If you were involved in the OpenGL ES specification, you are an idiot

http://www.jwz.org/blog/2012/06/i-have-ported-xscreensaver-to-the-iphone/

by cpeterso • 14 years ago

316 213 14 years ago

If you use green text on a black background you are an idiot.

tzs2• 14 years ago

Green text on a black background is very readable in normal lighting, yet doesn't assault the eyes if you are reading in a darkened room.

In addition, there are still people using CRTs. Green on black is often clearer than white on black or black on white, especially in a small font, on CRTs because it is essentially monochrome, and so cannot suffer from fuzziness due to color misalignment.

lucian3033• 14 years ago

Unrelated but interesting. I always wondered if it was just nostalgia, preference, or some other reason for green text. It makes sense with RGB pixels of course. Now what about orange? That's not monochrome ...

tptacek• 14 years ago

FWIW, greenscreen (and amber) CRTs used green and amber phosphors; color alignment didn't come into it, right? As I understand it, green and amber because they were cheap.

hermanhermitage6• 14 years ago

OpenGL ES is not designed to be OpenGL.

It has a different set of constraints. The point is to prune back the API for small devices - NOT - to make make migration of legacy code simple.

For sure it would be nice if it came with a client side library to emulate OpenGL to assist in migration where people dont care about foot print size or perf.

JWZ is a very smart guy and I respect his opinion, but he is coming from a narrow viewpoint and not considering the wider implications.

In my experience, backwards compatible APIs and languages are what makes development a pain going forward. This is not to say backwards compatibility should not be provided in some form - but ejecting it from the core is a sane decision.

Otherwise APIs and languages expand at an unfathomable rate. Imagine if every API or language you ever used had features both added and removed over time to make it a better language. Javascript without the bad parts for example.

An incremental only approach to design is non-design in my view.

Evolution both promotes and retires ideas.

StavrosK2• 14 years ago

I'm not knowledgeable about OpenGL at all, but how hard would it be to write a compatibility layer so older apps continue to work? It could be released as a third party shim.

drivebyacct2• 14 years ago

Isn't that precisely what he did here? (note, I know nothing about nothing when it comes to graphics/OpenGL stuff)

greggman7• 14 years ago

http://code.google.com/p/gles2-bc/

ktizo2• 14 years ago

An incremental only approach to design is non-design in my view.

Nearly all design is incremental. That's how design works. That's partly why chairs are still recognizably chairs, and other useful stuff like that.

hermanhermitage6• 14 years ago

Quite true. But there is another word there "only".

katabatic• 14 years ago

Core APIs and languages do not expand at an "unfathomable rate". How long as the Berkeley sockets API been with us? TCP/IP? Twos-complement arithmetic? Do you honestly think that those are going to go away for the sake of some vaguely hand-waved "wider implications" and "idea promotion"?

OpenGL has been, like it or not, the only open, widely-adopted, non-proprietary 3D graphics API around for quite some time now. Enabling it on mobile devices wasn't exactly a sea-change requiring tossing all compatibility with the past in order to make progress (especially not as mobile GPUs continue to get more powerful).

jwz's point was that this could have very simply been included as an optional compatibility layer, which he then went and did.

[edited to put in the "not" in the first sentence that my fingers skipped over, which kinda changed the whole argument]

Ralith8• 14 years ago

But OpenGL ES does not enable OpenGL on mobile devices; that's the entire point of its existence. It enables OpenGL ES, which is intentionally designed to be a simplified subset.

If mobile device manufacturers feel that full-on OpenGL is appropriate for their device, then they are free to implement full-on OpenGL. JWZ should be complaining to the manufacturer, not the spec authors.

I believe that on modern hardware, OpenGL proper is simply OpenGL ES style features (and then some) with a software compatibility layer.

MrFoof• 14 years ago

>It has a different set of constraints. The point is to prune back the API for small devices

I came here to say something along these lines, but right now I'm limited on time so I don't have time to go into specific examples.

In OpenGL, there might be 5 ways to do something. Three of them are very much suboptimal, one way worked but was incredibly kludgy to write, and one was performant and pretty clean.

With OpenGL ES, they got rid of all the suboptimal and kludgy methods. The benefit today is if you write an OpenGL ES application, porting it to OpenGL nowadays is pretty easy. The other way around? Yes, that can be tremendously difficult. Honestly, I wish a lot more in OpenGL 3.x and 4.x was deprecated. Working with ES and the reduced extension hell is a big step up from the mess the full OpenGL API can be.

CamperBob2• 14 years ago

The point is to prune back the API for small devices

I don't disagree with the general sentiment that it was time to clean out the cruft in OpenGL, but I find this part of the argument to be a bit humorous. These "small, constrained devices" we're talking about are probably 10x faster than a goddamned Reality Engine.

I tend to agree that the API should have been renamed entirely once it was pared down this far, as 3Dfx did when they created Glide.

mcbridematt• 14 years ago

Now that smartphones and tablets have respectable GPUs in them, is there any reason why they shouldn't implement the full OpenGL spec?

mattrobertso• 14 years ago

I thought the issue with immediate mode which prevented its inclusion (in ES) was that immediate mode is very inefficient for the CPU, resulting in increased battery drain on smartphones and tablets.

msbarnett12• 14 years ago

> Now that smartphones and tablets have respectable GPUs in them, is there any reason why they shouldn't implement the full OpenGL spec?

To what benefit?

OpenGL ES is basically OpenGL minus all of the bits you really really should have stopped using over a decade ago. Originally all of that crap was culled out because it was only realistic to write new software for such resource constrained devices anyways, so why burden driver authors and hardware with the need to support crap that should never be used anyways?

