First: It would be a big help for this discussion, if we could have the informal convention that people who were employed in an IT job before 1990 marked their post. I think it would show a quite clear divergence of attitude.

Second: It's very obivious, that a lot of you have never been anywhere near the kind of software project posited in the "cathedral" meme, instead you project into the word whatever you have heard or feel or particularly hate. That's not very helpful, given that there is an entire book defining the concept (I belive it's available online on ESR's homepage, how about you read it ?)

Third: No, I'm not of the "either you are with us, or you are against us" persuation. The bazaar is here to stay, but having everybody in need of transportation buy the necessary spareparts to build a car is insane.

Fourth: Related to point two really: A lot of you seem to have little actual ambition of making things better, I guess that is what happens if you grow up in a bazaar and never even experience a cathedral. I pity you.

Second: I worked on OS X. In fact, I worked on Snow Leopard, and the start of Lion. What's interesting about that, is that Snow Leopard was the last version of OS X developed according to the "Cathedral" model. Also, while the Snow Leopard cathedral was being built, iOS was being developed firmly using the bazaar model...

Third: You know, I wonder if you've ever been to a bazaar before? I live in Turkey, where the bazaar is a way of life (and where one of the largest, oldest bazaars in the world is located). I've never found "spare parts" at a bazaar. What I have found is some of the highest quality jewelry, tapestries, rugs, and other hand-made goods you'll find anywhere.

Fourth: I think you're conflating "good quality"=>Cathedral and "poor quality"=>Bazaar. The only thing that distinguishes the Cathedral and the Bazaar is whether or not there is one single individual in whose head the only valid vision of the completed project exists. You might do well to read up a bit on the history of Kapalıcarşı. Throughout its history there were guilds to enforce authenticity and all manner of quality control mechanisms. It is possible to have a Bazaar and a very high quality product.

> The only thing that distinguishes the Cathedral and the Bazaar is whether or not there is one single individual in whose head the only valid vision of the completed project exists.

Core OS and related groups has long consisted of VERY different teams each working in their own fiefdoms with their own methodologies. There was a somewhat coherent vision overall, but everybody achieved it in their own way, with very mixed results -- OS X has some tremendously terrible code. Security framework, Installer team, mDNS, anything related to Server ...

To be honest, I'm not even sure what your point actually is.

The point is that, if one of the workers building the Notre Dame had gotten hot after a day of work and said "You know what we should include here? A swimming pool!" well...he probably would've been committed.

Coherent vision and a priori design are not the same thing. Apple has coherent vision. Bazaars can have coherent vision (should have coherent vision if they hope to be successful). But that's not the same thing as a Cathedral's a priori design...

A priori design can adapt to new ideas. I'd argue that's what Apple did/does, in many cases.

Likewise, I'd argue that the truly ad-hoc bazaar development is responsible for some of the worst ideas and bad code that can be found at Apple.

The historical lack of good centralized vision on aspects of the Core OS -- such as Objective-C -- has led to staggering missteps and ridiculous inefficiencies on behalf of both the framework and compiler teams. This has been to nobody's benefit and the sum result is clearly inferior to better-designed language work done elsewhere (eg, MS).

Likewise, the ability for applications teams to drive forward ill-conceived OS and framework hacks has led to some terrible long-lasting implementation failures, which is something a coherent top-down vision could have prevented.

However, the fact is that products can succeed despite of their poor implementation. Costs may be higher, bug counts may be higher, and user satisfaction may be lower, but that hasn't always stopped Apple from building successful products. Where I take umbrage is in the notion that there's a dichotomy -- either you do things poorly and let intellectual lazy engineers take the lead, or your product does not succeed. That's not accurate.

I don't think Apple is a good case study for your point.

Compare FreeBSD kernel design versus Linux.

kqueue vs. dnotify/inotify/???

Mach-descended VM vs. a string of linux-vms

BSD scheduler, ULE scheduler vs. how many different schedulers?

Linux churns through ill-conceived solutions to problems until they find one acceptable enough. FreeBSD grinds on one until it definitively works.

FreeBSD almost invariably winds up with the better solution, in less time. See kqueue, for example -- the foundation upon which Apple's GCD is built.

Planning ahead can lead to great things. The Notre Dame, and the dozens of other cathedrals throughout Europe are positively stunning...

I love and I hate kqueue. I mean, I love kqueue. I love the way it can be used from C, I love the way it's integrated in MacRuby...I spent the weekend studying Clojure's reducers and was dying to have some time to work on ClojureC just so I could implement reducers with kqueue...

I hate kqueue because, in all likelihood, I'll never get to use it in a production system, because it's not in Linux.

Cathedrals can be nice to look at. Bazaars are often more functional.

Except that FreeBSD is functional in production, so what is the actual problem? That market effects and accidents of history resulted in Linux becoming more widely adopted?

What does that argue for, exactly?

http://en.wikipedia.org/wiki/USL_v._BSDi was hugely crippling at a very critical moment.

In a similar vein, PostgreSQL lost out to MySQL in no small part due to:

- PHP supported MySQL out of the box.

- MySQL was slightly easier to get running.

If there's a lesson we ought to learn, it's that market success may be partially or fully disassociated from actual merit relative to other market entrants. We've seen this in commercial software. It would be foolish to think it doesn't apply to open-source.

And what happens if it never works?

libkse was an attempt to implement M:N thread scheduling. This is something that can, in theory, provide significant benefits over 1:1 thread scheduling, but in practice, is extremely complicated to implement and has fallen out of favor across the board: Solaris abandoned the approach in Solaris 9, and FreeBSD abandoned it in (IIRC) FreeBSD 7.

Concurrent to libkse, libthr was developed by David Xu, and was also included in the FreeBSD base system. It implemented 1:1 threading, is far simpler than libkse, and has replaced libkse as the default threading library.

I would argue that in this case, FreeBSD's cathedral model failed; M:N threading ideally would have never been attempted, and it was wasteful to attempt to implement two distinct threading libraries. However, considerable thought and expertise went into the work, and the work (and decisions around it) were not made flippantly or taken lightly.

It was simple a case where despite best intentions and best effort, the wrong choice was made. At the same time, in what some might call "bazaar-like", David Xu maintained libthr as an alternative. It remained there, ready for adoption as the default threading library, until such time as it became apparent that M:N via libkse was a dead-end.

This was a mistake, but no entity is perfect, and the success rate of FreeBSD's considered decision-making remains statistically high. In this case where something never worked, it was replaced with something equally well-considered and far more successful.

Moreover, compared to Linux's gross missteps with their threading implementations (including such gems as setuid() being thread-local as an implementation side-effect), libkse was a minor fender-bender.

There seems to be some underlying assumptions of omniperfection hung of the "cathedral" meme in these parts. That is simply unfounded in both theory and practice.

Second, cathedral vs. bazaar is not really about governance, but about architecture, and they are two separate power-structures, although most organizations, including FreeBSD, mingle them, to their own disadvantage.

We can argue if attempting M:N was a wise or a factually based decision, but that has nothing to do with cathedral vs. bazaar, certainly not in this particular case: What happened was that M:N came first and that was that.

Only once it clearly transpired that it could not sustain its promises in practice (partly because every thread programmer assumed 1:1) did the actual decision making aparatus of FreeBSD kick into gear.

In Linux, (which is much more cathedral than FreeBSD) or OpenBSD (at the time even more so), that decision would probably have been executed practically instantly, but in FreeBSD which is mostly consensus driven, it took some time (and angry words etc.)

But overall the libkse vs. libthr saga has almost nothing to do with cathedral vs. bazaar, because architecture was not the driving force at any time during that saga.

Indeed, throughout this discussion it has become clear that you regard anything with a sense of design and some amount of quality control as a "Cathedral". If you want to discuss the benefits of design and quality control, that's a perfectly fine discussion to have, but it also seems to be completely orthogonal to the original discussion of cathedrals and bazaars, which was wholly focused on the development process.

In fact, I wonder if your arguments would be better focused on the concept of "craftsmanship" (or lack thereof) in programming...keeping in mind, of course, that both cathedral builders and bazaar artisans have historically had a notion of craftsmanship.

I think it is plausible to describe the kernel as much more Cathedral-like, at least in parts where Linus is very strict about how components should be integrated. I can see where parent could have got this impression.

