LWN: Comments on "No Mir by default in Ubuntu 13.10"

No Mir by default in Ubuntu 13.10

robclark — 2013-10-06T18:42:37+00:00

> All the information needed is there (and always was there), you just refused to connect the dots.

Well, I guess I'm just not reading between the lines as hard as you are. I was looking for something a bit more concrete and a bit less conspiracy-theory.

All the same, my opinion remains the same: it isn't really a terribly good process tool for open source projects.

No Mir by default in Ubuntu 13.10

khim — 2013-10-06T16:08:09+00:00

well, write-access control to prevent a developer from 'git push origin bonghits' is different from read-access, ie "developers can't see each other work".

Well, yeah. One is relevant to Rietveld, another is not relevant.

I didn't get this impression from either the wikipedia article or the 'project background' link.

Really? Please read it again. This quickly turned into a fork as we added access control features that Guido van Rossum did not want to see complicating the Rietveld code base. Then again. …access control features … Rietveld … code base. Got that? NOOO? I'll explain.

Just what is Rietveld? It's collaborative code review tool. It can not read code from subversion or git (this is work of depot_tools), it can not commit code to the repo (this is work of commit queue), it's strictly and specifically review tool. Heck it's not even involved in ownership decisions—this is work for the presubmit scripts. Any changes which are complicating the Rietveld code base by necessity must affect patch review process—because Rietveld does not do anything else. Just what kind access control for patch review process can you imagine which does not affect patches viewability status? No, really? You may forbid certain users from adding the comments, but this change is pretty localized: you only need to check these things when new comments are added. You may add OWNERS, but this change does not affect Rietvield at all (it's enforced by presubmit scripts and commit queue). The only kind of change which may significantly perturb Rietvield's codebase is pervasive access control which determines who and when can watch patches which are in process of review. This will be invasive: this will mean that you need huge amount of checks spread over the codebase.

All the information needed is there (and always was there), you just refused to connect the dots.

one of my long standing complaints about the gerrit workflow (vs, send-patches-to-public-list approach, possibly augmented w/ patchwork) is that by default only the people you choose to review some patch(s) get notified.

That's the whole point of gerrit! It's raison d’être! It's designed to facilitate parallel development of “open source” Android's code and vendor's proprietary “secret sauce” in parallel thus such decision makes perfect sense.

Vs. everyone seeing the patches and having a chance to review/comment.

Oh yeah. Imagine that. What SAMSUNG will do if patches needed to support their next innovative “air scrolling” will be seen by HTC or SONY? This will be huge brawl. Not pretty at all. This is what gerrit is dealing with: companies who actively try to destroy each other—yet are forced to cooperate. “Everyone seeing the patches and having a chance to review/comment” does not work for that.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-06T15:31:15+00:00

> What kind of “reference” do you need??? It's not a secret that enforceable access control was the reason for the Gerrit creation. Of course later it have gotten other properties, too, but initially it was just a “Rietvield with access control”. Heck, it's in description of the project background: Gerrit Code Review started as a simple set of patches to Rietveld, and was originally built to service AOSP. This quickly turned into a fork as we added access control features that Guido van Rossum did not want to see complicating the Rietveld code base.

well, write-access control to prevent a developer from 'git push origin bonghits' is different from read-access, ie "developers can't see each other work". I was curious if read-access was actually a motivation as you at least seem to have implied. I didn't get this impression from either the wikipedia article or the 'project background' link.

one of my long standing complaints about the gerrit workflow (vs, send-patches-to-public-list approach, possibly augmented w/ patchwork) is that by default only the people you choose to review some patch(s) get notified. Vs. everyone seeing the patches and having a chance to review/comment.

No Mir by default in Ubuntu 13.10

khim — 2013-10-06T14:45:12+00:00

What kind of “reference” do you need??? It's not a secret that enforceable access control was the reason for the Gerrit creation. Of course later it have gotten other properties, too, but initially it was just a “Rietvield with access control”. Heck, it's in description of the project background: Gerrit Code Review started as a simple set of patches to Rietveld, and was originally built to service AOSP. This quickly turned into a fork as we added access control features that Guido van Rossum did not want to see complicating the Rietveld code base.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-06T14:20:07+00:00

> Android developed Gerrit to make sure different developers can't see each other work yet still can develop common components

a bit off topic, but that would explain why I find gerrit so aggravating to use. Is that really true? Do you have a reference for that?