Now that mobile device CPU/GPUs are powerful enough to start being appealing as targets for porting OpenGL-based applications, I think the proper response is less "great, slather back on all of the deprecated legacy cruft from the desktop version of OpenGL" and more "for the love of god update your rendering pipeline to reflect the last 15 years of progress".

jes5199• 14 years ago

> In my experience, backwards compatible APIs and languages are what makes development a pain going forward

That's the exact opposite of my experience! Libraries that are constantly changing their APIs produce the vast majority of my work.

cpeterso2• 14 years ago

I think hermanhermitage meant that backwards compatibility makes development painful for the library developer.

mtgx• 14 years ago

Speaking of which, is OpenGL ES 3.0 supposed to launch this year?

codedivine• 14 years ago

Much respect to JWZ, but I don't agree here. I am not even sure what to say, except that he is wrong, so I will just leave it there. And no, I was not involved in anyway with GL ES specification.

phillmv• 14 years ago

The above comment is not particularly useful, interesting or relevant. Neither is this one.

qdog• 14 years ago

People that don't study history are doomed to repeat it, or something like that. There's pretty much no reason to drop the old API calls, if you don't want it clogging stuff up, fine, make a separate library you have to throw a flag for or something.

Favorite quote: "If there are old features that you would like to discourage the use of, then you mark them as obsolete -- but you do not remove them because thou shalt not break working code.

If you don't agree with that, then please, get out of the software industry right now. Find another line of work. Please. "

Ralith8• 14 years ago

This indicates a fundemental misunderstanding of OpenGL ES. It cannot break working code by not having OpenGL 1.0 features, because OpenGL ES is not a revision to or successor of OpenGL 1.0.

Note that OpenGL 4, which is a (distant) successor to OpenGL 1, does include OpenGL 1s features, marked as deprecated.

zokier2• 14 years ago

OpenGL > 3.1 does not require implementation of compatibility profile, which includes deprecated funtionality. OSX doesn't.

hoodoof• 14 years ago

Ruthless dropping of backward compatibility is a good thing in certain circumstances. I'm not saying this blog post is one of those cases, but for example Microsoft Windows pays a huge price in complexity, cruft, bloat and performance to obsessively maintain backward compatibility. Python 3 has dropped backward compatibility and experiencing significant pain for the decision but after getting to Python 3 it looks arguably like a better, cleaner, more consistent language.

lucian3033• 14 years ago

Amen!

quux2• 14 years ago

:) Back in 2008 I helped JWZ port daliclock to the iPhone and once that was working I proposed a port of xscreensaver. He couldn't imagine ever wanting screensavers on an iPhone at the time. Guess he changed his mind.

I also was the one that opened his eyes to the differences between OpenGL and OpenGL ES it seems. I apologize.

SeanLuke2• 14 years ago

re: your username. Like retired basketball jerseys, shouldn't "quux" be reserved for Guy Steele?

quux2• 14 years ago

Hmm, I didn't know the he uses quux too. I claim independent discovery.

opminion• 14 years ago

This rant should probably say "my iPhone could run a full OpenGL implementation, and instead I'm provided a subset. I proved that it could run full OpenGL by writing most of the missing parts".

"(OpenGL ES) is a subset of the OpenGL" according to http://en.wikipedia.org/wiki/OpenGL_ES

Ralith8• 14 years ago

Was there ever any question over whether an iPhone could run a full OpenGL implementation?

Perhaps the people JWZ should be complaining about are those who decided to provide OpenGL ES instead of OpenGL proper on the iPhone in the first place.

recoiledsnake• 14 years ago

Related interesting read if you haven't, the brouhaha a few years ago over OpenGL 3.0 preferring to support CAD/CAM programs over games.

http://www.tomshardware.com/reviews/opengl-directx,2019.html

Ralith8• 14 years ago

I've heard that since OpenGL 3.2 most of the issues that caused that drama got resolved. Does anyone know details?

Edit: Reading parent's link, it looks like there were two complaints: First, OpenGL 1 era features haven't been completely removed--merely marked deprecated, as JWZ loudly advocates. Second (and this took up most of the article) OpenGL 3.0 was not feature-competitive with DX11, which is a bit silly, given that DX11 wasn't available at the time. As of the present day, and as far as I know for most of the recent past, the latest version of OpenGL has maintained feature parity with current hardware and therefore DirectX.

rachelbythebay• 14 years ago

One committee, ten years. One jwz, three days.

This is the stuff "10X" is made of.

Ralith8• 14 years ago

You're comparing completely incomparable tasks. JWZ did nothing but write a thin wrapper. The committee didn't omit such a wrapper because they couldn't accomplish it, they ommitted it as a deliberate design decision.

nvoorhies• 14 years ago

Exactly. Forcing every embedded device that wants to do portable 3D to include megabytes of wrappers for over a decade of cruft wasn't worth it, when the alternative is people having to spend a few days' time if they can't find a suitable shim already written somewhere.

roel_v5• 14 years ago

It took NASA several decades to build a rocket to the moon, I build a Diet Code & Menthos rocket in 30 minutes. Guess I'm 1000 times the space scientist those moron at NASA are rollseyes.

protomyth• 14 years ago

He probably won't be very happy to know OpenGL ES 2.0 is not backward compatible with OpenGL ES 1.1.

lnanek22• 14 years ago

It's common for hardware that supports the newer version to have a driver to support the older versions as well, though. Sort of just like the article did, but faking fixed function hardware using programmable hardware.

orangecat• 14 years ago

I wasn't. I can understand getting rid of immediate mode; as jwz demonstrated it's not that hard to rewrite glBegin/glVertex code to use vertex buffers. But 2.0 has an absurdly steep learning curve if this tutorial is accurate: http://developer.android.com/resources/tutorials/opengl/open.... Multiple custom shaders to draw a single triangle, really?

TazeTSchnitzel5• 14 years ago

The first hurdle is a little steep, yes. You need a shader to project your points, and a shader to texture your triangles.

But once you have those, it can scale to far more triangles.

