Inline Review With Miguel De Icaza 198
Interview With Miguel de Icaza
Bringing a component architecture to the UNIX platform
By Dare (Carnage4Life) Obasanjo
Dare Obasanjo: You have recently been in the press due to Ximian's announcement that it shall create an Open Source implementation of Microsoft's .NET development platform. Before the recent furor you've been notable for the work you've done with GNOME and Bonobo. Can you give a brief overview of your involvement in Free Software from your earlier projects up to Mono?
Miguel de Icaza: I have been working for the past four years on the GNOME project in various areas: organization of it, libraries and applications. Before that I used to work on the Linux kernel, I worked for a long time on the SPARC port, then on the software raid and some on the Linux/SGI effort. Before that I had written the Midnight Commander file manager.
Dare Obasanjo: In your Let's Make Unix Not Suck series you mention that UNIX development has long been hampered by a lack of code reuse. You specifically mention Brad Cox's concept of Software Integrated Circuits, where software is built primarily by combining reusable components, as a vision of how code reuse should occur. Many have countered your arguments by stating that UNIX is built on the concept of using reusable components to build programs by connecting the output of smaller programs with pipes. What are your opinions of this counter-argument?
Miguel de Icaza: Well, the paper addresses that
question in detail. A `pipe' is hardly
a complete component system. It is a transport
mechanism that is used
with some well known protocols (lines, characters,
buffers) to process
information. The protocol only has a flow of
information.
Details are on the paper:
http://primates.ximian.com/~miguel/bongo-bong.html
[Dare -- check the section
entitled "Unix Components: Small is Beautiful"]
Dare Obasanjo: Bonobo was your attempt to create a UNIX component architecture using CORBA as the underlying base. What are the reasons you have decided to focus on Mono instead?
Miguel de Icaza: The GNOME project goal was to bring missing technologies to Unix and make it competitive in the current market place for desktop applications. We also realized early on that language independence was important, and that is why GNOME APIs were coded using a standard that allowed the APIs to be easily wrapped for other languages. Our APIs are available to most programming languages on Unix (Perl, Python, Scheme, C++, Objective-C, Ada).
Later on we decided to use better methods for encapsulating our APIs, and we started to use CORBA to define interfaces to components. We complemented it with policy and a set of standard GNOME interfaces for easily creating reusable, language independent components, controls and compound documents. This technology is known as Bonobo. Interfaces to Bonobo exist for C, Perl, Python, and Java.
CORBA is good when you define coarse interfaces, and most Bonobo interfaces are coarse. The only problem is that Bonobo/CORBA interfaces are not good for small interfaces. For example, an XML parsing Bonobo/CORBA component would be inefficient compared to a C API.
I also wrote at some point:My interest in .NET comes from the attempts that we have made before in the GNOME project to achieve some of the things .NET does:
- APIs that are exposed to multiple languages.
- Cross-language integration.
- Contract/interface based programming.
And on top of things, I always loved various things about Java. I just did not love the Java combo that you were supposed to give or take.
APIs exposed to many languages we tried by having a common object base (GtkObject) and then following an API contract and a format that would allow others to wrap the APIs easily for their programming language. We even have a Scheme-based definition of the API that is used to generate wrappers on the fly. This solution is suboptimal for many reasons.
The Cross-language integration we have been doing with CORBA, sort of like COM, but with an imposed marshalling penalty. It works pretty well for non inProc components. But for inProc components the story is pretty bad: since there was no CORBA ABI that we could use, the result is so horrible, that I have no words to describe it.
On top of this problem, we have a proliferation of libraries. Most of them follow our coding conventions pretty accurately. Every once in a while they either wont or we would adopt a library written by someone else. This had lead to a mix of libraries that although powerful in result implement multiple programming models, sometimes different allocation and ownership policies and after a while you are dealing with 5 different kind of "ref/unref" behaviours (CORBA local references, CORBA object references on Unknown objects, reference count on object wrappers) and this was turning into a gigantic mess.