-----

back on the main topic.. I think the short answer is success or failure of a phone ecosystem has more to do with other factors than development model. But the fact that there are so few successes makes it difficult to draw any sort of conclusion.

No Mir by default in Ubuntu 13.10

khim — 2013-10-06T13:51:47+00:00

I am not sure I can follow here. Either they shipped and failed or they were simply not used. Failure implies trying.

Of course they have tried! They have some code, bunch of repos, even some demos. Best case scenario: they managed to run their stuff on a couple of old smartphone models (usually with bunch of hardware non-functional). OpenMoko is in this camp, the only thing which places it in their own category is the fact that it had software developed and actually created specifically for OpenMoko.

Maybe, didn't followed them too closely. Doesn't change the fact that they failed while not being community developed.

Well, if “community” is so important then what can't it make these projects success? Are projects which were adopted by community are somehow more problematic then projects developed by “community” from the start?

What we can probably correlate is knowledge about number of prototype iterations with development process transparency.
A fully opque development process should in theory not allow to know that prototypes even existed, a fully transparent process should allow to see all stages, maybe even including experiements.

Hmm... So by now we know that opaque approach succeeds from time to time (even most iPhone suppliers had no idea that Apple does iPhone, Android developed Gerrit to make sure different developers can't see each other work yet still can develop common components, etc), while transparent approach fails (all “transparent” efforts are faileres. What does it say about “community” chances?

No Mir by default in Ubuntu 13.10

krake — 2013-10-06T13:04:22+00:00

> GPE, OPIE, Enlightenment and many, many others.

Ah, good to know. Didn't know about the first two and I thought Enlightenment was a desktop software project.

> Very few reached the stage where you can actually use them on the real hardware

I am not sure I can follow here. Either they shipped and failed or they were simply not used. Failure implies trying.

> These are developed by “community” nowadays, too.

Maybe, didn't followed them too closely. Doesn't change the fact that they failed while not being community developed.

There is no correlation and certainly no causation between single vendor vs. multivendor developed product and market success.

What we can probably correlate is knowledge about number of prototype iterations with development process transparency.
A fully opque development process should in theory not allow to know that prototypes even existed, a fully transparent process should allow to see all stages, maybe even including experiements.

No Mir by default in Ubuntu 13.10

khim — 2013-10-06T12:17:47+00:00

Dozends? OpenMoko and?

GPE, OPIE, Enlightenment and many, many others.

Very few reached the stage where you can actually use them on the real hardware (OpenMoko is rare success by the standards of phone projects developed by “community”), most died before reaching that stage.

plus a couple that were not community developed and failed, e.g. Maemo, WebOS

These are developed by “community” nowadays, too. As Mer and Open webOS. These, too, have about the same level of success as all other “community” projects.

No Mir by default in Ubuntu 13.10

krake — 2013-10-06T12:01:39+00:00

> XMir runs on Mir. It is a major achievement to replace Xorg with XMir.

It would be a major achievement if it had been achieved :)

You might have missed it but a short while back there was a huge flamefest because some XMir specific patches were removed from a Xorg driver.

Those patches wouldn't be necessary if XMIr ran on Mir due to Mir doing the actual input/output hardware/driver access.

No Mir by default in Ubuntu 13.10

krake — 2013-10-06T11:51:15+00:00

> That's why we have dozens of Linux phones projects which were developed by “community” and failed

Dozends? OpenMoko and?

> plus one “developed in isolation” which succeeded

plus a couple that were not community developed and failed, e.g. Maemo, WebOS

No Mir by default in Ubuntu 13.10

khim — 2013-10-04T16:22:36+00:00

Mea culpa. Of course Motorola's project was called EZX. You are right, OpenEZX is community project not supported by Motorola.

No Mir by default in Ubuntu 13.10

BlueLightning — 2013-10-04T09:19:17+00:00

Er, did Motorola actually sponsor OpenEZX or was it simply a community project?

No Mir by default in Ubuntu 13.10

khim — 2013-10-04T01:12:11+00:00

Heck, if things designed by committee are doomed to fail, C++ has been a failure for years and years... ;)

C++ was not designed by committee. It was polished by committee. It was designed by a small group of developers. Heck, the infamous ARM carries only two authors on it's cover—yet contains pretty good description of the whole langauge including bits and pieces which were not actually implemented when it was published!