At the same time, there are parts of kernel that are bazaars. Virtualization/Hypervisor support is the area I'm most familiar with; the various hypervisor vendors basically crammed their paravirt calls and passthrough drivers in, without unifying the interface at all. Yes, Linus and others were very aggressive with maintaining the code quality, but less with the architecture. There are other areas of kernel (particular drivers), where we've seen similar situations.

That said, given your response, I think perhaps still don't understand the meaning behind your cathedral/bazaar dichotomy.

While it's been years since I mucked about in Linux kernel space, I always remember it as being a very open and egalitarian community. Although I never made the attempt to contribute to any of the BSDs, I got the impression that they were much less open to newer contributors and divergences from "accepted wisdom" (this could be put more politely as being more discriminating (as in taste) and careful). That being said, the majority of decisions I was witness to on LKML appeared to be well-considered, and make no mistake that patches and design ideas were not just accepted willy-nilly. While Linux has stumbled across many missteps, I still personally prefer the wild possibilities and endless choices that eventually shake out a solution, at the same time keeping with a distribution (Debian stable) that filters out the issues for the end users (myself included).

I also find very interesting the parallels between this discussion and the "worse is better" essay, where UNIX was supposed to be in the "worse" camp, and now it's Linux.

It definitely is, in my experience. Most importantly, despite the obvious presence of "celebrities," even completely unknown people can jump into the middle of a conversation on an important topic, and if they have their shit together, will be accorded pretty much instant respect and be treated as an equal. There's very little sense of "needing to pay one's dues."

But you definitely have to have your shit together. If you don't, you will be quickly eviscerated.

It's really rather nice, especially compared to many dev communities where there's often much more sense of entrenched factions.

I was hoping someone would bring up OS X and how horrible its cathedral-born API became around 10.5, but I was born in '85 and therefore have no right to speak in this thread.

This article concludes that quality only happens if someone takes responsibility for it. Yes. You'll find no argument on that point from me. But cathedral's are not (edit: should say "not only") about one person taking responsibility for quality. They are about a priori design.

Quality control can happen ex post facto but creativity, once strangled, dies.

It seems to me that quality is about having someone on the top who is willing to give direction and vision to a project, instead of having every person who scratches their own itch full influence. Contributors are great. But letting everyone pull in their own direction doesn't lead to something that feels well engineered.

PHK seems to be calling this vision and cohesive design (a-priori or not) the "cathedral model". I agree strongly that for a project to work well and feel "high quality", it needs some source of a unified idiom.

If the suggestion of the article is that the only way to retain quality is to move back to the "single vision" world of a priori design, I'm sorry...that ship has sailed, the cat is out of the bag...whatever your favorite analogy, PHK is very right that the new generation has gotten used to not having instructions handed to them.

Of course, anyone is free to start a project with a single vision, recruit new members, and do their best to grow the project. I suspect, however, that such an effort would loose out to one that figures out how to develop a coherent vision without the need for a priori design.

Nobody belives in "a priori design" and I somehow doubt that anybody did. Brooks points out that the original publication of the "waterfall model" was meant as "how not to..." and people got that wrong.

But cathedrals are not about a priori design, they are about style, elegance, economy of means and coherency of design.

But my point in the piece is that the lost generation doesn't even know what a cathedral is in the first place, having grown up in the bazaar.

ESRs original prototype for the "Bazaar" was Linux. Do you feel that Linux is lacking in coherency of design? You refer to the dot-com bubble, but the problem with the bubble wasn't, I think, that it was the "Bazaar". Indeed, I don't think the dot-com bubble of the late 90s was characterized by much open source development at all!

It was, rather, consumed with "flashiness" and "wow factor". I definitely see the continued obsession with these things as a problem. I would say that, for example, much of the obsession with Node.js today is a consequence of this obsession. But that isn't a Bazaar.

It's a disco.

This is true. The OP has made several somewhat-arbitrary and vague assertions about what may or may not constitute a "Cathedral", on top of lambasting others for not agreeing with his definition ("Read Brooks Book")

Excerpts from the HN discussion :

"Windows and Office are actually not examples of cathedrals, because the architectural focus of Microsoft was not on software but on a near-monopoly market." -phkamp

"...iOS is very much a Cathedral and has a designer and architect who cares and who is in control." -phkamp

Does Apple not care about achieving market dominance? Did no version of Windows or Office have a lead architect who "cared"?

Frankly, the entire concept of a "Cathedral" - rigidly defined as it appears to be here - is a little nebulous and hand-wavy.

You're calling a ot of people out on using Microsoft as a cathedral example, so I assume you want the original bazaar-vs-cathedral reading of the paper. But if you do that, how can autoconf be a bazaar example, when it's GNU software?

With more open development communities there are tradeoffs, but often the result is that you can do stuff you simply wouldn't have the manpower to do in closed communities.

I think most of us would love it if every library and tool we used was pretty and elegant, but in the end, it's often better to have something, however much you'd ideally wish for it to be prettier.

Morever, a community in which different components are often the result of disparate teams with different thinking probably results in more attention paid to robust and simple interfaces between them, simply because that's the only way you can get everything to work. Systems composed of heterogeneous components with robust and simple interfaces between them, are, I think, a good thing. I suspect they tend to be more future proof than systems with everything designed by one person from the top down, because in the end, change is inevitable, so having to deal with heterogeneity from the beginning is an advantage.

It is not just that M4 is a horribly ugly, badly, inconsistent language. You also have about a million leaky caches in the build process to deal with. Autoconf rewrites the build scripts in three stages, each which is cached and also keeps an autom4te.cache directory around. So what happens is that you update something in the M4 build script and then have to spend hours hunting down why the change didn't change anything because some stupid autoconf cache wasn't flushed.

What autoconf has done is to separate free software developers into two categories. Those who can hack the source (the .c and .h files) and those who understand the "autoconf magic." It should be telling that people refer to it as "magic" - it's a fing build system, not rocket science.

The only reason this ugly piece of software has lived on is because people treat it as some kind of magical device and inertia. Configuring and building software is an easy problem. It gets even easier if you finally decide to drop support for 30 year old unices that no one uses. The autoconf team consistently refuses to do that which why the software wont ever become better.

CMake, SCons, Waf and probably half a dozen more tools all do a much better job configuring software than autoconf ever will. People really need to try those alternatives and I'm sure they will realize autoconf needs to die, die, DIE.

CMake, SCons, etc, all have their pros and cons. They are hardly paragons of virtue though. [Yes, even as a mere "consumer" of CMake (someone who occasionally needs to build packages that use it), I've been bitten and frustrated by it.]

As far as I've found, there really isn't any build tool that really gets everything right -- which suggests that the problem is harder than you suggest. [Although judging from the number of people who try to write their own build tool (most of which languish and eventually die, but cause some pain along the way), it's clear enough that many people think it's a simple problem...]

> autoconf needs to die, die, DIE

Decaf?

It doesn't matter for a "works fine" person how many faults their current tool has or how much smoother "your" tool solves all those problems, they won't listen. Logic doesn't work maybe because it is very hard for them to get used to something new. Maybe because if you haven't experienced any alternatives it is very hard to imagine that something could be better.

I think bazaar's like the Turkish one referenced above (with guilds and standards of quality) are the best of all worlds. but to get there, a guild needs a strong, opinionated leader. Torvalds perhaps. (PHK seems to fit the bill too :-)

So, what I suggest is the issue here, is too many people trying to build a cathedral in a bazaar - they want everyone to build it their way but cannot persuade enough others to follow them. As such the bazaar is full of many half-finished carpets instead of less, perfect carpets made by engaged teams.

Software needs many talented people, working together, mentoring new ones in "their way". A bazaar is great for getting different talented people to find each other and choose between many different projects. A Cathedral is great for encouraging apprenticeships, the promulgation of culture and so on.

I think that one of the best lessons to take away from this whole flame fest is that history is something to learn from, and not just blindly revere. But first, that requires knowing about history, which far too many people don't. The whole issue of whether cathedrals or bazaars produce better software is orthogonal, and needs to be addressed to different situations.

First, are there still places where cathedral style development is a very good idea, hinging on being necessary? Yes, but it has to be done right, and unfortunately, cathedrals aren't (usually) very flexible. The Mars rovers were all cathedral programmed and work pretty darn well; on the other end of the spectrum you can find all sorts of big projects that went awry, even with "coherent design".