We have of course been trying to fix all these issues, and things are looking better (the GNOME 2.x platform does solve many of these issues, but still).
.NET seemed to me like an upgrade for Win32 developers: they had the same problems we had when dealing with APIs that have been designed over many years, a great deal of inconsistency. So I want to have some of this new "fresh air" available for building my own applications.
Dare Obasanjo: Bonobo is slightly based on COM and OLE2 as can be gleaned from the fact that Bonobo interfaces are all based on the Bonobo::Unknown interface which provides two basic services: object lifetime management and object functionality-discovery and only contains three methods:
module Bonobo {
interface Unknown {
void ref ();
void unref ();
Object query_interface (in string repoid);
};
};
which is very similar to Microsoft's COM IUnknown
interface which has the
following methods
HRESULT QueryInterface(REFIID riid, void **ppvObject);
ULONG AddRef();
ULONG Release();
Does the fact that .NET seems to spell the impending
death of COM mean that
Mono will spell the end of of Bonobo? Similarly
considering that .NET plans
to have semi-transparent
COM/.NET interoperability, is there a similar plan
for Mono and Bonobo?
Miguel de Icaza: Definetly. Mono will have to interoperate with a number of systems out there including Bonobo on GNOME.
Dare Obasanjo: A number of parties have claimed that Microsoft's NET platform is a poor clone of the Java(TM) platform. If this is the case why hasn't Ximian decided to clone or use the Java platform instead of cloning Microsoft's .NET platform?
Miguel de Icaza: We were interested in the CLR because it solves a problem that we face every day. The Java VM did not solve this problem.
Dare Obasanjo: On the Mono Rationale page it is pointed out that Microsoft's .NET strategy encompasses many efforts including
- The .NET development platform, a new platform for writing software.
- Web services.
- Microsoft Server Applications.
- New tools that use the new development platform.
- Hailstorm, the Passport centralized single-signon system that is being integrated into Windows XP.
Miguel de Icaza: Not at this point. We have a commitment to develop currently:
- A CLI runtime with a JITer for x86 CPUs.
- A C# compiler.
- A class library
All of the above with the help of external contributors. You have to understand that this is a big undertaking and that without the various people who have donated their time, expertise and code to the project we would not even have a chance of delivering a complete product any time soon.
We are doing this for selfish reasons: we want a better way of developing Linux and Unix applications ourselves and we see the CLI as such a thing.
That being said, Ximian being in the services and support business would not mind extending its effort towards making the Mono project tackle other things like porting to new platforms, or improving the JIT engine, or focusing on a particular area of Mono.
But other than this, we do not have plans at this point to go beyond the three basic announcements that we have made.
Dare Obasanjo: There are a number of other projects that are implementing other parts of .NET on Free platforms that seem to be have friction with the Mono project. Section 7.2 of Portable.NET's FAQ seems to indicate they have had conflict with the Mono project as does the banning of Martin Coxall from the dotGNU mailing list. What are your thoughts on this?
Miguel de Icaza: I did not pay attention to the actual details of the banning of Martin from the DotGNU mailing lists. Usenet and Internet mailing lists are a culture of their own and I think this is just another instance of what usually happens on the net. It is definitely sad.
The focus of Mono and .NET is slightly different: we are writing as much as we can in a high level language like C#, and writing reusable pieces of software out of it. Portable.NET is being written in C.
Dare Obasanjo: There have been conflicting reports about Ximian's relationship with Microsoft. On one hand there are reports that seem to indicate that there may be licensing problems between the license that will govern .NET and the GPL. On the other hand there is an indication that some within Microsoft are enthusiastic about Mono. So exactly what is Ximian's current relationship is with Microsoft and what will be done to ensure that Mono does not violate Microsoft's licenses on .NET if they turn out to be restrictive?
Miguel de Icaza: Well, for one we are writing everything from scratch.
We are trying to stay on the safe side regarding patents. That means that we implement things in a way that has been used in the past and we are not doing tremendously elaborate or efficient things in Mono yet. We are still very far from that. But just using existing technologies and techniques.