When C++ was passed to a committee it was pretty mature and more-or-less finished work. Well, except for the STL, but that work, too was made by a small group of developers.

Heck .. posix, the internet, the web ????

In all cases you can name few principal authors who did initial work and then large committees which polished the result. This is fine and this exactly as it should be. There is very simple reason for why it must be done that way: efficiency and freedom of choice.

Initially when you try to create something new efficiency (both efficiency of your team and efficiency of the end result) dominates. You can not please everyone but that's not a problem at this stage: you must make sure this thing actually works. At this point you don't have luxury to involve committee and you don't have spare capacity to add unnecessary connections, bells and whistles. Custom-made pieces rule at this stage! They may be proprietary or open source, but they are designed to precisely adhere to requirements of your team, they are not generic and you don't discuss them for years. You just go and do what you need to do. Later, when thing more-or-less works it's time to involve “community”: this will mean that the end result may not be as holistic or pretty, but it will be more customizable and will give users more choices. If you'll not do that then someone else will do that.

Sometimes process is repeated if you need serious redesign of your work. Android gives us great example with it's Honeycomb 3.x release: it was supposed to be “tablets-only” and “proprietary”, but why? The answer is obvious: 3D acceleration. When Android was conceived and implemented it was developed for a systems without 3D accelerator but that made effects slow and clunky. Android 3.0 introduced hardware acceleration and this basically required to redo a lot of things in Android. In a sense Honeycomb was beta release of Android++ and ICS was basically the first stable release of 3D-accelerated Android++. That's why they repeated the same thing they did with the initial release of an Android: first close-source beta, then bugsfixes and eventually new release.

This is similar to what Wayland guys are doing and Mir guys are doing, but it took about three years from the initial idea to the full-blown implementation to the world-wide adoption. Wayland was conceived in about the same time and it's still not in production. Mir was probably started as response to this slowness but apparently Canonical underestimated complexity of the task and overestimated availability of skilled personnel.

The only thing Steve Jobs ever personally developed was his own ego, his bank account, and some crazy, effective people-manipulation skills.

It's true that Steve Jobs have not written the iOS code, but, on the other hand, he was the guy who managed to release iPhone in 2007—complete with “every frame is perfect” ideology and usable finger-driven GUI. Yes, all the pieces which he used were actually invented by others (often they were simultaneously invented by different guys in different companies), but he was able to choose the right combination of hardware and software to make this thing actually usable. Which BTW distinguishes Steve Jobs before his exile and Steve Jobs after his exile: first Steve Jobs was able to design pretty things (Apple III, Apple Lisa, Apple Macintosh), but he had no idea how to make sellable. People talked about his creations a lot but few have bought them. Second Steve Jobs was able to invent something people would instantly like—and not just like enough to gossip, but like enough to buy.

No Mir by default in Ubuntu 13.10

ssmith32 — 2013-10-03T23:54:51+00:00

> Steve Jobs taught us all that the great products are the ones that are NOT developed by committee.

Hah! As long as we're being flippant about it:

No, Steve Jobs taught us that you could take something developed by committees and the community (C,C++, BSD), put some shiny on top of it (objective-C, Cocoa, OSX), sell it to suckers as a status symbol, AND convince them that the set of ALL GREAT PRODUCTS consists of three things: the iPhone, the iPad, and OSX - ignoring clean running water, sanitation, the roads, security... gee, if only the Romans had brought iPad's too..

Heck, if things designed by committee are doomed to fail, C++ has been a failure for years and years... ;)

Heck .. posix, the internet, the web ????

I would say all great things were developed by a team of cooperating people operating by consensus (even if the consensus is let the iCon take all the credit, while we do the work), sometimes with the official label of "committee" on their name. The only thing Steve Jobs ever personally developed was his own ego, his bank account, and some crazy, effective people-manipulation skills.

Of course, I'm just being flippant, but one silly comment deserves another!

No Mir by default in Ubuntu 13.10

khim — 2013-10-03T21:31:00+00:00

Google has pretty much no real concern whether Android gets broader uptake in the wider Linux development world.

It does, it does. And it gets said uptake. E.g., there used to be pretty diverse and wild world of embedded systems - yet today Android is there and it's slowly crushing incumbents.

When you have large enough ecosystem developers will come—like kernel developers did.