Thing is without shaders it's still fixed-function, wich is bad.

AshleysBrain2• 14 years ago

My understanding is OpenGL ES is intentionally designed to be a subset of OpenGL so mobile devices can be simpler and therefore cheaper. It shouldn't be surprising mobile hardware and software is more limited. If deleting 80% of the platform features made your iPhone $50 cheaper and last another hour on battery, surely that's worth it?

Also, I believe the reason glVertex was removed is because it is a very inefficient way to draw. It wastes a lot of CPU time in "jump in to glVertex function, do a tiny amount of work, jump out of function, jump in to glVertex function, do a tiny amount of work...". It's to the extent that glDrawArrays is always faster. Take a buffer, fill it with data as fast as memory can transfer it, then a single function call to send all the data in one go, resulting in negligable overhead. Interesting anecdote: at least in my work in 2D games, modern GPUs are so fast they can render a quad faster than you can write its vertices to a vertex buffer on the CPU. So performance is limited by the CPU overhead! So this really makes a huge difference to the performance of some applications. And on mobile performance is usually more of an issue. So by removing the old functions, you're forced to do it in a more efficient way which may boost your framerate. Not so bad, huh? Unless, of course, you write a compatibility layer on top, which will reverse the performance gains.

Another reason there isn't a compatibility layer is it isn't OpenGL's job: it's supposed to be a super thin layer on the hardware so you can interface to the capabilities of the hardware as efficiently as possible. I also expect writing a compatibility layer that is standards-compliant in the general case is extremely difficult - check out all the effort that went in to ANGLE, for example.

So I think it's just a misunderstanding of the purpose and design of the tools. In future, I guess porting from OpenGL ES to desktop OpenGL would be a lot easier. 0.02

mwd_• 14 years ago

Part of the story here is that OpenGL ES isn't a new version of OpenGL, it's a different API for different platforms. It's nice if you can take old code and make it run on a mobile device easily but overall that's probably a small consideration compared to having an clean API that performs well.

If we want to focus on immediate mode specifically, it arguably shouldn't be used even in regular old OpenGL. It's slower and results in overly verbose code. Most people seem to use immediate mode because that's what most of the tutorial examples use.

Maybe one way to tackle the switch from immediate mode -> vertex arrays (and maybe quads -> triangles) is just to make some macros that take old code blocks and generate new ones.

AshleysBrain2• 14 years ago

Are you sure it's fair to call it a different API? OpenGL ES is officially described as "well-defined subsets of desktop OpenGL" [1], which would suggest it has a great deal in common with desktop OpenGL.

[1] http://www.khronos.org/opengles/

nikic• 14 years ago

I don't know much about OpenGL ES in particular, but backwards compatibility arguments are very tricky in general.

On the one hand you have all the people maintaining legacy apps and not interested in improving their code. They'll scream at you for breaking BC. On the other hand you have other people complaining that the library sucks due to all the old stuff and why they don't just remove all the crap.

I noticed this in particularly in the context of PHP. People really hate some parts of the language (for good!) and commonly demand a big BC breaking release that fixes all the bad parts. But every time something is fixed (obviously breaking BC in some way) there is a big outcry about wtf the developers have been thinking and whether they are all braindead - well, the usual stuff.

So really, before you start calling people idiots because they didn't keep comparability with an older version (or here even a completely different version), think again. There probably was a lot thought put into the decision. It's not like people just say "Oh, let's drop this, just so everyone can change his code!"

Bjartr• 14 years ago

Wasn't the whole point of the reduced API surface area in ES so that implementors only had to write the code that actually interfaces to the graphics hardware and let higher level libraries and engines deal with abstracting it appropriately for the task at hand?

kmm• 14 years ago

>If there are old features that you would like to discourage the use of, then you mark them as obsolete -- but you do not remove them because thou shalt not break working code. > If you don't agree with that, then please, get out of the software industry right now. Find another line of work. Please.

Arrogant and wrong. There was no working code in the first place, OpenGL ES is not OpenGL.

I, for one, am very glad with the changes the OpenGL board made. The glBegin/glEnd combo is so inane it should have been removed ages ago. Maintaining all these outdated codepaths carries its cost and clutters up documentation.

dkarl• 14 years ago

I don't think xscreensaver is a good example of valuable OpenGL 1.3 code that users want running on the iPhone. I think a person ought to have such an example before they start calling people idiots. Not to say that porting xscreensaver to the iPhone isn't cool, but it's only cool because it is anachronistic and whimsically impractical, like playing GameBoy games in a vintage arcade chassis in 2012.

There are a few things I couldn't figure out how to implement:

This after a year of percolation and three days (plus a few hours) of intense work by jwz. Given jwz's reputation as an insanely good programmer, I think this says a lot. As in, there's a nontrivial amount of work to do to claim near-compatibility with OpenGL 1.3, and an undetermined amount more work to claim true compatibility. Plus there may be more functionality from 1.4 and 1.5 to claim backwards compatibility with all of OpenGL 1.x. And all that to grant a (deservedly) dead desktop API new life on a mobile platform. No wonder they didn't bother.

Ralith8• 14 years ago

The thing that particularly bothers me about this post:

> People defend this decision by saying that they "had" to do it, because the fixed function pipeline is terribly inefficient on modern GPUs or some such nonsense. These people don't know what they're talking about, because the contour of the API has absolutely fuck-all to do with what goes over the wire.

As far as I can tell from what he's written, all he's done is forcibly enact the same inefficiencies that the paradigm shift away from immediate-mode rendering was intended to eliminate. The contour of the API in this case has a great deal to do with what goes over the wire: most critically, correctly-designed OpenGL ES programs will not retransmit static geometry to the GPU every single frame.

koide2• 14 years ago

And what I've understood he's done is not to retransmit static geometry every single frame, but to use an array to send batches behind the scenes.

This is, the exposed API doesn't force any particular implementation.