Second, is cathedral design necessary for all software today? No. I'm fortunate enough to have the perspective of a second generation programmer; my father started in middle school by sending punch cards to the university and would get his results a week later. Careful planning was not just a good idea then; it was absolutely necessary. These days he has a computer with more power and connectivity in his pocket than was available when he was born. His turnaround time for seeing results on software changes these days is well under five minutes.

So the computing industry has come so far so fast that it's not just unnecessary to make grandiose plans in advance, it may actually be a bad idea in a number of cases (what was it Paul Graham said? "If you want a recipe for a startup that's going to die, here it is: a couple of founders who have some great idea they know everyone is going to love, and that's what they're going to build, no matter what.").

It's kind of nice that anyone can just pick up a computer and start writing software these days. Sure, sometimes it's scary too, to think of all the inefficiencies, and wrong results and security holes that plague software. But it's so nice and liberating to think: "you know what? I can do it better, I have the freedom to try to make something better." And you can develop all cathedral style if you want!

Could software be improved? Definitely! Should people learn from history? Yes, including the mistakes. Would more design and coherent vision help make better software? Maybe; go ahead and prove it (a quick note: unfortunately, the market doesn't select the best software or designs, so you will have to prove it some other way, but being successful market wise is a good start).

I will tell you, though, that Snow Leopard missed its ship date by 8 months (of course, you never heard about that because Apple is smarter than to announce an OS ship date before it goes GM). It was also nearly un-usable for about a month, and only marginally usable for two months beyond that, during development. Lion suffered from none of these issues...

I guess post Snow Leopard era OSX and Apple apps just feel cheap, unpolished and rushed. It actually feels a lot like Windows, where quality of different parts of the system varies wildly.

Examples: Finder and anything file browsing related beachballs much more than before (particularly when dealing with network shares). Visual design seems to have taken a hit - everything seems to just consist of either different murky shades of grey, or jarring skeumorphism. The 10.6-era AHCI kernel panic hasn't happened for a while now, so maybe that was fixed in a newer revision of 10.7 or 10.8, but userspace stability seems much worse. Some of it is crashes, some just weird UI glitches that require a restart or Force Quit. Old problems (like Mail.app reliability or the heap of shit that is iTunes) are ignored completely in favour of sexing up the UI. (the runaway dynamic pager issue FINALLY seems better on 10.8, but again we get loads of other regressions)

Add to that what you say: the "improvements" leave most people I know completely cold or even get in the way, and that's not just power users/developers. Since we're "paying" for the (subjectively) unnecessary bling with regressions, the overall impression is that of a negative change.

Not that any of this is particularly relevant to the development model. You can end up with festering layers of crap with either model if nobody feels responsible for overall quality. I suspect the objective for Lion was "make OSX look kind of like iOS and get rid of all of that GPLv3 software" and that's basically what was achieved. I guess bug-free software isn't good business, Microsoft were printing money for decades.

You can say that again. Frankly, after using Mac for 10 years, this is my last one.

Hackintosh or something even worse (and I am with you 100% on the horrible state of OS X) -- I wouldn't want to discount the quality of the hardware.

In fact, it actually reminds me a lot of the change from FreeBSD 4.x to 5.x ... things worked, people were happy ... and then boom.

That was when all of my desktops and laptops stopped running FreeBSD and I "switched" to the mac. I am fairly certain that once running SL stops being practical, I will switch again.

Nothing against a good conspiracy theory, but if you truly think this is intentionally planned then this thought is a bit … dumb.

It does feel that Apple's development model is changing to an annual release cycle, wherein larger/systemic bug fixes are deferred to the next major release, rather than in a point release. In fact, most OS X release these days feel like "minor tweaks that make consumers happy + a truckload of bug fixes and plumbing". The fact that they deferred so much 10.0-era plumbing to 10.6 and beyond is indicative of why there's a lot of regressions, IMO.

Minor irritations like reworked Spaces and full screen mode making a second monitor pointless are irritations, but I wouldn't regard them as reflective of the quality of Lion and ML.

Whether products are developed according to what users want vs what builders want has almost nothing to do with bazaar vs cathedral.

Indeed, many bazaar-style projects (the linux ecosystem in general, gimp, etc) are almost cautionary tales about "builders creating what they want for their own niche purposes and if anyone disagrees they're free to fork off".

Their inscrutable beauty is buried under tons of libraries nobody will ever touch for fear of breaking 20 years of development efforts, exactly like what happens in the Unix world. Their move to the 64-bit world was painfully slower than what their fellow merchants accomplished in the Unix bazaar. They still provide compatibility layers for programs built with technologies that have been thought of as extinct, like monks still praying to the gods of ancient Greece. Whenever they went for the "total reuse" mantra, they built terrible and insecure specifications (DCOM) that still saddle us 20 years later. And let's not even talk about portability, which is anathema: to each his own Cathedral and his own Faith, touch ye not any unbelievers!

So yeah, making mistakes and keeping around the cruft is something every long-running IT project can experience. Unixes are, arguably, the longest of them all, so the ones that naturally tend to show it more. Besides, there's a whole new world of applications to be built out there, if we were rewriting libtool every three months we'd move even more slowly than we do now.

<ad-hominem>Oh, and I wish I could say your rant is unbefitting of professionals of your age, but I'm afraid it's actually quite matching the grumpy-old-man stereotype you're clearly striving for.</ad-hominem> Hey look, I can do ad-hominem too, and I was born in the late 70s!

You diagnosis of its qualities are spot on.

"ad hominem" means to "attack the man", ie: I single identified man, saying some generalities about identifiable groups is not "ad hominem".

And yes, I am a grump old man, and a surprisingly cheerful one at that.

Your remarks are essentially Ad Hominem regardless of whether you single out 1 man or 100 individual men.

The words "cathedral" vs. "bazaar" have obviously gotten too vague for people to meaningfully argue about. But "coherent design vision" has a much clearer meaning, and an enormously important one.

I think the pendulum is bound to swing back from incoherent, hypercomplex software, because at some point the reductio that we've got today simply won't be able to adapt. Who knows what will trigger that, or when. But intelligent people will always care about simplicity, beauty, efficiency, and the other qualities that come from good design. And among the ignorant programmers there are more than a few whose eyes light up when they are finally exposed to good design, and want to learn to work that way. I'm an example. So there's no point of no return here.

The UNIX design philosophy -- small utilities loosely joined into a coherent whole -- is a coherent vision for a platform, under the definition you've repeated. And the Microsoft Office design philosophy, if what you're saying is true, is a bazaar because it's "incoherent."

If the UNIX ecosystem is a Cathedral and the Windows ecosystem is a Bazaar, I think we can safely say that you're not using the words that way the classic essay used them, and that the definitions you're using are malleable enough that any argument about aesthetics could bend and twist them into synonyms for "stuff I like" and "stuff that annoys me."

Coherency cannot be measured, and it's extremely subjective. The *nix world can be extremely coherent: for example, you'll always find libtool, whether you want it or not, and your '70s-like filesystem layout. Isn't that "coherent" with its history?

This is not true even of Android and iOS, the modern wonderchildren he cheers on: a quick look at the Android filesystem hierarchy will show its adherence to outdated conventions (etc?) and incoherent repetition (sys? system?), and I bet you'd find something similar in iOS. So there might be a "coherent vision" behind, but the practice is quite incoherent; and mind, we're talking about very basic OSs that delegate any but the most basic functionality to third party apps. I'm happy to bet that in 10 years, the Android codebase will be as shitty and crufty as any Unix.

The truth is that a cathedral, in phkamp's rant, is "software that can all fit in one's head". For all operating systems, we're well past the stage where a single genius architect could envision the totality of a massive cathedral, for a number of reasons (time, working culture, legacy tech etc). Clearly phkamp, being more intelligent than most, only just passed that threshold with the very latest FreeBSD release, and felt the urge to tell the world.

"Software that can fit in one's head" and "coherent design vision" are closely related things. I'm in favor of both. If you're saying we're past the point where that's feasible in real-world systems, that's just assuming the conclusion – the wrong conclusion, in my view.

So, you are defining "cathedral" as "good"?

Microsoft didn't architect their software as software, they shaped it as tools of monopoly enforcement.

There are numerous accounts from people involved about how marketing decisions relating to cutting of 3rd parties or trying to bludgon somebody into submission caused them to cripple their own software and its architecture.