Dare Obasanjo: It has been pointed out that Sun retracted Java(TM) from standards processes at least twice, will the Mono project continue if .NET stops being an open standard for any reason?
Miguel de Icaza: The upgrade on our development platform has a value independently of whether it is a standard or not. The fact that Microsoft has submitted its specifications to a standards body has helped, since people who know about these problems have looked at the problem and can pin point problems for interoperability.
Dare Obasanjo: Similarly what happens if Dan Kusnetzky's prediction comes true and Microsoft changes the .NET APIs in the future? Will the Mono project play catchup or will it become an incompatible implementation of .NET on UNIX platforms?
Miguel de Icaza: Microsoft is remarkably good at keeping their APIs backwards compatible (and this is one of the reasons I think they have had so much success as a platform vendor). So I think that this would not be a problem.
Now, even if this was a problem, it is always possible to have multiple implementations of the same APIs and use the correct one by choosing at runtime the proper "assembly". Assemblies are a new way of dealing with software bundles and the files that are part of an assembly can be cryptographically checksummed and their APIs programmatically tested for compatibility. [Dare -- Description of Assemblies from MSDN gloassary]
So even if they deviate from the initial release, it would be possible to provide assemblies that are backwards compatible (we can both do that: Microsoft and ourselves)
Dare Obasanjo: Looking at the Mono class status page I noticed that a large number of .NET class libraries are not being implemented in Mono such as WinForms, ADO.NET, Web Services, XML schemas, reflection and a number of others. This means that it is very likely that when Mono and .NET are finally released apps written for .NET will not be portable to Mono. Is there any plan to rectify this in the future or is creating a portable .NET platform not a goal of the Mono project? Similarly what are the short and long term goals of the Mono project?
Miguel de Icaza: The status web page reflects the classes that people have "requested" to work on. The status web page is just a way of saying `Hey, I am working on this class as of this date' to avoid code duplication. If someone registers their interest in working on something and they do not do something after some period of time, then we can reclaim the class.
We are on the very early stages of the project, so you do see more work going on the foundational classes than on the end user classes.
I was not even expecting so many great and talented programmers to contribute so early in the project. My original prediction is that we would spend the first three months hacking on our own in public with no external contributions, but I have been proved wrong.
You have to realize that the goals of the Mono project are not only the goals of Ximian. Ximian has a set of goals, but every contributor to the project has his own goals: some people want to learn, some people like working on C#, some people want full .NET compatibility on Linux, some people want language independence, some people like to optimize code, some people like low level programming and some people want to compete with Microsoft, some people like the way .NET services work.
So the direction of the project is steered by those that contribute to it. Many people are very interested in having a compatible .NET implementation for non-Windows platforms, and they are contributing towards filling those gaps.
Dare Obasanjo: How does Ximian plan to pay for the costs of developing Mono especially after the failure of a number of recent venture funded, Free Software-based companies like Indrema, Eazel and Great Bridge and the fact that a sizable percentage of the remaining Free Software based companies are on the ropes? Specifically how does Ximian plan to make money at Free Software in general and Mono in particular?
Miguel de Icaza:Ximian provides support and services. We announced a few of our services recently, and more products and services have been on the pipeline for quite a while and would be announced during the next six months.
Those we announced recently are:
- Red Carpet Express: a subscription service for those who want a reliable high speed access to the Red Carpet servers.
- Red Carpet Corporate Connect: We modified our Red Carpet updater technology to help people manage networks of Linux workstations easily and to deploy and maintain custom software packages.
- Support and services for the GNOME desktop and Evolution: Our latest boxed products are our way of selling support services for the various products we ship.
The particular case of Mono is interesting. We are working on Mono to reduce our development costs. A very nice foundation has been laid and submitted to ECMA. Now, with the help of other interested parties that also realize the power of it, we are developing the Mono runtime and development tools to help us improve our productivity.
Indeed, the team working on Mono at Ximian is the same team that provided infrastructural help to the rest of the company in the past.