The one thing Google does not do is playing favorites: “uptake in the wider Linux development world” is important but of course “uptake in the wider gamedev development world” is important too and “uptake in the wider mobile apps development world” is even more important. Not because Google hates “the wider Linux development world” but because “the wider Linux development world” is actually smaller than “gamedev development world” and much smaller than “mobile apps development world”.

You can call it successful anyway, but that's not the kind I meant.

Why not? What kind of “success” you've meant? I'm not jeering at you, that's a honest question. Usually success of a given endeavour is measured in monetary terms, but it's pretty poor measure for open source platforms thus I tend to think more in terms of “number of users” or may be more of a “percentage of users” (not many users use Linux on supercomputers, but it's still a success since most users who do anything with supercomputers at all do that with Linux nowadays). If you don't like that criteria then offer your own… with explanation for why do you think this must be measure of success.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-03T21:18:47+00:00

yes.. well, I actually worked for an SoC vendor at the time android was starting, so maybe had a bit of a different perspective. But seems to me like their success was:
(a) somewhat neutral ground (ie. not associated with one or small # of handset makers or carriers), with a big enough backer that all the carriers and handset makers pay attention
(b) get something out there... doesn't matter that it really sucks in the beginning, google is big enough to catch developer interest to start getting app ecosystem going
(c) rinse, repeat until you have something decent

But don't underestimate the google power.. a lot of people inside the industry were pretty much assuming android would be the next big thing back when android was in diapers and way inferior on a technical level compared to alternatives like maemo/meego.

No Mir by default in Ubuntu 13.10

khim — 2013-10-03T20:27:41+00:00

Really all? Why Google (an outsider company!) could do that but Nokia (with Meego), Motorola (with OpenEZX), Samsung (with SLP) and even dozen of high-profile mobile companies together (with LiMo) could not do that?

Yes, “corporate credibility” is important (and Ubuntu is sorely lacking in this department), but something else is clearly needed, too.

No Mir by default in Ubuntu 13.10

pboddie — 2013-10-03T20:19:42+00:00

Indeed. They went and closed bug #1, reflecting the situation on the ground in such an accurate fashion that when I had reason to go and purchase a computer in a retail outlet the other day there was a healthy choice of software pre-installed on the available models.

Just kidding, of course: there was a row of Macbooks and everything else was running Windows 8. Canonical's "mission accomplished" moment, without a doubt.

No Mir by default in Ubuntu 13.10

smurf — 2013-10-03T20:04:14+00:00

Canonical started out with a somewhat different vision. Microsoft did not.

This is not trolling. It's disappointment.

No Mir by default in Ubuntu 13.10

k8to — 2013-10-03T17:47:31+00:00

Google has pretty much no real concern whether Android gets broader uptake in the wider Linux development world. And they haven't gotten it.

You can call it successful anyway, but that's not the kind I meant.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-03T17:25:23+00:00

android won because it had the corporate credibility with carriers and handset manufactures which comes from the google backing. That is really all.

No Mir by default in Ubuntu 13.10

smurf — 2013-10-03T17:23:39+00:00

>> but then Mir will be ignored by developers not using Ubuntu.

I would assume that Mir will be ignored by developers not using Ubuntu PERIOD, no qualification necessary.

I have't heard of a single advantage of Mir, compared to Wayland.

No Mir by default in Ubuntu 13.10

khim — 2013-10-03T17:14:38+00:00

May be. But this goes against Occam's razor principle.

Android have won because it was there first. In “good enough” state, I mean. OpenMoko, Maemo/Meego, Tizen and many other projects have failed for one simple reason: they were late. And it's obvious that “proper” “community project” moves slower then “developed in isolation” project. Simply because you need more interactions to achieve consensus. Of course at later stage (when basic functionality is done) you need “community” (or at least “alliance”) because it's the only way to create lots of small tweaks, but start doing that too early and your project will not be done in time.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-03T12:01:51+00:00

I think you may find that the success of android is a result of factors other than the development model.

No Mir by default in Ubuntu 13.10

ovitters — 2013-10-03T11:40:49+00:00

I share your opinion. I want open source / free software companies to succeed (all of them!). No matter if that is Google where you can argue if Android is free software, or just any other company. For Canonical, I think they're not big enough to work isolated and this might cause a failure. My criticism is because I want them to succeed and I think there are things they could work on. At the moment only Google seems to have success (talking solely about Android). Copying another company is not enough, should figure out why they're a success.