See the evidence in the M$/Novell case for ironclad evidence of this.

This kind of passive-aggressiveness only reinforces the grumpy-old-man stereotype and makes it really hard to take you seriously.

I do think that was a mistake. But MS didn't (doesn't) have the luxury that Apple did (and still does). MS can't just shut 30 years of compatibility off and expect to keep their corporate customers happy.

Oh and just for the record (since this thread seems to be grouped by age, I was born in 1963. I remember when MS was cool and "us" nerds were running from CA and IBM.

It's been twelve long years that OpenOffice moved from the cathedral model to the bazaar model. And it still suffers from the same criticisms you level against Office and more.

It is important to note the distinction between FreeBSD's package system and say Debian apt. In FreeBSD I can make from source some package, and in Debian I can apt-get install a package, because the Debian packages are prebuilt it just comes over in several chunks and doesn't need to build. (yes you can pull prebuilt packages for FreeBSD too). But my point is that the packaging system of FreeBSD, as used, conflates building the packages and using the packages.

So if I write a perl script that goes through the source code and changes all the calls to fstat() to match your configuration then to build I need perl even though you do not need perl. (as an example). But to run I don't care if you have perl or not.

But lets get back to the volunteer/paid thing again. People who volunteer rarely volunteer to clean the shit out of the horse stalls, no they volunteer to ride in the races and build a new horse. So you end up with a lot of stuff around you don't want.

Sadly for operating systems, and the original rant is really operating systems, there really isn't a cathedral/bazaar model, its more of a democracy / feudalism kind of thing. Nobody 'owns' the final user experience for the OS in FreeBSD/Linux discussions.

It's also a matter of what people are willing to pay for. Even (or perhaps especially) in commercially supported software, cleanups only happen if there is a strong business case for the cleanups. And if the cost of buying an extra build machine to run long complicated configure scripts is significantly less than the engineer time to re-engineer a new autoconf system from scratch, in many companies (and certainly most starupts) --- it won't happen.

The OP cares very much about code quality as a good and important thing in and of itself. But that view isn't shared by many business people, or by many programmers in general, for better or for worse. Some have argued that OSS code tends to actually be _better_ about code cleanliness, because it's public, and people do care about making sure that their code is clean. (I've never seen the proprietary source code for the Oracle DB, but there are many stories out there about how horrible it is from a code cleanliness perspective.)

Also, the OP seems to care a lot of lots of extra library dependencies. The big problem here is that a lot of people don't really care if their package uses perl or python as part of their build scripts/makefiles --- or even if their package uses perl _and_ python scripts. The OP cares, but for better or for worse, most people don't. And I would wager this is likely true at most companies where programmers are paid to maintain the source tree as well!

Yes. Much of his poor experience is due to build systems.

Anyone who's developed anything can tell you how fragile and finicky build systems can be.

Since build systems are used by developers, who can deal with complexity much better than "ordinary" users, they tend to be rough around the edges.

They're also unglamorous infrastructure, so volunteers tend not to spend time on them if they're not paid. And they have a tendency to break the entire application if broken, and are often part of the "interface" of the software (since every distro who builds package X from source uses build scripts that rely on for example ./configure with particular options), so maintainers tend to take an "if it's not broke, don't fix it" attitude.

GNU projects in particular -- whose autoconf the author complains of -- have quite a bit of built-up cruft since many of them are very old, and also vital infrastructure of most FOSS OS'es.

Of course, I'm biased, because I have such a degree. However, when I compare what I build to what is built by a business school grad who thinks anyone can learn computers, so everyone should be taught business, I really cringe at loss of formality in the industry.

Needing to build a computer processor from AND and OR gates really drives the concepts home. Building an operating system in C++ really drives the concepts home and creates a framework for thinking about computer based problem solving that's lost in many of the systems I look at today.

I really think it is a matter of exposure -- at one time all anyone ever saw were cathedrals, and so that was all anyone ever built. I feel now we've swung too far the other way, where there are many in the craft who've never seen a cathedral and have only known the bazaars. The reality is we need both, and the craft of building cathedrals is becoming endangered.

You are a terrible example because you come from a formal background of a hard science. Even worse, you come from one where proofs (the cathedral in the mathematics sense) are required or you aren't taken seriously.

The parent was talking about joe schmo off the street or the high school wunderkind who is building "twitter for teens" or "pinterest for social good". They are just slapping it together from the get go without thinking about any aspect of the design. Eventually they'll have to hire the formally schooled to come in and clean it all up, if they actually get anywhere with it.

Which makes me wonder, if more formally educated individuals were in the start up game would the failure rate be so high as it is now?

Uh, we are talking about the same Cathedral and Bazaar thing here, right? Cathedral in the sense that only the anointed master architects get to make decisions; everyone else does what they're told and keeps their mouths shut; bazaar in the sense that everybody brings whatever they've got and hopefully the good stuff hangs around?

Everyone brings whatever they've got, and the anointed master architects sift through it, determine the best course of action, redirect known-bad avenues of exploration and possibly anoint new master architects from the best of the candidates.

I asked a question, you are making a statement. If you have numbers that prove that the formally educated are failing more then I think we'd all love to see them. Otherwise, this is just conjecture.

As for 'science projects' failing it could be a number of things. I would assume that hard science startups fail a lot of the time because of cost. "Twitter for teens" is much cheaper to start than one doing alternative energy. Also, hard science is less sexy than a lot of the social stuff that currently dominates the landscape. Thusly, it can be much more difficult to find the money you need due to visibility. The social startups will probably pay out sooner than long-tail hard science companies.

This isn't to say that they don't exist, but there are a wealth of examples demonstrating the transition of major progress in those fields to companies like Google and Amazon. For anecdotal examples, see GFS, MapReduce, and Dynamo; there are many unpublished/trade-secret examples that will take years to come out.

Additionally, many schools have reduced their requirements for what exposure students should have to systems and OS:

A Harvard CS concentrator, for example, is not required to take either the introductory systems or operating systems classes; he/she may opt to take a high-level mobile programming class instead. Indeed, the operating systems and distributed systems courses are taught every other year, further reducing enrollment. The last time CS 161 (intro OS) was taught, there were only 23 students. 25 in CS 153 (Compilers). This is a stark contrast to the 100-200 students who took other upper-level CS classes (189 in mobile, for example).

While I will not contest the value of the formal CS education, I see some of the 'naive'(for lack of a better term) CS proliferating even there.

From googling around, might you be referring to:

"Systems Software Research Is Irrelevant"

http://doc.cat-v.org/bell_labs/utah2000/

And there's a little slashdot discussion:

http://interviews.slashdot.org/story/04/10/18/1153211/rob-pi...

I don't agree strictly with everything he says. It is a polemic rather than a careful argument. But still, he has a point and the data about OS research is strong (even beyond his data).

Of course, I may be biased because I've seen more poorly-built results from cathedral approaches.

My company ships quite a bit of FreeBSD (as pfSense). Could have gone linux, but linux is a mess (much worse than ports.)

I think OpenBSD is a mistake, at best it belongs as a group focused on security inside the netbsd project, but of course, Theo got kicked out of netbsd, thus: OpenBSD.

I'm also the guy who appointed Russ Nelson the "patron saint of bike shedding." Just FYI. :-) (None of Eric Raymond, Russ Nelson or Theo de Raat like me much.)

The first time I read of 'Cathedral & Bazaar' I thought ESR was illustrating and contrasting the BSD .vs linux development model. Only later did I understand that he was pointing fingers at GNU/FSF, not BSD.

There exist several larger OSS projects, such as Apache, Boost, the Kernel, etc, which accept contributions but are also curated. Thus they represent sort of hybrid between cathedral and bazaar. People who use these projects know they are getting some (varying) standard of quality.

I think these sorts of projects -- often shepherded by some kind of noncommercial Foundation or Organization -- are the best way to get a mix of openness and quality going forward.

The only example that jumps out at me is the original Unix, and I think you'll agree that comparing software from 30 years ago that does vastly less than ... pretty much anything out there these days is not an entirely fair, nor useful comparison.

Android is the most original-UNIX-like in it's development but I imagine you could extend it to systems in which large portions of source are available (since that is the kind of project C&B discussed) such as iOS and perhaps Java. Google Chrome. vBulletin. The Pine MUA. The ssh.com RFC 4253 implementation. Various parts of RHEL. And probably numerous "Free" programs that end up following the Cathedral model merely due to the culture of their maintainers.

