain in criticisms of the FOSS movement that for the types of software that creatives need and/or simply desire to use the proprietary apps tend to have no equals.

See situated software.

At this point, when you're writing things like this, it's worth re-examining what in the hell you've got all this stuff underneath that you don't even understand and you're certainly not using correctly.

"It basic reviver with multiple features"? That makes no sense.

Great term of art for e.g. scraping.

Inheritance seems to be the odd item out in this list—inheritance is not essential for OO.

It's strange that there is a ready metaphor that isn't used—

The practice not of manufacturing parts (which is readily automated), but the labor that goes into designing the machines that enable the automation but for which there is no known way to automate the design work itself.

As I mentioned in another annotation, Cox's article in American Programmer was actually published with the title '"No Silver Bullet" Reconsidered'.

Hsu again cites only Mahoney for this, and the passage here is presented as one quote, but it's actually a quote within a quote: first Bemer and then Mahoney. The original Bemer quote ends with the second sentence ("I contend that the production costs are affected far more adversely by the absence of such an environment than by the absence of any tools in the environment…" which ends prematurely here but ends with a parenthetical "e.g. writing a program in PL/1 is using a tool"), and the remainder is Mahoney's commentary.

The Bemer source is:

R.W. Bemer, "Position Paper for Panel Discussion [on] the Economics of Program Production", Information Processing 68, North-Holland Publishing Company, 1969, vol. II, p. 1626.

This is such a strange thing to see in a paper from 2009.

The source cited as support is a book from 1995.

Strong Python vibes.

Specifically, it's:

Gabriel, Richard P. “LISP: Good News, Bad News, How to Win Big.” AI Expert 6, no. 6 (1991): 30–39.

Wikipedia says it was written in 1989. Gabriel says it was published in 1991. Presumably the latter refers narrowly to the actually publication when it was cut down to 9 pages for AI Expert in June 1991.

Ted is referring to the "The Rise of Worse Is Better" section that appears in the version published on Gabriel's site but that was simply called "Worse is better" in the version that AI Expert published.

The header image here doesn't load. But I made sure that it was archived. If and when Medium stops syndicating this article, you can find a copy of the image that was used here:

https://web.archive.org/web/20250614131832/https://miro.medium.com/v2/resize:fit:640/format:webp/1*WmQCEd5vo4ijVXrThaFSlA.png

See this note for the same essay, only qualified by Guo's UCSD subdomain instead of his personal one:

https://hypothes.is/a/22JaWMu5Eey2UvchosEz6Q

This is:

Gabriel, Richard P. “LISP: Good News, Bad News, How to Win Big.” AI Expert 6, no. 6 (1991): 30–39.

... and a copy (in HTML) can be found at https://www.dreamsongs.com/WIB.html

Note that the actual title as it appeared in the American Programmer Journal is «"No Silver Bullet" Reconsidered».

Irony (cf https://hypothes.is/a/BeEXTkk3EfCERutnCr7MPw)

No.

From Ted Nelson's "Computer Lib Pledge":

• The purpose of computers is human freedom.
• I am going to help make people free through computers.
• I will not help the computer priesthood confuse and bully the public.
• I will endeavor to explain patiently what computer systems really do.
• I will not give misleading answers to get people off my back, like "Because that's the way computers work" instead of "Because that's the way I designed it."
• I will stand firm against the forces of evil.
• I will speak up against computer systems that are oppressive, insulting, or unkind, and do the best I can to improve or replace them, if I cannot prevent them from being bought or created in the first place.
• I will fight injustice, complication, and any company that makes things difficult on purpose.
• I will do all I can to further human understanding, especially through the new visualizing tools of interactive computer graphics.
• I will do what I can to make systems easy to understand, interactive wherever possible, and fun for the user.

Guo's addendum (mentioned in an earlier annotation) addresses this: so many readers seem to have come away from Guo's piece with weird assumptions that he's not in the process of trying to teach his student any of this stuff and that he is instead just trying to set things up for them. That's not what he's doing, and that's not what upsets him.