Dare Obasanjo: It is probably little known in some corners that you once interviewed with Microsoft to work on the SPARC port of Internet Explorer. Considering the impact you have had on the Free Software community since then, have you ever wondered what your life would have been like if you had become a Microsoft employee?
Miguel de Icaza: I have not given it a lot of thought, no. But I did ask everyone I interviewed at Microsoft to open source Internet Explorer, way before Netscape Communicator was Open Sourced ;-)
What does user interface have to do with Mono? (Score:5, Insightful)
Considering that they are currently working on the compiler, the language runtime and base class libraries for Mono I fail to see what user interfaces have to do with anything at this stage in the development process.
On the other hand if this was an interview about GNOME, which it isn't then I assume he would have mentioned the user interface issues.
Re:What does user interface have to do with Mono? (Score:4, Insightful)
That's exactly the problem. It's called "user-centered system design" for a reason. User experience is upstream from engineering in a user-centered project. You don't bring designers in late in the game to slap some icons on the system. Instead, you have a set of designs that engineers work towards implementing.
On the other hand if this was an interview about GNOME, which it isn't then I assume he would have mentioned the user interface issues.
I believe your assumption is in error. I have seen de Icaza discuss GNOME in exactly the same way -- naming lots of libraries and implementation strategies, but saying almost nothing about user-facing issues. That's why I noted the continuing pattern in my message.
Tim
Re:How will mono solve inter language problems ? (Score:1, Insightful)
Re:API Wrappers (Score:2, Insightful)
One good thing to come out of the Gnome C++ bindings is libsigc++: signals and slots done within the language. To be fair to the Qt guys, at the time most compilers didn't have support to do what libsigc++ does and they were forced to use a workaround.
On the one hand we have Gnome with an incomplete C++ API and on the other we have KDE with a non-ideal signalling mechanism.
Re:Obsoleting unfinished software... (Score:2, Insightful)
So why would anyone use Bonobo instead of Mono? Lots of reasons. For one thing, it's easy to wrap legacy applications and languages in Bonobo than Mono. Since most of the world's applications are legacy applications and Mono won't be finished and stabilized for at lest 2 years, this is a big deal. Also, not everyone will be developing for Mono, so Bonobo is a good way of creating components. And Bonobo isn't GNOME specific. You can use it on Windows without including any GTK+ libraries.
BTW, KParts is Qt specific. You can't write an UNO (OpenOffice's component) KPart. You can write a Bonobo-UNO bridge so all UNO objects appear as Bonobo components and can be used by Bonobo applications.
This isn't about KDE (Score:2, Insightful)
Gnome sucks.
Computers, in general, suck.
This interview wasn't about Gnome; it was about a component model that might be better than the one Gnome currently uses. Although MS doesn't often come up with good ideas, it does employ some extremely bright people; if some of those bright people come up with a good idea, it behooves us to learn.
In this way, perhaps computers will someday suck less.
Re:What does user interface have to do with Mono? (Score:3, Insightful)
All software is made of many layers, and there are different concepts at each layer. The highest layer deals in concepts that are exposed to the user, and should be designed in a "user-centered" way. The other layers, by and large, deal in concepts that the user shouldn't have to think about, so their design should be based mostly on engineering considerations.
The notion that every layer of design needs to be "user-centered" is a gross distortion that harms software development. It would enormously complicate the design of the lower-level functionality, impeding the development of clean, simple layers that can be used by a variety of applications (some unforseen). I think there are enough examples to make this obvious.
By your standard, it seems we should criticize CPU designers for not considering the end-user. This is the same fallacy that leads people to say foolish things like, "You can't build a beginner-friendly interface on top of Unix".
I have seen de Icaza discuss GNOME in exactly the same way
Well, even though GNOME overall is a user desktop, large parts of it are not at the user-facing level, so it is entirely appropriate for Miguel to talk about those parts of GNOME from an engineering viewpoint. Hopefully, though, other people in GNOME think more about the user-facing parts. :-)
Re:Obsoleting unfinished software... (Score:4, Insightful)
All of it. The icons in KDE are much clearer and easier to distinguish than the Gnome ones. Icons should *not* be mini-photographs -- they should be clear simple representations. The Gnome icons give me a headache.