And that makes a lot of sense.

Ralith8• 14 years ago

If he's reimplemented the OpenGL 1 API, then he must be retransmitting geometry every frame, because the API is not sufficiently expressive to allow for retained state of that nature. He describes glBegin and glEnd as accumulating an array to batch out, but that batch gets re-accumulated and transmitted every single time the glBegin/End block is executed, i.e. every single frame.

syncopate• 14 years ago

Just some numbers, based on my hardware:

Lion OS X install on disk: 7GB

iOS 5.0.1 install: 1.7GB

My MacBook's RAM: 8GB

iPod Touch: 512KB

MB's disk size: 100GB

iPod Touch: 8GB

So: Maybe they removed 80% of the API in general because the iPhone/iPod touch is only 10-20% of a normal computer? Something had to go.

fpgeek8• 14 years ago

jwz is complaining about OpenGL 1.3 support. It might be worth asking yourself what computers were like when OpenGL 1.3 was released (August 2001).

P.S. I'm guessing your iPod touch has 512MB of RAM, not 512KB.

CrazedGeek• 14 years ago

It actually has 128MB or 256MB of RAM: http://en.wikipedia.org/wiki/IPod_Touch#Models

marshray9• 14 years ago

It's not like there's this massive base of OpenGL code just waiting to be ported to embedded devices if there were just a few more API calls. Yes, there's glxgears, screensavers, and handful of open source games. You could probably get some old CAD programs to run on your Android.

But if we compare the massive number of newly written OpenGL ES applications to all the old OpenGL 1.x fixed pipeline apps, the latter seem insignificant.

As he so ably demonstrates, it's just not that hard to port the older apps either. This "All you engineers are idiots because I had to work THREE WHOLE DAYS to port this 20 year old code to an iPhone" just sounds childish and silly.

The OpenGL board made the right decision for ES.

malachismith• 14 years ago

You kind of miss the point. This is a philosophical argument ILLUSTRATED through OpenGL and a port. To quote, "thou shalt not break working code"

debacle• 14 years ago

If your architectures are completely different, you're going to have to break some fucking code. That's just the way it is.

marshray9• 14 years ago

No, I understand that point. I just don't agree with it in this case.

* "thou shalt not break working code". That's not some Universal Code of Software Engineering, that's just something he made up.

* No one in their right mind would expect you could port 20 year old C using OpenGL 1.x code to an iPhone with no effort.

* There are plenty of reasons APIs can benefit from breaking changes. Security fixes, major architectural improvements, platform porting, even general evolution and modernization. The question is always one of cost/benefit, relative pain vs relative gain.

* Sorry JWZ, but your X screensaver project is actually not an overriding concern driving the evolution of OpenGL ES.

* I didn't mention it again because so many other have pointed this out, but this isn't even the same API. OpenGL ES is not the same thing as OpenGL 1.x.

* Changing the major version number is a common and accepted way to indicate breaking changes are present in an API. The OpenGL board went even farther and gave it a different name and a different versioning scheme specifically because it is different.

rphlx• 14 years ago

That is a noble goal, but it's one of hundreds of often-conflicting noble goals. It's not inherently wrong to design a toaster that fails to work as a refrigerator.

semanticist2• 14 years ago

Except he didn't port his code to OpenGL ES - he ported the missing API from OpenGL 1.3 to OpenGL ES.

If it takes one developer three days to implement the missing API, then perhaps it suggests that this API wasn't in any way damaging to the new system and could have been left in place originally?

marshray9• 14 years ago

* It costs money to document.

* It costs money to implement. Three days for a lone coder, much more than that for a careful consistent development process.

* It costs money to develop regression and validation tests.

* It costs money to test, including collecting all the different hardware it needs to be tested on.

* It costs money to fix bugs in it.

* It costs money to answer questions from developers about it.

* These costs are multiplied by every future version that will continue to support it.

* These costs are multiplied by the many vendors that will be required to implement it.

* Some of these costs are multiplied by every language that will support bindings to the API.

* It costs code space in embedded devices.

* It costs as much as a vendor wants to spend trying to make it perform as fast as possible.

* It costs conceptual complexity in developers' heads.

* It costs developers when they inadvertently do things the old inefficient way.

* It costs reputation when somebody compares benchmarks of your product using the old inefficient way against another vendor's product using the new recommended way.

* Requirements to continually support it at the same level of performance constrain your choices when developing new hardware.

* Sometimes it costs precious die space which could be better used for other things.

* The inefficiency sometimes increases power consumption and reduces battery life.

* Users are disappointed if every version of new hardware doesn't run the old inefficient code even faster than the previous version.

So if it doesn't have a real need to be there in ES 1.0 then, yes, it is actively damaging and should be thrown out in the rewrite.

msbarnett12• 14 years ago

It didn't take him three days to port all of the cruft from OpenGL 1.3 to Open GL ES; it took him three days to port the precise calls and precise semantics he relied upon in OpenGL 1.3 to OpenGL ES, and even then he punted on some of it.

The latter is a far, far cry from the former. The former is basically what a modern OpenGL (non-ES) driver does to support all of the legacy cruft on modern hardware. Recall that the OpenGL API is an API for a state machine. All of those layers of cruft built up over the years have incredibly arcane state interactions, are a bitch to support, and are why modern OpenGL (non-ES) drivers have gotten rather quite bloated.

Edit: to tackle the second part of your comment:

> perhaps it suggests that this API wasn't in any way damaging to the new system and could have been left in place originally?

The Immediate Mode calls he complains about not being there? They were "removed" (never introduced) way back in OpenGL ES 1.0, which IIRC originally debuted in 2003.

It's a pretty big stretch to argue that because reimplementing Immediate Mode on top of OpenGL ES doesn't have a crushing performance penalty on 2012's iPhone, it wouldn't have been a problem on what passed for mobile hardware back in 2003.