We can think of many but the point is to make sure we're on the same page. It's not like all software projects in the world are neatly divided into "bazaar" and "cathedral". I don't know that Google Chrome is a "cathedral", for instance.

It's ridiculous that the author refuses to give a single example, instead opting to say basically "see! You people don't know what a cathedral is, just like I said."

I will cite as evidence two of my favorite programs ever:

WriteNow (for Macintosh) an excellent early word-processor written entirely in assembler. Unequaled for many years for its combination of stability, raw performance, and ease of use, ultimately it simply could not add new features (let alone make the jump to PowerPC) and died on the vine.

HyperCard, which was perhaps one of the most dazzling, innovative, and influential products ever to ship, which pretty much stopped evolving once its original programmer lost interest.

Today we have the phenomenon of the incredible version one product, usually developed by one person, which never really makes it to 2.0. These seem very much like cathedrals.

Most projects go nowhere, but once a project gains momentum it's going to split into pieces and each piece has a chance of "going bazaar" at each step.

Now, I've not been working in the software industry since before 1990, so I'm obviously not qualified to comment, but I'll say this: I can understand why a cathedral model might be warranted. Even using something like CMMI might be a good idea in some cases.

But for a lot of things, especially exploratory/experimental things, the bazaar model is really nice and can reap significant benefits. And the nice thing about the bazaar model is that if you want to follow a cathedral model, no one is stopping you! Go off and be your own little dictator with a "grand unifying vision". Come to think of it, that seems to be what many of the most successful open source projects are: one (or a few) people have a vision of an itch they want to scratch, and they pursue it with a bloody-minded persistence. The bazaar only comes in when someone forks or in the fact that anyone can compete or (try to) contribute.

And BTW, I do know who you are, and have a lot of respect for you, but in some ways this article (and your comments here) could be read as reactionary against the success of Linux and other more open OSS; the BSDs have always been more insular (or discriminating depending on your POV) and developed like cathedral models than Linux; oddly enough this has resulted in three distinct BSDs while there is still only one Linux kernel. I will agree that reading Brooks (and other computer history) is almost always a good idea; just MMM was enough to open my eyes to how little the industry has progressed (VirtualBox/VMWare? That's nice; IBM was designing full system emulators for hardware that didn't yet exist in the sixties).

I think people do want to make things better, but its happening much more decentralized, with small teams taking small safe steps on tools, libraries & frameworks.

Media, on-line banking, 3d printing, accessible hardware hacking, e-commerce sites, scientific/engineering software. The majority of these things are made by small teams plugging the best libraries and platforms out there together, where they already solves part of their software problem. Android, Cloud, Linux etc. It would be silly to say these were without ambition and don't contribute to the community.

(The average software devs Impact Factor may be diluted by the increasing number of people working with computers, but computers are so globally useful its inevitable. If good things are still getting made then who cares.)

In this comment I will equate "Rug Market" with "Ready, Fire, Aim" and "Cathedral" with "4 Year Plan".

Firstly, the "Rug Market" beats the "Cathedral" when you haven't formulated the problem properly, and so you have bad specs.

Secondly, the "Rug Market" beats the "Cathedral" when bad software is more profitable than good software. Google for "Worse Is Better" and "The Innovator's Dilemma".

Sad but true.

I've been equally disgusted by the evolution of programming in the last few years. HTML - which is, to a first approximation, always invalid. JS - which needs jQuery to make it mostly-but-not-completely cross-compatible among browsers. CSS, which hides so many hacks on top of each other, and which even needs a reset to be compatible. And now - distributed systems with pieces in PHP, Python, Objective C, Dalvik-Java, etc... and sustained by awesome-but-hackish fixes like Varnish and FastCGI and Nginx. where everything seems to be put togetther with duct tape.

But I'm slowly getting to the next stage, and having to admit - no, actually, admitting - that if it's spread like fire, there must be something to it, no matter if it hurts our taste so much.

And if you look well into it, it's very similar to biological evolution. Our own genetic codes and body plans are full of ancient pieces that are not needed any more (or at all, male nipples anyone), but it seems that it was more economical to "patch" things than to fix things properly. Or at least, it didn't hinder the current designs enough that they wouldn't succeed over the alternatives evolution probably also tried.

And now Google translates text statistically, without even trying to understand things.

One fear I have is that we will be able to create a working AI in a few years... and due to the way we do it, we may even not understand how it works.

I find jQuery more analogous to tools like configure and autoconf. (in that they aren't actually needed for "modern", standards compliant browsers.) As an example: there are already many lightweight drop-in replacements that assume a sane browser to begin with.

CSS is remarkably hack-free and the resets are just to remove the default styling that browsers have built in. Obviously browsers need builtin styles otherwise all the old pre-css pages would stop working.

And why are you calling programming languages and web servers hackish? How is nginx hackish and apache not? How is Objective C hackish and C++ not? Why stop inventing new languages at assembly, c or c++?

Other than that I generally agree with the rest of your comment. I think if we ever get to some kind of technological singularity we'll almost certainly and just about by definition not understand how it works. And true/strong AI would definitely be a singularity. Even if we understood version 1 we would likely not be able to understand whatever it dreams up 5 seconds after we overclock it. And we will - moor's law and all that. ;)

Back in the day, a "serious" application usually involved a large amount of quite homogeneous source code in a single language (or 2 or 3 different languages for different tasks: C/C++ for the main code/engine/logic, assembly for performance-sensitive code, and maybe some custom high-level script for high-level application logic). This was built usually on a single machine with a single build script, and resulted in some binary which could be deployed.

Even early web apps were more akin to this model - Java apps, or even Perl apps.

Nowadays, an "app" lives distributed among dozens of servers and client types. People describe what they have: "12 memcaches, 5 varnishes, 8 nginxes, 5 app servers with Ruby, 4 static content servers, and a MySQL master server with a fallback master and 5 slaves for reads. Separately, an Android app, an iPhone app and an HTML5 front-end."

These systems started as a single app in a single box, and have "grown" for scalability, reliability and security, being patched up with different pieces of technology put together with often unreliable methods but fallback mechanisms that make it more resilient that if reliable methods were used but no fallback. It's not so important nowadays that the application code be clean, elegant, or failproof, but that measures are put in place so that the service will keep chugging along most of the time.

And setting up the whole system from scratch involves a manual with probably hundreds of steps putting together haphazard technologies, installing different types of Linuxes and packages for each piece. And may even involve difficult-to-replicate steps, such as using an AMI to launch EC2 instances, where the engineer that created the original AMI doesn't even work here any more and recreating it from scratch would involve 20 packages, Googling for 5 of them as they're not available in normal repositories, 5 manual patches, two secret incantations and a tribal dance around the chair while singing and praying to arcane gods of long-forgotten package managers.

{{Side-note: although it's not the main point, I'd definitely say Objective C is more hackish than C++. C++ has its own amount of weird stuff due to being built on top of C and its evolution with templates, etc... but the evolution of Objective C is done much more in the form of "patches". Even the original syntax takes advantage of awkward gaps in the syntax of C (@interface, @implementation, #import, [object msg:param]?). But it's even worse how new features are piled on top of old ones: properties, ARC 2.0, etc... it's distinctly noticeable when you just wouldn't feel comfortable teaching programmers new to Objective C how to use ARC, without learning the underlying memory management model first. In C++, people can learn "new" and "delete", and never hear about malloc()/free(), and there's no problem at all.

Anyway, my point about hackishness was not referred to ObjC, but to how whole systems are engineered nowadays.}}

It also comes back to "don't reinvent the wheel" and "not invented here syndrome". You _could_ implement many of these things in your homogenous codebase or you could just reuse something that already works. I'm just saying there are serious and good counterarguments to the "large amount of quite homogenous source code written in a single language" approach.

I agree with some of the details (like AMIs, EC2, different types and versions of linux, different package managers in the same system), but I think in general the things you think of as hacks aren't hacks at all and just the signs of progress. Or at least the nature of large, complex modern distributed systems. You basically described the term "agile development" and you're remembering a time many developers would rather forget with rather rose-tinted glasses.

(I don't really know enough about objective-c to comment.)