No Mir by default in Ubuntu 13.10

khim — 2013-10-03T11:05:19+00:00

That's why we have dozens of Linux phones projects which were developed by “community” and failed plus one “developed in isolation” which succeeded. Paint me unimpressed.

I have my doubts about Mir, but they are related not to the fact that it's developed in isolation but to the fact that it's developed by a small company which does not have the ability to actively promote the end result, push large companies around, etc.

No Mir by default in Ubuntu 13.10

Neowin — 2013-10-03T05:10:21+00:00

When people refer to mjg59's ideas, please link to a more recent one:
http://mjg59.dreamwidth.org/28032.html

No Mir by default in Ubuntu 13.10

Neowin — 2013-10-03T04:56:36+00:00

Many Linux trolls have problems seeing Canonical, a normal, for-profit company having any business secret and/or using diplomatic language. Their heads are full of FUDs and conspiracy theories. They prefer to pull down Canonical rather than Microsoft.

No Mir by default in Ubuntu 13.10

Neowin — 2013-10-03T04:49:33+00:00

Trolls always come early. It's none of your business, Sir. Continue doing your work using MS Office.

No Mir by default in Ubuntu 13.10

k8to — 2013-10-03T03:07:09+00:00

Open Source development is not the only way to do things, nor in my opinion always the best. However, in the context of Open and/or Free systems, doing things in isolation is well-documented as substandard.

No Mir by default in Ubuntu 13.10

mjg59 — 2013-10-02T22:09:43+00:00

Not really. XMir is a Mir client, it depends on there being a Mir server process running. It just doesn't use that Mir server for performing drawing operations.

No Mir by default in Ubuntu 13.10

drag — 2013-10-02T21:31:58+00:00

> XMir runs on Mir. It is a major achievement to replace Xorg with XMir.

It's more accurate to say that Mir runs on XMir.

http://mjg59.dreamwidth.org/26254.html

It's a bit old article, but I doubt a huge amount has changed.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-02T21:20:01+00:00

well, considering that an nv employee talked openly at XDC about how EGL could potentially be extended to let them step out of the mir vs wayland debate... it doesn't really feel they they treat this as their secret sauce (and rightly so.. EGL is boring)

No Mir by default in Ubuntu 13.10

tuna — 2013-10-02T17:46:02+00:00

The only point I found in the comment was that it would be technically possible to have better VT switching, but that is not enabled in the current Mir/XMir setup.

Did you find something else?

No Mir by default in Ubuntu 13.10

rsidd — 2013-10-02T17:04:23+00:00

I assume you read it too, as well as the comments? There is a benefit to pushing (Mir/Wayland) out to users, even if they are using X(Mir/Wayland) to actually run their programs, and even if Xmir is not very different at this point from Xorg. Apart from wider testing of Mir itself, several other points are brought up in the comments.

No Mir by default in Ubuntu 13.10

ovitters — 2013-10-02T16:33:40+00:00

Because there is no benefit in doing that. You might want to read http://mjg59.dreamwidth.org/26254.html.

No Mir by default in Ubuntu 13.10

rsidd — 2013-10-02T16:13:57+00:00

XMir runs on Mir. It is a major achievement to replace Xorg with XMir. If it were trivial, why didn't Fedora or others ship with X-Wayland as default years ago? It would have been the best way to kickstart porting to Wayland.

No Mir by default in Ubuntu 13.10

ovitters — 2013-10-02T16:06:26+00:00

Canonical and Ubuntu rely on the work of others. The other companies have far more control. For Mir Canonical needs EGL, but Wayland needs this as well. The patching within just their distribution is possible, but then Mir will be ignored by developers not using Ubuntu.

Anyway, would be nice if you don't assume all kinds of things about me. E.g. starting about "The youug'uns" is funny and inaccurate. Probably you'll say that despite replying to me that you didn't mean me, but then it is a bit meaningless to include that tidbit, no?

No Mir by default in Ubuntu 13.10

ovitters — 2013-10-02T16:00:24+00:00

Why can't you discuss like a civilized person?

I am a member of the community and I already explained in another reply why this information is useful. If you have any real questions about this, feel free to ask in a normal way.

No Mir by default in Ubuntu 13.10

robclark — 2013-10-02T15:50:50+00:00

there is a pretty important distinction between "developed by committee" and "developed by community"