(which is ignoring the fact that even if the performance back then wouldn't have been garbage, taking the opportunity to make breaking changes and throw out cruft for an API that wasn't expected to host legacy apps was absolutely the right move)

retrogradeorbit3• 14 years ago

And lets not forget he's also unwilling to help people out by uploading his OpenGL wrapper to github (or by extension any revision control system).

http://www.jwz.org/blog/2012/06/i-have-ported-xscreensaver-t...

greggman7• 14 years ago

He didn't implement OpenGL 1.3. He ported an OpenGL 1.3 app to OpenGL ES 2.0. There's a HUGE difference.

jlarocco• 14 years ago

This guy needs to get over himself. The world doesn't revolve around his pet project.

His rant is stupid for at least four reasons.

First, OpenGL ES is not regular OpenGL. If it were, it would just be called "OpenGL". It's a different thing, based on OpenGL, but targeted at mobile phones and "small" devices. It's almost like complaining that DirectX and OpenGL have different interfaces. They're different things, therefore they will be different. At the time of OpenGL ES's release there was no backwards compatibility to consider, because it was a new thing.

Second, few applications or games used the immediate mode drawing code in regular OpenGL. It's slow and inconvenient to use for the data formats used in real life. It might be great for the OpenGL equivalent of "Hello, World", but other than that nobody uses it.

Third, there was a lot of discussion about what should be in OpenGL ES, and he could have contributed his opinion when the spec was being drafted. Where was his outcry then, when he could have made an impact? Honestly, though, it probably would have been ignored, because immediate mode is so lame.

Fourth, immediate mode is deprecated even in regular desktop OpenGL, as of version 3.0.

astrodust5• 14 years ago

If you were porting GL-based screensavers that were written a decade ago you'd probably be feeling the same way. You don't want to rewrite code that already works on an earlier version of the spec.

TazeTSchnitzel5• 14 years ago

What earlier version? GL ES is NOT GL.

rhizome• 14 years ago

I bet people would have less of a problem with it if had been called "OpenGL SE" instead.

angersock3• 14 years ago

I respect jwz, but this is very much a step backwards, and while I appreciate his complaint about breaking existing APIs, this would be for programs near 20 years old targeted at a different platform.

OpenGL is a terrible, awful, crufty API, and the reason those methods were removed is that they are comically suboptimal. They do not reflect anything remotely like modern card capabilities, and their use directly causes harm to the ozone layer, kittens, and infants. I'm pretty sure that glBegin() gave a coworker cancer, and the matrix stack has claimed more lives than Kevorkian.

Building a shim to port over old OpenGL 1.3 apps is kind of like translating the Necronomicon into English--possible, of questionable utility, and likely to bring about insanity and demons.

As others have pointed out, OpenGL ES is not intended to be an extension of OpenGL--it was a chance to break out a lot of the dumb cruft that had accumulated into the API. Most of the features he's complaining about are either bad practice or should be gotten rid of entirely.

Compare the length of the API listings for GL 1.x, 2.x, and modern 3.x / 4.x. Remember that the whole thing is a hissing, clanking state machine, and that interactions between functions can be arcane--and threading presents additional issues.

Immediate mode rendering with glBegin()/glEnd()/glVertex()/glNormal/etc. is ugly. Any shim that collects that information still has non-trivial work stuffing it into a buffer, and the overhead of drawing anything with more than a few hundred triangles soon becomes absurd. Worse, this style of programming discourages storing geometry on the card, and that causes additional inefficiency--and trying to use those calls remotely over X causes all kinds of stupid as glx can barely do indirect rendering anyways.

Additionally, we have additional vertex stream attributes available now which are very flexible and don't map onto that anymore. It's time to let go.

tl,dr: jwz is complaining about a fork of an API that removed cruft people depended on, but the cruft needed removing. :(

fpgeek8• 14 years ago

jwz makes it clear that his deeper complaint is not about removing the cruft per se, it's about removing vast swaths of the API and then continuing to call it "OpenGL". If they'd called it "MobileGL" or some other nonsense, he probably wouldn't have liked the result (based on this rant), but he also wouldn't have complained that they broke working code.

EDIT: 'continuing to call it "OpenGL"' -> 'using "OpenGL" in the name at all' Effectively, he's complaining about a form of false advertising.

city41• 14 years ago

But it's not OpenGL. It's OpenGL ES. Windows 8 is also not Windows 95.

msbarnett12• 14 years ago

They didn't call it "OpenGL". They called it "OpenGL ES". The ES is a significant part of the name. The fact that the name happens in part to contain the word "OpenGL" is not a promise of source compatibility, and even the most cursory glances at the documentation would have made it clear that the API has a different name because it is a different API.

It's incredibly daft to argue that part of a name amounts to a promise of backwards compatibility with a different API, in perpetuity. The X11 protocol isn't backwards compatible with that of X9 simply because they both share a name-fragment.

bcowcher• 14 years ago

So OpenGL ES (Embedded systems) was not different enough? I actually think OpenGL ES didn't go far enough in the initial spec.

ES 1.1 still has alot of the old fixed function pipeline, but then in 2.0 they scrapped it all entirely, opting for a smaller, more modern API. That, to me, was a bad move. I think they should've made the shader based pipeline in 2.0 part of the 1.1 spec.

saurik2• 14 years ago

Right, but again: those are the parts of the API that are already discouraged and downright deprecated as of OpenGL 2.0. If you were developing an app targeting OpenGL 2.0, following what I understand are the best practices for that version of the specification, OpenGL ES will require little to no porting.

The problem then, arguably, is not that OpenGL ES is called OpenGL but that OpenGL 2.0 added an entirely unrelated pipeline to that used in OpenGL 1.3: you could then claim "how dare they reuse the name for what is now two APIs stuck into one library". However, they shared a lot of underlying conventions, and they were honest about bumping the major version number.