This comment seems to mostly be, "I'm going to complain about software disciplines and lines of CS research that I don't understand because their practical approaches don't conform to my finicky aesthetics."

If you want to formulate Strong AI to solve NLP as a problem, go for it. The rest of us are happy to use Google as it is.

He's probably right that lots of software could be designed better, but I think he's wrong that that's of paramount importance. We need lots of software these days and we simply don't have the resources to built it to Kamp's standards. Also, we've learned that our requirements change so fast that his beautiful design would quickly be twisted into the pile of hacks that he hates.

He notes how long it takes to compile the software that runs on his work machine---how long would it take to compile the software on a windows machine? (Or whatever he would claim is the standard-bearer of his cause).

He also attacks open-source software. The truth is, we have an amazing amount of free and open source software available. Some of it may be flawed in design or usability, but it enables us to solve so many problems (and look at/modify the software when needed). I don't think that all software needs to be "free", but I do think free software has made us all much richer. I have a hard time seeing all this value that the Bazaar has created as inferior to slow-moving, centralized, big, up-front design development.

You seem to assume that cathedrals are "slow-moving, centralized, big up-front" designs, where did you get that idea ?

Ever looked into how USA put a man on the moon ?

Maybe you should. Also: Read Brooks book, if you can.

Why do you think most of the software we build now is like putting a man on the moon? I think it's nothing like that.

Maybe I should read Brook's book (I've enjoyed MMM), but your article doesn't make a strong case for it.

Because you claim he's attacking open source software, and you said that maybe he thinks MS is the standard bearer of his cause.

The (easy to find) facts are that he has committed a lot of code to open source and has been involved in freeBSD for many years. These are not small trivial bits of code used by a few people or for insignificant reasons. Crypt and varnish are bits of code getting billions of uses.

To make a convincing argument, he should have made it clearer what he was arguing against ("the bazaar" is too nebulous and refers to many things, .com development refers to other things) and what he was arguing for (which was entirely missing, other than a reference to Brooks' book). If he doesn't have a clear vision of how development should work (or a "standard-bearer"), he should have at least provided some examples of things working the "right" way (or better).

He should probably also have left off a bunch of the insults and hyperbole: "clueless" / "hacks" / .com period a "disaster" for code quality. I'm frankly pretty surprised that ACM Queue would publish this.

PHK is one of the most prolific contributors to Queue. Most regular ACM readers know who he is.

> I'm frankly pretty surprised that ACM Queue would publish this.

Again, see above.

OP's site bikeshed.org (http://bikeshed.org/) has some nice proposed features for software that posts to large audiences.

     +------------------------------------------------------------+
      | Your email is about to be sent to several hundred thousand |
      | people, who will have to spend at least 10 seconds reading |
      | it before they can decide if it is interesting.  At least  |
      | two man-weeks will be spent reading your email.  Many of   |
      | the recipients will have to pay to download your email.    |
      |								   |
      | Are you absolutely sure that your email is of sufficient   |
      | importance to bother all these people ?                    |
      |								   |
      |                  [YES]  [REVISE]  [CANCEL]                 |
      +------------------------------------------------------------+

      +------------------------------------------------------------+
      |	Warning:  You have not read all emails in this thread yet. |
      |	Somebody else may already have said what you are about to  |
      |	say in your reply.  Please read the entire thread before   |
      |	replying to any email in it.                               |
      |								   |
      |			     [CANCEL]                              |
      +------------------------------------------------------------+

Why is it wrong? For example, consider the second message. Suppose you haven't read all the messages in a thread because you've been on holiday for the previous few days. But a friend pointed you to a message in the thread which specifically mentions your name, asks "could you verify this for me?", and doesn't have any followups?

Why should your newsreader force you to read all of the thread in order to answer something which isn't in the rest of the thread? Yet the only option there is "cancel".

Furthermore, "at least 10 seconds" is completely wrong. People killthread, plonk people, and develop other ways to ignore discussions. Assuming 4 hours per day means people can handle at most 1440 messages. There are 276 comments already in this thread, but I expect most people sampled a few messages on each major branch, skimmed a bit, and perhaps did a text search to see if someone mentioned a key word. Most assuredly, the entire HN readership did not read this entire thread.

You took the time to write a long essay, and present a novel idea. A lot of posters here disagree with you, but the mindset on HN has been to build first ask questions later. Convincing people in a single day to change their world view would be a huge undertaking regardless of whether you are thinking about it correctly or not.

But some topics would require entire books to present in a context free way, this is one of them, and I don't have the time and money to write that book, so you will have to make do with a column @ACM.

When I read something, and have the nagging feeling that there is context I'm missing, I go looking for that context, usually starting with "who the heck wrote this", rather than assume that the writer is a clueless bozo.

But whatever blows your hair back buddy...

You didn’t even consider that what I wrote was meant as a complement to the fact that you are taking a view that no one else currently takes. An idea that has real merit and is worth discussion. You took my comment as an insult and wrote:

“But whatever blows your hair back buddy...”

To me it means: If you're happy what you're doing, then who am I to tell you're wrong ?

I don't think the appeal to authority fallacy will fly here. We are all too smart for that.

You arguments should be evaluated for their coherence and, in the end, it really doesn't matter who you are or what you did if you are wrong.

To use your example, it doesn't matter if you are Von Braun himself. If you don't bring enough propellant for your descent engine, you'll still crash on the Moon.

He also mentions 'one person having responsibility'. Does he mean a benevolent dictator? Linux and Python have that. They're imperfect in their own ways too.

I understand where you stand and agree some of your criticism is valid, but "quality" doesn't have the same meaning for everyone.

Second, just because there is one bad example of a cathedral, doesn't mean the cathedral model is bad.

Many years ago, a fella named Gettys (look him up!) wrote as part of sage advice for OSS philosophy:

"The only thing worse than generalizing from one example, is generalizing from no example at all."

right, but this could be applied to your example of the "configure" script as an indictment of the "bazaar" model too. Surely there's software that's come from the "bazaar" model that you feel is of high quality ? There's none whatsoever ?

Why don't we employ rigorous software engineering principles to, say, iOS games? It would be a waste. It could be done, but in practice, people just don't care as much if a game on their phone crashes as they do if an airplane they're riding in crashes.

There's a wide spectrum of software out there, needing varying levels of robustness. There's not a lot that can be said about software development as a whole along these lines; inadequate design for an online banking site might be excessive design for a llama costume competition voting site.

Nobody says you have to write iOS games using the same methodology as avionics, what somebody is trying to say is that random walk will only get you so far...

iOS is a pretty bad example btw, because iOS is very much a Cathedral and has a designer and architect who cares and who is in control.

That's precisely why don't need to use autoconf to compile stuff for iOS.

You don't need it to compile stuff for one specific version of Debian with packages X, Y and Z installed, either. You're comparing apples to oranges.

The fact that something only runs on iOs means it's a fairly limited piece of software in terms of portability - which is an acceptable tradeoff for many things, but not for others.

It's very difficult to learn when everyone is just ranting.

If you try too hard to design a rigid structure around anything, it can come back and bite you in the ass. Part of the reason why this seemingly disorganized and haphazard collection of buildings makes a working and thriving metropolis is because of its diversity and resilience to change and progress. If you attempt to over-engineer something of such complexity, you might just end up with Pyongyang instead of New York. Instead, we have a competing ecosystem of libraries and options with the good ones theoretically rising to the top. Better communities and communication of these values make this process work even better, just as in the greater economy. For all its complexity, this process works surprisingly well.

For all the explaining Kamp did in this article, the one thing he failed to account for was the resounding success of the current model. He did use a surprisingly apt metaphor, however. Cathedrals are no more; their old place as the center of society had to be explained by a series of myths and lies that placed ideals above reality. The thriving Bazaar easily replaced them at the center of any thriving modern city, based on the simple truth that the dynamic edge of reality was ever-changing, and that quality based on that reality would indeed be more successful.

Could it be better? Of course. But it is reality and truth that will move us forward—not mythology. Good riddance to cathedrals.

Have you never wondered about their surprisingly coherent APIs and wondered why that was so different from, say, UNIX ?

See my other comment about knowing what a cathedral is to begin with.

I guess you're saying that there's no reason UNIX should be different—that it has somehow become a disorganized mess of libraries and code re-use, and that it is broken.