What of the lack of perseverance of "the others"—to contribute their work in a way that can be integrated? Wouldn't it be a lot more productive if everyone aimed their perseverance at dealing with the problems that fall out of their own contributions—work that they are intimately familiar with—rather than trying to grapple with problems in N of the M components that they are building on and don't have the same level of familiarity—and for this effort to be duplicated by everyone else trying to build on top of it, too?

Indeed—which should really cast the norms around scientific computing in stark light as being clearly the wrong way to go about doing things. (Which itself isn't to say that there's anything special about scientific computing here—there are plenty of programmers (working on open source or otherwise) that get things just as wrong. Most of them, even. It's virtually everyone.)

This doesn't nail it either. There are lots of people who are software developers who have (or at least have been told or have convinced themselves) that they have been trained in the best possible procedures. Adar obliquely acknowledges "open-source developers", but in a strange way that seemingly implies software endeavors that aren't FOSS are better "incentivized to keep documentation up to date or the software running".

My initial reaction was to the "This is inevitable" part that follows, before backing up and focusing on this sentence, to which my response was originally going to be, "Is it? No, it isn't." That is, until I skimmed Guo's addendum/update (which I'd never seen before) in one of the two copies I kept of the piece which this piece by Adar is a response to. And indeed, it seemed, based on a surface-level reading of that addendum that Guo largely agreed. But then a close reread of that proves that there isn't any explicit or implicit agreement on Guo's part about the diagnosis offered here. And I don't think the diagnosis is correct.

And now to focus on the next part:

This is inevitable. It is inevitable because

I don't think it is. (If nothing else, it hasn't been proven.) Even from the perspective of the author, this falls squarely in the "accidental complexity" versus "essential complexity" bucket.

(And to pick up a thread of mine from earlier this year, I wonder if we have perhaps made a mistake in focusing too much on the notional of incidental (or "accidental") versus inherent (or "essential") complexity, and whether or not we should be trying to address the fact of incidental versus intentional complexity—which surely exists. And that isn't to say that's the culprit everywhere, but to reiterate: it surely exists and surely accounts for some of what's going on.)

I don't see how the examples offered are helpful in conveying the distinction. The instruction to "type git clone" seems procedural.

Based only on my inference of what is actually intended by "factual" versus "procedural" (without first checking the link here), I'd guess that, "The command for copying a Git repository's contents to the local machine is git clone" would be a real example of a factual (versus a procedural) lesson style.

Too meta, too messy, too personal, and too much detail.

(NB: this actually comes from export of my annotation of Eytan Adar's piece "On the Value of Command-Line “Bullshittery”" originally published on Medium. See https://hypothes.is/a/zmhLwkkqEfCyyo_A6dOGrw for the original annotation. I used the aforementioned export to annotate the annotation—i.e., to add a true, first-class (though perhaps second-rate) annotation, rather than a reply.)

A major purpose of this style is Malleable Software, or software that is more supportive of change, reusability, and enhancement.

Is this the first occurrence of this term?

samsquire is https://www.halfbakery.com/user/chronological.

Does this mean arguments against verbose "boilerplate" languages are going to be given less credence?

So why not just stop doing that? None of the churn was necessary.

Pfft. Lame.

The earlier breakdown involved being dependent on the bus for transportation into the city and relying on the library for entertainment. Is the bus subsidized? Is the library?

Before I got to the end of this sentence I was expecting to see something like $70 per person.

$300 per month on food is "laughably low"? You can eat cheaper in the city!

Quintessential boomer shit.

Interesting problem decomposition with shades of Engelbart re https://dougengelbart.org/pubs/papers/scanned-original/1960-augment-133181-Special-Considerations-Individual-re-Information.pdf.

Hey, traveler. You're looking for https://dl.acm.org/doi/pdf/10.1145/948093.948095.

For example, the syntactic analysis stage builds a conventional expression tree, but this tree is expressed as objects—instances of various classes relevant to compilation: Class, Method, Message, Selector, Argument, etc. […] The code for these operations is encapsulated inside each object, and is not spread throughout the system so changes are easily made. And all these specialized classes rely heavily on reusable code inherited from more primitive classes like Object, HashTable, Symbol, Token, and LinkedList.