If anything, the fact that jwz is happy he proved--that you can build an OpenGL 1.3 emulation library over OpenGL ES--would argue to me to /not/ include the 1.x features as part of the standard, but to instead encourage third parties to distribute such libraries. The fixed function pipeline wasn't removed from the API as it is unimplementable, but because it is a ton of obsolete code that 2.x coders avoid anyway.

wanderr• 14 years ago

I'd be interested in seeing an argument explaining why these API calls smother puppies when the premise of the original article is that you can in fact offer them as an interface to the shiny new better way of doing things, without accidentally summoning cthulu. If he's wrong on that point I'd like to see a clear explanation of why.

tung• 14 years ago

I'm not an OpenGL expert, but my understanding is that the fixed function pipeline (FFP) is like a set of big, generic shaders and state that everybody had to go through to actually do the work of displaying 3D graphics. You could write shaders a fraction of the size that does just the work you need without having to pay the performance price for features you don't use. FFP-using code looks nice in tutorials, but performs terribly outside of demos, never mind the complexity tax it imposed for implementations as a 3D graphics layer for people who don't understand how 3D graphics works.

Also the old OpenGL API had immediate mode functions which encouraged people to trickle in interleaved data and operations; the exact opposite of what 3D APIs need to run fast.

angersock3• 14 years ago

Sure, here's my attempt at explanation:

OpenGL is a gigantic mess, one which only somewhat recently has started to get better. For those that don't know, it's lineage goes back to IrisGL and big-iron Silicon Graphics machines. There's a wonderful recap of its history on Stack Overflow ( http://programmers.stackexchange.com/questions/60544/why-do-... )--long story short, design-by-committee and squabbling vendors (especially the CAD folks, whom I until recently counted myself among) resulted in bloated, sad, crufty APIs.

Having to maintain a codebase to mimic old OpenGL functionality, especially when in some cases it wasn't particularly well-defined/standardized, in addition to coming up with a small profile for new features on embedded systems, would present a nontrivial burden on the driver and hardware writers. Hell, even Intel has only somewhat gotten it right recently--and they've had the OS community via Mesa do most of the work for them (as I understand it)!

These aren't features that are hugely important, these aren't features that are game changing, these are a lot of things that simply obsolete or unnecessary. jwz laments the lack of quads support, so let's start there:

OpenGL 1.x supported the following primitive types: points, lines, line strips, line loops, triangles, triangles strips, triangle fans, quads. quad strips, polygons (see http://www.opentk.com/doc/chapter/2/opengl/geometry/primitiv... for examples). Several of these options are quite redundant, and supporting them is not really helpful. Moreover, several of them present interesting questions for a driver writer: what is the preferred way of decomposing quads or polygons? Strips? Fans? Discrete triangles again?

Sphere mapping has, I believe, been replaced with cube mapping. OpenGL ES 1.1 has cube mapping as an extension, but I don't know if Apple decided to implement it or not--such is part of the evil of OpenGL, this use of extensions.

1D texturing (and 3D texturing) were omitted, again presumably to make implementors' lives easier. To face this, fill a whole 2D texture with a gradient, and clamp on the edges when sampling (glTexEnvi I think should do this...?). Hopefully that would work. Only recently have 1D and 3D textures gotten really useful, for clever tricks in passing LUTs and such to the programmable shader pipeline; I think the older use for them was ghetto cel-shading and palette mapping--cool but not critical.

Anyways, the problem with requiring that the library writers support all that is again that they would have to create most of the OpenGL environment (which is terrible), and then map it onto their new environment (even more terrible), as well as develop the new environment. This is nuts.

It's similar to asking if people could write a portability layer atop Win32 to support Win16 to support old DOS system calls--anyone can do a subset of that and complain that "Hey, it's easy!" but to do it right (and you must do it right, or else somebody else will complain!) is very nontrivial.

For a more timely example, consider the issues folks have had getting people to move on to Python 3--and contrast that with what the Rubyists have accomplished by just moving fast and fixing things as they break.

Or think about the amount of time/money spent on keeping the COBOL infrastructure up, or supporting legacy VB6 installations.

Honestly, sometimes we should applaud vendors for Doing the Right Thing and trying to force users into fixing outdated code.

c3d2• 14 years ago

By the same reasoning, we should remove printf from libc. It's a terrible, awful, crufty API with threading issues, and it's overhead on modern windowed systems is just horrible.

angersock3• 14 years ago

Come now, be reasonable.

If we released some sort of libc for embedded systems, specifying only fprintf(), your analogy would be valid.

gaius5• 14 years ago

OpenGL is a terrible, awful, crufty API, and the reason those methods were removed is that they are comically suboptimal. They do not reflect anything remotely like modern card capabilities

But the guys who wrote it weren't idiots. They were in fact super smart engineers working at the cutting edge company of the day, SGI. And they made a philosophical call, which was that OpenGL should be an abstraction of geometry and an idealized rendering pipeline with just enough hardware-specific hackery in it to make it perform[1]. They did the best they could operating under the constraints of the state-of-the-art of the time and the resources they had available. And people still use OpenGL, decades later, and have done amazing things with it.

Isaac Newton said "If I have seen further, it is by standing on the shoulders of giants". Kids these days, talking about legacy technologies, would be wise to remember that.

[1] Whereas Microsoft believed in an abstraction of physical hardware, with just enough geometry in to make it useful.

alexlarsson• 14 years ago

Thats not at all true. The original OpenGL API is very much a direct mapping of the original SGI graphics hardware. Most openGL calls on SGI are single cpu instructions feeding the data to the hardware, which fully implements the whole OpenGL state machine.

Its just that modern GPUs work in completely different ways, so this kind of API is useless for them.

hesdeadjim• 14 years ago

God what flamebait, how is this near the top of the front page?