I don't see this. Android and iOS are like brand new cities; Dubai and Abu Dhabi. UNIX is Paris. Perhaps you can make some updates to the city infrastructure, but in the end the streets are still cobblestone underneath, designed for horse carts and not cars. But you have the Notre Dame at the center (ironically, a cathedral), the architecture, tons of history, you have a thriving culture and countless people working and playing in this city every single day with great success. You can complain about it all you want, but there is a reason that people are fundamentally attracted to it.

I think you're right about a lot of things, but if you propose rebuilding Paris, well, you may have some opposition. I think you might want to build a new city, rather than trying to build over very old history with some ideal of the cathedral you think it should have been. And to use your examples again, that strategy has been very successful for Android and iOS.

Read Brooks book.

Will do, thanks for the thought-provoking article.

[1] http://en.wikipedia.org/wiki/Haussmanns_renovation_of_Paris

Citation needed. :) What metric did you use to measure the efficiency of the software bazaar?

Efficient?

Since 2001, we've started in on the XP/Agile movement, which, think what you will of it, test-drives and version-controls relentlessly, and fosters a constant dialog about what quality is and how to achieve it.

Furthermore, some marvelous tools have been written in the last decade; I can hardly see how the author can complain about version-control systems, as Git and Mercurial--both miles better than what was available in 2001--are not least among them.

Other quality-enhancing things that have happened since 2001: continuous integration, build pipelines, Selenium and friends, behavior-driven development, REST architecture triumphant over the cathedral-like SOAP and XMLRPC, JSON over XML, lightweight messaging, the adoption of small, focused open-source tool over "quality-focused" large vendor bullshit-enterprise tools.

To the extent that UNIX sucks now, which it doesn't, it's because the hackers who work on it now have lagged the industry--not the other way around. So go find some other group of kids to shoo off your lawn.

Really many of ideas which we've come to associate with "Agile"-like methods only really started to become codified about the end of the 1980's in more traditional manufacturing. I don't think it's too surprising that it's taken the software industry another 10 years for this stuff to become mainstream.

Also from your other comment above...

> Ever looked into how USA put a man on the moon ?

I'm sure FreeBSD would be a lot better if you gave it $25 billion USD over 11 years as well.

It takes more than just good will and intelligent people to produce high quality results.

see: Tom Gilb Principles of Software Engineering Management (1988)

We do have an awful tendancy to reinvent the wheel, but I think that building Cathedrals vs building Bazaars is the wrong metaphor - I prefer building simple vs building complicated.

The "bad" people PHK describe are building complicated. The "good" people build as simple as possible, but no simpler.

edit: 1992. Only been building bungalows.

UNIX was originally written as pair-programming.

Brooks refers to agile in The Mythical Man-Month (he didn't invent it, check his references)

You are being overly optimistic about the pace of human cultural development. The rest of the world is only just starting to realise again that perhaps we have cultural issues that lead to low quality results. Bell Labs have numerous advantages over the rest of the world when it came to fostering the sort of culture that produces high quality results. Just reading the history books really isn't enough for that to come about.

TLDR; I blame Windows ME on World War 1.

From http://www.drdobbs.com/open-source/interview-with-ken-thomps... :

KT: I did the first of two or three versions of UNIX all alone.

Later, he emphasizes that they didn't even look at each others' code:

DDJ: Was there any concept of looking at each other's code or doing code reviews?

KT: [Shaking head] We were all pretty good coders.

I remembering reading somewhere (Dennis Ritchie's C History article?) that once they both programmed a solution without consulting each other, and later when they read it, the solution was exactly the same, even down to the variables' name.

Power of C? Two people who thought alike?

>>KT: [Shaking head] We were all pretty good coders.

Which the modern world doesn't have enough of. Demand is so high that quality of supply doesn't matter, just enough to satiate the demand; which is in stark contrast to the early days, when computer science was a research field only privy to scientists and later to hackers and crackers. When it went mainstream, and grew to an industry, its no wonder that it acquired all that which comes with it.

I found the autoconf comments amusing, we have supported a pretty broad range of platforms, from arm to 390's, IRIX to SCO, as well as Linux, Windows, MacOS, and our configure script is 157 lines of shell. Autoconf had its place but I think it was much more useful in the past and now it is baggage. The fact that it is still used (abused?) as much as it is sort of speaks to phk's points.

I think the jury is still out on whether the bazaar approach is better or not. It sure is messier but it also seems to adapt faster. I worked at cathedral places like Sun, and while I continue to emulate their approach I also question whether that approach is as nimble as the bazaar approach.

I've voted for the older style of more careful development and I think it has worked pretty well for us, we can support our products in the market place and support them well. The cost of that is we move more slowly, we tend to have the "right" answer but it takes us a while to get there. Bazaar approaches tend to come up with answers of varying quality faster.

It's really not clear to me which way is better. I'd be interested in hearing from someone / some company that is supporting a lot of picky enterprise customers with some sort of infrastructure product (database, source management, maybe bug tracking) and making a success of it with a bazaar approach. Seems like it would be tough but maybe I'm missing some insight.

The way to address oversights in the bazaar model isn't to cram everyone back into the cathedral but to build support for change by showing where something is clearly better.

One area where PHK might see this is the way Linux has flown past FreeBSD - not due to endlessly-debated questions of kernel superiority but rather because Linux distributions like Debian provided a clearly superior experience for the overhall system by rejecting the decades of accumulated hacks (i.e. the ports system, monolithic config files, etc.) which had been the status quo for years and building better tools to reduce management overhead.

Debian contains tons of accumulated hacks that I wish would be rejected.

See also: MySQL vs. PostgreSQL.

This comparison doesn't hold true: while we might debate the technical superiority of the BSD kernel (and I have some NFS deadlocks to trade against ZFS), running many BSD systems is so much more work that it's more like comparing Postgres to Berkeley DB, where you have to take on responsibility for everything outside of the innermost core.

I ran FreeBSD and OpenBSD on a number of systems for years but was increasingly unable to justify the sysadmin overhead. There were only a couple of times where kernel issues were significant (NFS stability and SMP scalability, esp. when amd64 came out) and since Linux actually came in ahead of FreeBSD for our applications it was really unrealistic to locally duplicated all of the Debian infrastructure.

A lot of the commentary seems to be confusing the waterfall model with the idea of the cathedral.

Another way to make this point is to note how few programmers these days read Dijkstra, or Knuth's TAOCP. (Knuth would have us believe that even he doesn't read it. See Coders At Work.) Among other things, Dijkstra taught understanding the entire program before setting down one line of code. Contrast this with TDD (which, believe it or not, some take to mean Test Driven Design).

Lately I have been in the Application Security business, and nowhere has the issue highlighted by the article been more obvious.

Edit: Mark Williams Company, inventor of Coherent OS, was not a paint company. It started out as Mark Williams Chemical Company, manufacturing Dr. Enuff, a vitamin supplement.

His Discipline of Programming http://www.amazon.com/Discipline-Programming-Edsger-W-Dijkst...

There are some here http://cs-exhibitions.uni-klu.ac.at/index.php?id=31 that might overlap with the texas ones. One famous one is about the cruelty of teaching computer programming: http://www.cs.utexas.edu/users/EWD/ewd10xx/EWD1036.PDF.

His wikipedia page has links to a number of his seminal ideas.

But Evolution copes better than intelligent, top-down design with the evolving constraints of a market landscape that is constantly shifting.

Not to mention the fact that the bazaar is never going to put a man on the moon.

And yes, I wrote that piece.

What you have to consider is that individual cells evolved, until they created a whole greater than themselves. No individual cell was in control of that first step Armstrong took onto the moon. Likewise, what we end up creating from so much "bazaar" development will almost certainly accomplish something amazing...

...without any of us being able to take credit for it.

I think we can see abundant proof evolution can go a long way towards that specific goal.

The reason your work machine, running BSD, works the way it does (and self-assembles that way) is because each part of its software environment is independent, designed for different goals and constraints and it just happens to work together well enough you can work with it. It would be easy to standardize Unixes, but then you wouldn't be able to compile M4 for Windows or use one of the 26 FORTRAN compilers absent from your system. Also, what happens when you realize your design led you to a dead end? Evolution usually has plans B to Z built in. It's really difficult to make something evolve into a corner.

Having plans B to Z built-in has an impact on the overall elegance and purity of the design.