This is an interesting claim, given that one of the common criticisms of the classic, inheritance-heavy style of classic OOP is precisely that it leads to code that is indeed "spread throughout". Adele Goldberg commenting on the legacy of Smalltalk made a well-known quip, "In Smalltalk, everything happens somewhere else."

It's notable that although there many common word sequences between this paper and the later, similarly titled one that appeared in IEEE Software (1984), this passage doesn't appear to be present (based on a quick skim—I could be missing something), and the closest corresponding statements are significantly reworded.

This is:

Cox, Brad J. “The Message/Object Programming Model.” In Proceedings of Softfair: A Conference on Software Development Tools, Techniques, and Alternatives, 51–60. Arlington, VA: IEEE Computer Society Press, 1983.

In Hints and Principles for Computer System Design, 2020, Lampson describes his original 1983 paper (https://dl.acm.org/doi/10.1145/800217.806614) as being "[r]eprinted with some changes in IEEE Software".

This is:

Lampson, B.W. “Hints for Computer System Design.” IEEE Software 1, no. 1 (January 1984): 11–28. https://doi.org/10.1109/MS.1984.233391.

This presupposes that you'd settle on opaque sequential blobs of said instructions as the preferred archive format—a consequence of the cultural milieu of the writer, and not the product of careful thinking and logical decisionmaking.

The modern incarnation of the fallacy of bulk-is-better is judging the quality of a project (esp. a component) by being overly concerned with whether it is more or less "active" than some other project.

The question posed here is an obvious response for when you encounter instances of the bulk-is-better mindset (although it's not as easy to automate as Cox suggests here).

I've often mused about the thought of pulling a Reddit* and automating some level of churn within a codebase for a project that's hosted on e.g. GitHub for the sole purpose of making sure that it appears "active".

* A story about the early days of Reddit has become well-known after the creators volunteered some information about the early days when they made use of a bunch of "fake" accounts to submit links that the creators had aggregated/curated for the purpose of seeding the site.

Are they even well-formed hypotheses? That is, are they falsifiable?

It's strange that this basic truth is something that has to be taught/explained. It indicates a failure in analytical thinking, esp. regarding those parts of the question which the quoted passage is supposed to be a response to.

In other words: actually practicing version control.

I'm baffled how much mindshare Matrix was evidently able to gain (and maintain) given how crummy it is as a product and frustrating to use.

This is:

Cox, Brad J. “Message/Object Programming: An Evolutionary Change in Programming Technology.” IEEE Software 1, no. 1 (January 1984): 50–61. https://doi.org/10.1109/MS.1984.233398.

The actual name of the paper referenced here appears to be "Prospects for an Engineering Discipline of Software". "Towards an Engineering Discipline of Software" is the last chapter/section.

Consider a derivative: Slim Source Code—

… in the context of the problem described in "The Cost of Selective Recompilation and Environment Processing" (Adams, Rolf, Walter Tichy, and Annette Weinert. ACM Transactions on Software Engineering and Methodology 3, no. 1 (January 2, 1994): 3–28. https://doi.org/10.1145/174634.174637.)

This is:

Franz, Michael, and Thomas Kistler. “Slim Binaries.” Communications of the ACM 40, no. 12 (December 1, 1997): 87–94. https://doi.org/10.1145/265563.265576.

Infuriatingly, obnoxiously dismissive.

Claims like these need supporting evidence. Are they really "weird limitations"? Or are they actually reasonable limitations—on developers' unending attempt to exercise what they feel as their manifest destiny to control the user's device and experience?

So? To reiterate:

Fuck "developers".

This is a terrible example.

That is 100% a failing of the folks making those decisions at Wikipedia, not HTML.

This flavor of fatigue (cf "JavaScript fatigue") is neither particular to nor an intrinsic consequence of static site generators. It's orthogonal and a product of the milieu.

It can also create pressure to hold your tongue on things that you might otherwise be comfortable expressing were it not for the fact that there was available a meticulously indexed, reverse chronological stream of every sequence of words that you've authored and made ambiently available to the world.