When your primary argument that the ES designers were idiots is lack of immediate mode, I'm sorry, you are the idiot. These are embedded systems with highly constrained resources and that immediate mode API is horrible for a lot of reasons:

* Requires tons of driver calls.

* Stupidly hard to optimize on the driver side when you have no idea just how many vertices or other per-vertex data are to follow your call to glBegin.

* Trivial to replace with a much better, and much, much higher performing vertex representation either through vertex buffer objects or simple calls to glVertexPointer/etc.

* Teaches beginners the Wrong Way of doing things -- you won't use this API for anything beyond a toy program as the last thing you would do is load an exported mesh from Max or Maya and then iterate through every vertex.

Having learned OpenGL initially with the ES 1.1/2.0 spec, then transitioning back to the desktop version, I couldn't believe how bloated the API had gotten. There is a reason they want to deprecate most of it and move to a spec that is similar to ES in its simplicity.

ecaron• 14 years ago

> God what flamebait, how is this near the top of the front page?

Because the source is widely respected and has an amazing demonstrated grasp of what good programming involves.

greggman7• 14 years ago

Yes but in this case he's an idiot. Try asking Carmack. Any GPU programmer knows GL 1.x was utter crap. Good riddance.

roel_v5• 14 years ago

Eh, not any more. JWZ is a has-been, who admittedly did good things back in the 90's, but who has lost touch with the industry when he became a hobby coder. This post is a clear demonstration of that - if he knew even a little bit what he was talking about, he wouldn't make the outrageous claims he does. But no, he just want to hang on to his toys he wrote 15 years ago and have them work with just a recompile on platforms that don't even remotely resemble the platforms he wrote them for back then. That's just absurd. And yes, I too complain (loudly) when I have to change code to accommodate changes in the platform (like how VS12 doesn't support WinXP) but in the end, deep down, I recognize that that's just how it has to go, and so should somebody who knows (or should know) much better than me, like JWZ.

anonymous• 14 years ago

> These are embedded systems with highly constrained resources Gigahertz quadcores with several gigabytes of RAM is "constrained resources" nowadays? I agree with you though - it's flamebait. But it's also true.

potkor• 14 years ago

These "embedded systems with highly constrained resources" are machines with 512MB+ of memory and monster CPU/GPUs. It's perfectly OK to write code for them that you haven't bled over to optimize the hell out of.

And JWZ just showed you don't need "tons of driver calls" unless you mean simple function calls that don't cross the kernel boundary.

lucian3033• 14 years ago

A spec that's not backwards compatible. Oh my! I've never seen that before. Get over it. Some software isn't backwards compatible. OpenGL ES != OpenGL. Get over it.

jiggy2011• 14 years ago

Next Week: JWZ tries to port xscreensaver to webgl.

K2h• 14 years ago

<As with all things, the first 90% took the first 90% of the time, and then the second 90% took the second 90% of the time.

That math is totally consistent with my experience.

Anm• 14 years ago

Stealing this quote:

"As with all things, the first 90% took the first 90% of the time, and then the second 90% took the second 90% of the time."

derpmeister2• 14 years ago

Which is of course a thinking error, he meant: "As with all things, the first 90% took the first 90% of the time, and then the remaining 10% took the second 90% of the time.".

roel_v5• 14 years ago

Sorry to be that guy, but that quote is over 25 years old - http://en.wikipedia.org/wiki/Ninety-ninety_rule .

dbbolton• 14 years ago

All I have to say about this post is "Thank goodness for Readable."

Groxx• 14 years ago

So, what happens when all this less efficient drawing code now drains the batteries of devices more quickly?

Good effort and all, and it's pretty cool that you can do this, but why all this work to avoid improving something?

fatjokes• 14 years ago

I'm home sick right now with a mild headache and dizziness. I thought it wouldn't hurt to browse hacker news for a bit. Then that black/green color scheme hit me.

willvarfar• 14 years ago

JWZ is so right.

The whole "FFP is slow" argument is completely wrong.

Most graphics drivers are broken and inefficient. Its just a lack of decent dedicated programming effort coupled with over-management at NVidia etc.

My own brush with Khronos group back when this was being done still makes me shudder.

http://williamedwardscoder.tumblr.com/post/14011115100/openg...

TazeTSchnitzel5• 14 years ago

Yeah, graphics drivers are broken and inefficient, and one reason is they have to implement ALL of the LEGACY, DEPRECATED SPEC in SOFTWARE, because some people's code could break.

gcb• 14 years ago

someone send that post to the gnome lists... hopefully they will be using xfce and hence be able to read it. Then just hope they get the analogy

immy• 14 years ago

jwz, how about porting the ~Aaron extension to Mac OS X now. Too bad SGI isn't around to hold a funeral like Apple did for OS 9 :)

haxe2• 14 years ago

Wow ... this guy is just an arrogant idiot ...

You first need to understand that opengl specification are not made to be easy to use, but efficient. And the api design is highly constraint by the hardware. You also need to understand that opengl 1.3 is FCKING 11 YEARS OLD !

Back in the day you had immediate mode. You first call glBegin, then make one call to glVertex for each vertex and finally call glEnd. So if you want to render one frame of a cube, that's 1+6*4+1=26 calls, and for each call your cpu send data to the graphic card. There are two huge problems with that : you need to the transfer the same vertex data to the graphic card for each frame and your cpu processing power limit your graphic card processing power (because cpu calls take way more time than your graphic card takes to render a vertex).

Why was it designed like that in the first place if it's inefficient ? Easy : there were no graphic card. All the 3D rendering was done by the cpu.

So with the development of more powerful graphic card, they updated the specification. With the new api you allocate memory buffers of vertex directly on your graphic card memory and you just need to call one function to render the whole buffer. So you don't need to transfer all the vertex data at each frame, and you don't need to do gazillion of function calls to render a cube. THIS WAS IN 2004 !