No, that's not the case at all actually. FreeBSD like the other BSDs is a complete OS, designed and built as a complete OS. It is not a collection of random tools by various people thrown together like a linux distro is.

  arcueid ~ $ du -hs /usr/src
  728M    /usr/src
  arcueid ~ $ du -hs /usr/src/contrib/
  307M    /usr/src/contrib/

Compared to that, in something like Debian the GNU tools, designed and built by the GNU with the intention to form a complete OS, probably constitute a greater proportion of the core OS.

Except that GNU ls was not designed to run on top of the Linux kernel, which is, on my book, a feature rather than, as you imply, a bug.

And if, next week, I write a better ls, I can install it and use it.

Also I know NASA is famous for their solid code, but personally I still question the amount of effort that goes into it. Just because their approach works (if you can afford it) doesn't prove that there couldn't be a more effective way.

I suppose with infinite money you can make quite a lot of approaches work. You could have an army double check every line of code. Instead of pair programming, why not have 10 programmers look over each others shoulders?

Lastly, lots of intelligent beings happily drive over cliffs all the time, just look at the financial crisis.

Bingo: it's all about the economics of various tradeoffs. If I have a company that puts together web sites for small companies in the area, and we have NASA style coding standards, we're going to completely price ourselves out of the market. If NASA applied "little web shop" coding standards to their projects, they'd be lucky to get anything up in the air before it exploded.

The important thing is to be on the curve though, not behind it, meaning to make the best of the tradeoffs the economics of what you're doing push you towards.

Spacex's "careers" page has a whole lot of "linux" jobs listed, so the Bazaar at least seems to be contributing heavily to potential future space travel: http://www.spacex.com/careers.php

Unfortunately, those statements seem to be contradictory; the United States ran the Apollo program for national pride, to cow the Soviets, not out of some grand desire to visit the moon. The science done by Apollo was secondary, and if you believe exploring space was a primary goal, you are going to have a hard time with the question, "why haven't we done it lately?"

Maybe you will not accept this statements for similar reasons like the ones driving your essay.

Article: This is a horribly bad idea, already much criticized back in the 1980s when it appeared, as it allows source code to pretend to be portable behind the veneer of the configure script, rather than actually having the quality of portability to begin with. It is a travesty that the configure idea survived.

Good high-minded notions here. But configure, with it's standardized parameters for how to do stuff, is near irreplaceable at this point. Certainly a more stripped down version, one not built on M4, would be wise, but libtool/autoconf itself is used too broadly & with trepid familiarity by developers & upstream maintainers: in spite of so much of it being indeed old deprecated no longer useful cruft, the best we can hope for is duplicating a wide amount of the existing functionality in a cleaner manner.

But at what cost would reimplementation come? How many weird build targets would for months or years go unnoticedly broken?

The place where we escape these painful histories is where we leaving the old systems programming languages behind. Node's npm I'd call out as a shining beacon of sanity, enabled by the best most popular code distribution format ever conceived: source distribution, coupled with a well defined not-completely-totally-batshit algorithm for looking for said sources when a program runs & at runtimes goes off to find it's dependencies: http://nodejs.org/docs/latest/api/modules.html#modules_all_t...

And trust me, libtool is replaceable, all it takes an agreement about a compiler and loader flag for producing shared libraries and you suddenly don't need it at all.

That's why things like libtool or autoconf evolved (or, better, were "iteratively designed") to be able to grow and encompass varying and different goals.

We understand the pitfalls and complications, and can come up with some solutions on how to handle them in a forward-thinking way.

If things break in the future, we can revisit this then.

In the meantime, libtool is one of the most ridiculous time-sinks I've ever had the displeasure of working with.

http://publib.boulder.ibm.com/infocenter/comphelp/v7v91/inde...

Anyway, I get the impression that for the original example of libtool, cathedral vs bazaar is the wrong question to be asking. [1] The problem isn't one of design or implementation. It seems a problem of distribution. I posit you could build a simplified, cross-platform library linker via either a centralised or distributed process. The question is rather, could you get everyone to agree to actually use it? Considering the success of clang, I think you'd stand a reasonable chance. (speaking of which, clang doesn't exactly stand out as a pure-bred bazaar-model example; neither is GCC for that matter)

Actually, I think that's what's wrong with this whole thread and the article that triggers it. I'm fed up as anyone with the layers of crap that we're building our software on. I seem to spend most of my time yak shaving because of some legacy decision, not contributing new stuff. In fact, my main output seems to be adapters for crappy interfaces. So while I find myself nodding along with the original essay, I think cathedral vs bazaar has nothing to do with the quality of the code we use all the time. It's a question of having the resources and balls to replace the layers of crap when necessary instead of carrying on piling more of them. Whether or not that's possible is mainly a question of whether the systems you absolutely need to retain are open or not. (and open source software can still be developed cathedral-style - I'm pretty sure ESR considered GNU to be cathedral software)

As an aside, when was the last time the format of .a/.so/.dylib/.lib/.dll files drastically changed on the respective operating system? We change even prevailing processor architecture a lot more often than that!

[1] I'll have to admit ignorance on the specific detail on the problem that libtool solves (inelegantly). I hope that doesn't derail my argument too much.

Of course not! We should try to get as much of what we are doing right. We also must realize we are not as clever as we think we are (or would like to be). Cathedrals are monuments to imaginary deities. As such, they aren't subject to the laws of reality. Our efforts should, OTOH, be guided by what's real and measurable and take into account that what we "know" but can't measure is nothing more than an educated guess.

My fear is that the awful hack-on-hack-ad-infinitum we do have at least binds us together, even if it is in misery. I'm not at all uncomfortable with replacing libtool, and I'm certain a very good job could be made. But I can hear 18 m68k afficianados crying out in pain that the new toolchain doesn't work on their NetBSD system, or the guy who wrote his own C compiler that needs some extra rules feeling totally shafted because now he was to write M4 and "M5," which for some reason he cannot stand any more than M4 even thought it looks nothing alike.

We stand unified. Not by good things. But by the woe we all suffer. To those out there seeking to crush that & make better, I wish you the best! Please keep in mind a simple mantra as you go about inventing the future, do no harm. Libtool maybe awful but it's how distributions are made.

phkamp, sorry for missing the reply. M4 is ugly, libtool takes forever, everything in it is a hack. It's proven to be an at least adequately flexible hack that has kept *nix unified, more or less. I respect that, but I'm not a systems programmer that gets burned by it on a regular basis either, and I don't mind that my OpenWRT compiles of all packages take 32 hours and it's all libtool's fault.

A lot of the problems are due to poor integration between languages, so for example the JVM people have reimplemented almost everything, as have the C++ people.

I'm not sure I understand how the Peter Principle applies here? Autoconf seems a rational solution in the economic sense: to standardise Unix, you need to have lots of influence to effect buy-in, whereas to write autoconf, you need a big dose of hacking talent. If you have the latter and not the former...

Agree, but once Linux became the dominant Unix-like, the major Linux distros like Debian and Redhat were probably in a position to replace uses of libtool in upstreams with a distro-wide standard for ld flags.

If anyone ever managed to write a genuinely better build system, the bazaar would let it rise to the top; the gradual rise of e.g. cmake is testament to this. Trying to impose one solution top-down, e.g. the LSB standardization of RPM, has a far worse track record than letting the bazaar do its thing.

As for doing its job well, the failure mode of configure "you can't build this" is abysmal. Just give me a fucking Makefile, I'll fix it myself. I love packages that come with Makefiles that don't work. I pop "-I/usr/local/include" into CFLAGS, run make again, and boom. Done. Trying to do the same with configure? Forget about it. --with-include-dir or whatever doesn't work, because it's really running some bogus test in the background which expects /bin/bash to exist and so on and so forth.

Sounds like the bazaar in action. I hope they succeed, but I know two gentoo projects that tried to do the same thing and were eventually abandoned as unworkable.

>As for doing its job well, the failure mode of configure "you can't build this" is abysmal. Just give me a fucking Makefile, I'll fix it myself. I love packages that come with Makefiles that don't work. I pop "-I/usr/local/include" into CFLAGS, run make again, and boom. Done. Trying to do the same with configure? Forget about it. --with-include-dir or whatever doesn't work, because it's really running some bogus test in the background which expects /bin/bash to exist and so on and so forth.

Sounds like you know make better than you know autoconf; I find it easier to fix autoconf problems on the rare occasions when they fail.