Mmm… half of what's described here is an episode of Black Mirror…

The screenshot here (p. 33) reads:

Introduction

The intent of this paper is to describe a number of studies that we have conducted on the use of a personal computing system, one of which was located in a public junior high school in Palo Alto, California. In general, the purpose of these studies has been to test out our ideas on the design of a computer-based environment for storing, viewing, and changing information in the form of text, pictures, and sound. In particular, we have been looking at the basic information-related methods that should be immediately available to the user of such a system, and at the form of the user interface for accessing these methods. There will undoubtedly be users of a personal computer who will want to do something that can be, but not already done; so we have also been studying methods that allow the user to build his own tools, that is, to write computer programs.

We have adopted the viewpoint that each user is a learner and have approached our user studies with an emphasis on providing introductory materials for users at different levels of expertise. Our initial focus has been on educational applications in which young children are taught how to program and to use personalizable tools for developing ideas in art and music, as well as in science and mathematics.

It is titled "Smalltalk in the Classroom" and attributed to "Adele Goldberg and Alan Kay, Xerox Palo Alto Research Center, Learning Research Group". Searching around indicates that this title was re-used in SSL-77-2, but neither introduction in that report matches the text shown here.

Searching around for distinct phrases doesn't turn anything up. Is this a lost paper? Is there some Smalltalk image around from which this text can be extracted?

Is that actually true? Surely there are examples where people play the game that they're less suited for—where the decision is driven by desire?

I don't think that's right—

My life experiences during the (now bygone) Shirky era and the loose school associated with it really inculcated (in me, at least) the value of small, cumulative improvements contributed by folks on a wide scale. See something wrong? Fix it. Can't fix it (like, say, because you don't have the appropriate authorization)? File a bug so the people who can fix it know that it should be fixed. This matches exactly the description of seeing the "little flaws in everything you see in the world, and focus[ing] on those".

Looking at those flaws and thinking "this thing isn't as good as it could be" is a necessary first step for optimism. That belief and the belief in the possibility of getting it fixed is the optimist approach.

When I think of miserable people (and the ones who make me miserable), it's the ones who take the attitude of resignation that everything is shit and you shouldn't bother trying to change it because of how futile it is.

The at-oddsness of the two things mentioned here—spewing bullshit and private beliefs that someone could correct you about—is hard to skip over.

This is my experience working with other people, generally. Another way to put it besides people not "priortizing the truth" is that they just don't care—about many things, the truth being one of them.

I don't think this is a good definition of liar. The bullshitter described here would be a liar if what they're saying is lie.

Check out https://tinlizzie.org/VPRIPapers/m2004001_power.pdf

Are we sure?

This is:

Bobrow, Daniel G, and Bertram Raphael. “New Programming Languages for AI Research.” Palo Alto, California, August 20, 1973.

An early mention of Smalltalk.

That's interesting. This paper is Copyright 1977 (dated June of that year) and is using the term "message-oriented" rather than "object-oriented".

Rochus maintains that Liskov and Jones are the originators of the term "object-oriented [programming] language".

Here's Goldberg in an early paper (that nonetheless postdates the 1976 paper by Liskov and Jones) writing at length about objects but calling Smalltalk "message-oriented" (in line with what Kay later said OO was really about).

We're at the point in humanity's development of computing infrastructure where the source code for program texts (e.g. a module definition) should be rich text and not just ASCII/UTF-8.

Forget that I said "rich text" for a moment and pretend that I was just narrowly talking about the inclusion of, say, graphical diagrams in source code comments.

Could we do this today? Answer: yes.

"Sure, you could define a format, but what should it look like? You're going to have to deal with lots of competing proposals for how to actually encode those documents, right?" Answer: no, not really. We have a ubiquitous, widely supported format that is capable of encoding this and more: HTML.

Now consider what else we could do with that power. Consider a TypeScript alternative that works not by inserting inline type annotations into the program text, but instead by encoding the type of a given identifier via the HTML class attribute.