So when the khronos consortium (just a reminder, the "idiots" are : Apple, AMD/ATI, Nvidia, Intel, Google, Id Software, ...) decided in 2007 to make a mobile version (ES stand for embedded system) of opengl, did they choose to base their api on something conceived in 1992 and dropped in 2004 ? OF COURSE NOT !

But why not keep it compatible ? Well it's a waste of time. Developers shouldn't use deprecated functionality since 2004 in 2012, especially when it's highly inefficient.

This guy need to grow up, understand that opengl ES main objective is not screensavers, understand that he sometimes need to update a little bit his programming knowledge and acknowledge that he's not an expert in the very complex field of gpu.

tldr; he's the idiot

gaius5• 14 years ago

Easy : there were no graphic card. All the 3D rendering was done by the cpu.

Err, what? SGI, who wrote the OpenGL spec, had rendering hardware before OpenGL existed...

haxe2• 14 years ago

Yeah you're right, they were graphic cards, my bad, I made a shortcut. But at the time (1992), opengl implementations were mostly on the cpu (except for SGI graphic cards that were reserved to professionals). And the graphic card processing power was not enough to make the cpu a bottleneck.

the_mat• 14 years ago

You missed a key point: with OpenGL 1.3 you don't need to follow the glBegin/glVertex/glEnd paradigm. You can create a display list or you can use vertex arrays. In fact the reason for including display lists in the first place was to make it so you didn't have to send one vertex at a time.

I don't agree with JWZ, regardless. OpenGL ES was a good simplification to the overall system. But he is right in that an optional compatibility layer that lives on top of OpenGL ES would have kept the original API valid for those who need it (and without slowing down the core of OpenGL ES).

forgottenpaswrd2• 14 years ago

Maybe the idiot is jwz.

There was a change in the technology that needed a change in the API. Not a single GPU use a fixed pipeline anymore, so the API needs to change.

I see OpenGL as a testing platform that learned by trowing to the wall and learning what sticks. Companies create extensions and if people find it useful they became part of the language, but in any API there is a need for removing what is not used anymore or can be done much more efficient with new ways.

Maintaining code takes a lot of resources and is a pain in the ass to program, I can tell you(you have to emulate a lot of things that do not exist anymore, including BUGS or HACKS).

I could understand not being Apple extreme here, e.g look at the Apple TV and you only see new digital ports, but the other extreme, don't touch anything for 20 years because it works is equally non sense..

I remember a brand new computer that was hanged 5 seconds on the start up by a 1.4M floppy disk because the people who assembled the computer were afraid to remove the thing. I had to remove the thing, left a horrible hole and I discovered the machine had the floppy check hardwired in the BIOS!!!

You buy a computer today and it has parallel, serial and PS2 ports with an outdated BIOS with 1981 timings they can't change so they don't "break legacy standards".

You buy a computer in 2012 with a 1900x1080 pixels and the first thing you see is a horrible black screen with fuzzy letters.

Non sense. openGL needs to be way more clean that what it is now, where it is impossible to remove anything because some company of the consortium find the feature "essential" because they have some legacy code they are too lazy to update.

leothekim• 14 years ago

"I wrote this because you are all idiots."

Speaking as an idiot, why do I want to run a screensaver on my iPhone?

TazeTSchnitzel5• 14 years ago

You don't, especially an ancient one written for a deprecated graphics API optimised for SGI Graphics workstations.

exDM69• 14 years ago

disclaimer: I write OpenGL ES implementations for living. I occasionally deal with some of the guys writing (parts of) the GL specs.

This article is a load of bollocks. It actually isn't the first article I read that complains about the removal of immediate mode in GLES. This article, like the others I've read, leaves me with the impression that the author is rather clueless.

The article essentially complains that a feature from a 20 year old API should have been included in an API designed 10 years ago. The glBegin/glEnd API is a horrible mistake from the start and it should have never existed. It's a good thing it's removed from the GLES API. OpenGL 1.x and GLES 1.x are both deprecated, more than 10 years old and should not be used any more. In any case, he's complaining about mistakes(?) made more than 10 years ago.

One major flaw in jwz's reasoning is that something was removed. Immediate mode was indeed removed from the spec but an implementation of GLES w/ immediate mode never existed. It was dropped from the spec to avoid having the GLES implementers to waste time adding a legacy drawing API that is essentially useless and has awful performance.

GLES1 was a stripped down API for early mobile 3d applications that used rather primitive hardware or software.

There are lots of flaws in the GL(ES) API, but removal of immediate wasn't one of them. If anything, they should have fixed/broken the API more and intentionally destroy backward compatibility.

The the committees designing the API's are understaffed and have too much work on their hands. They're trying to make the interests of hardware manufacturers, content creators and OEM's match. Because of all these pressures, they're not really doing a great API but at least we have some kind of well-specified standard. Calling these people idiots doesn't help anyone.

With modern 3d API's, the vertex data is pushed into buffers in video memory. If you're writing 3d code in the 21st century, that's what you should do.

So a legacy feature that the author depended on was removed and rather than updating his code to run on modern software and hardware he re-implements it in software. He spends 3 days doing it, gets pissed off and writes a blog post.

ajuc• 14 years ago

JavaScript is also not "backward" compatible with Java, and their names are similiar.

sriram_malhar• 14 years ago

Speaking of idiocy, a screen saver for the iphone? Jeez!! What a waste of talent and battery.

I do agree with the sentiment that APIs should be deprecated over time.

paines• 14 years ago

This all makes me wonder, is it hindering to learn open gl even if you want to develeop prmarily for android/iphone. On one hand I would say: Yes it is hindering. On the other hand, it feels like a good idea to learn open gl, in order to now how and what a rendering pipeline is, gl state machine, etc. What do you guys think ?

If you were involved in the OpenGL ES specification, you are an idiot

Related Stories