KDE had issues with look and feel back in the KDE 1 days. It doesn't any more. Gnome has the advantage of a larger community developing themes and styles, but the default in KDE 2 is perfectly acceptable, and the recent point releases have greatly increased the 'style candy' aspects over the original 2.0.
--
Don't take the sniping of random Slashdot trolls as a reason for not helping to theme KDE -- but don't go into it with the attitude that you are saving KDE from some horrible design mistake, because there isn't one there.
Re:What does user interface have to do with Mono? (Score:3, Insightful)
Exactly. If you start with programming concepts and then slap an interface on top of it, you find that the underlying layers won't support the interface you need. For instance, in a network application framework, there are serious issues about timeout. If they're approached from a traditional programming perspective, you'll be left with a system that drops connections out from under the user and loses their work in progress.
This kind of issue is pervasive throughout any software system. You've mentioned a few other good examples; I'd add things like timeout management (as above), exception handling, module installation, and refresh timing. Engineers looking at these problems will generally take paths of least resistance which wind up hurting the user. The only way to prevent these problems is to start with the user and work back to the system. If you try to fix them after the foundation is already laid, engineering will always report that it's too late to tear up the foundation.
Tim
Re:dammit anyways (Score:4, Insightful)
I hope you are trolling, but you probably aren't.
Miguel is building Mono because he A) thinks it is cool, B) will probably be popular, C) Microsoft did much of the hard work of designing and documenting the system :).
Basically Gnome has always been about being able to reuse Gnome libraries and components in your language of choice. That's a pretty darn good goal, but it is definitely trickier than it looks. Micrososft and Miguel have both come to the conclusion that the easiest way to solve this problem is via a virtual machine.
Basically, it would allow Python hackers like me to reuse any Mono component using a simple:
import foo
Not only that, but Perl hackers could then import my Python package using a simple:
use bar;
These packages would likewise be available from any other language that had been ported to the CLR. Now, that's some pretty cool stuff.
The fact that Microsoft sponsored .NET, and that they have tied the CLR and the virtual machine with a lot of tech that is basically evil (Passport and Hailstorm), doesn't mean that the idea behind Mono isn't pretty cool.
When it's all said and done Mono will probably be compatible with .NET in the same way that gcc is compatible with Visual C++ (ie. not very), but that's still good because it will give Gnome hackers another tool. Miguel's canonical example is reusing an XML parser. Such a thing isn't really possible with Bonobo, but it will be possible if the XML parser is written as a Mono component.
Personally, I am content using a mixture of Python and C, but the idea behind Mono is intriguing, never mind who wrote the specification.
Re:What does user interface have to do with Mono? (Score:3, Insightful)
I don't know what planet you are from. But here in the real world the only "users" that are likely to use the Mono compiler and language runtime are developers.
In other words, the developers working on the product don't need to query a bunch of lawyers, accountants, and fast food personnel to find out how they should work. They already know how they should work.
The fact of the matter is that the compiler should work very similarly to gcc, or the command line java compilers from Sun and IBM, or every other compiler written in the last 10 years. Especially since their target audience is currently Linux developers.
When Miguel talks about Gnome to hackers, of course he talks about libraries, APIs, and implementation. On the other hand, Miguel makes no bones about modeling his User Interfaces from Microsoft's software. Gnumeric's interface, for example, is close enough to Microsoft Excel that I recently used it to ace a College course in the use of Excel. As far as I am concerned, when it comes to spreadsheet usability aping Excel is the only way to go. I have also seen lots of articles by Miguel where he praises Windows UI, and badmouth's Linux's UI (including Gnome).
Believe me, when it comes to the "user centered" design of a compiler Miguel is as qualified as anyone I can think of to choose the UI design.