Now consider program parametrization where a module includes multiple options for the way that you use it, and you configure it as the programmer by opening up the module definition in your program editor, gesturing at the thing it is that you want to concretely specify, selecting one of those options, and have the program text for the module react accordingly—without erasing or severing the mechanism for configuration, so if another programmer wants to change the module parameters to satisfy some future need—or lift that module from your source tree and use it in another one for a completely different program—then they can reconfigure it with the same mechanism that you used.

I feel no strong urge to disagree, but on one matter I do want to raise a question: Is "accidental" even the right term? (To contrast against "essential complexity", that is.)

A substantial portion of the non-essential complexity in modern software development (granted: something that Brooks definitely wasn't and couldn't have been thinking about when he first came up with his turn of phrase in the 1980s) doesn't seem to be "accidental". It very much seems to be intentional.

So should we therefore* highlight the contrast between "incidental" vs "intentional" complexity?

* i.e. would it better serve us if we did?

In hindsight, this would be a better outcome than what we ended up with instead: very large software systems that, relative to their size—esp. when contrasted with the size of their (often more capable and featureful) forebears—accomplish very little, but demand the resources of extremely powerful computers.

Considering this post is about Swift, describes Lattner's departure, etc., it would have been opportune to mention Graydon Hoare's departure re Mozilla/Rust and his subsequent work at Apple on Swift.

"Pre-RFC" doesn't sound like there's "not going to be an RFC" for it. It rather sounds like the opposite.

I haven't been using LLMs much—my ChatGPT history is scant and filled with the evidence of months going by between the experiments I perform—but I recently had an excellent experience with ChatGPT helping to work out a response that I was incredibly pleased with.

It went like this: one commenter, let's label them Person A posted some opinion, and then I, Person B, followed up with a concurrence. Person C then appeared and either totally misunderstood the entire premise and logical throughline in a way that left me at a loss for how to respond, or was intentionally subverting the fundamental, basic (though not unwritten) rules of conversation. This sort of thing happens every so often, and it breaks my brain, so I sought help from ChatGPT.

First I prompted ChatGPT to tell me what was wrong with Person C's comment. Without my saying so, it discerned exactly what the issue was, and described it correctly: "[…] the issue lies in missing the point of the second commenter's critique […]". I was impressed but still felt like pushing for more detail; the specific issue I was looking for ChatGPT to help explain was how Person C was breaking the conversation (and my brain) by violating Grice's maxims.

It didn't end up getting there on its own within the first few volleys, even with me pressing for more, so I eventually just outright said that "The third comment violates Grice's maxims". ChatGPT responded with the level of sycophantism (and emoji-punctuated bulleted statements) a little too high, but also went on to prove itself capable of being a useful tool in assisting with crafting a response about the matter at hand that I would not have been able to produce on my own and that came across as a lot more substantive than the prompts alone.

Then don't use the word "upload".

This URL is reference from https://ignore.pl/2025/04/a_pair_of_examples_of_simple_interfaces.html

Traveller:

You are looking for https://ignore.pl/2022/01/playing_around_with_simple_interfaces.html.

This is a perfect use for a mostly-single-file browser-based literate program.

Or—get this—it was because despite the fact that he "hasn't been for several years", he used to "ski a lot", and it was the start of ski season.

You don't have to assume any sophisticated conspiracy of ad companies listening in through your device's microphone or location-awareness and user correlation. This is an outcome that could be effected by even the dumbest targeted advertising endeavor with shoddy not-even-up-to-date user data. (Indeed, those would be more likely to produce this outcome.)

Fucking stupid.

It is a tragicomic fact that our proper upbringing has become an ally of the secret police.

This is:

Engelbart, Douglas C. “Special Considerations of the Individual as User, Generator, and Retriever of Information.” American Documentation 12, no. 2 (1961): 121–25. https://dougengelbart.org/pubs/papers/scanned-original/1960-augment-133181-Special-Considerations-Individual-re-Information.pdf

If only there were some sort of highly accessible information system that was designed to make resources available from anywhere in the world without any higher requirement besides relatively simple and ubiquitous client software. Developers might then not be compelled to churn out bespoke programs that effectively lock up the data (and prevent it from being referenced, viewed, and kept within arm's reach of the people that are its intended consumers).