[vwnc] Platform priorities

Andre Schnoor wrote:

>Am 29.11.2009 um 10:43 schrieb Antony Blakey:
>> Another important choice is to start with OSX, not with Windows or 
>> GTK as your baseline.
>
> Indeed. MacOS X can safely be considered the
> most potent superset of all platforms.

That may or may not be true. Basing development too strongly on any one platform generally ends up being to the detriment of cross-platform behaviour. It's like writing a class hierarchy with the superclass being a particular concrete case.

 

I do worry occasionally about the increasing percentage of Mac OS X fans in these discussions, both within Cincom and without. Not that I have anything against the platform or those people - far from it - but I feel that personal preference shouldn't unduly influence the direction of the product. IMHO that direction should be determined by current and projected revenues from the different platforms, from Cincom's customers' customers. Obviously, Cincom have little chance to influence the platform choices of those people - even amazingly cool and accurate UI support for the wrong platform won't make them switch. However, the quality of VW's UI support on the platform they already use will have a significant influence on how many decide to buy.

 

IIRC I was told some years ago by Cincom that the total VW revenue was around 90% Windows-based. I'd imagine that figure has changed somewhat, but if it follows the percentage of users of platforms in general, maybe not by much.

 

In the last 5 years or so most new platform-specific VW development effort seems to have been expended on the Mac, and that actually makes sense to me. Linux isn't much of a market in terms of people actually paying for products. The (commercial) Windows world has stayed largely in XP, so business apps have been able to survive without a Vista L&F. (Which gives rise to another interesting question: what is the split of VWrevenue between apps for people at work, and apps for people at home?)

 

But with the coming of Windows 7, which looks likely to be a success for both home and business users, and the faltering of Snow Leopard, I imagine the market the majority of VW apps will be selling into will be Windows 7. It would be nice to see some action in VW development in that direction (ribbon, multitouch...), on top of the things that are de rigeur on all platforms (alpha, antialiasing...).

 

All the best,

Steve

 

 

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Am 29.11.2009 um 21:55 schrieb Eliot Miranda:

And for that you get to discard the event queue road block in the VM and hook up to native event pumps.  What would make that much easier is throwing away my overengineered and underperformant threaded FFI in place of David Simmons' architecture that I've recently implemented in Squeak.   But that can be done independently in parallel.

That'll be a huge step towards a much smaller much more focussed VM which doesn't get in the way of APIs and concentrates on providing effective fast and simple access to them.

I absolutely agree, but, at least on the Mac this would need a natively multi-threaded VM, as all GUI code is required to run on the thread that owns the event loop. This can not be achieved with green threads.

Andre

[hidden email] wrote:

Steven Kelly wrote:
> > In this case, basing the fundamentals on the Mac would lead to
> > higher costs and lower revenues on ~90% of PCs, with corresponding
> > upsides on <10% of PCs.
>
> Hm. I guess we are probably talking about different things.

Maybe. I was relying on what Eliot said would happen, if the basis for
VW graphics became the Mac:

> On Windows and especially Linix that may
> require lots of support.

You and Eliot have far more experience than I do in messing around with
graphics in the VM. If I understand Eliot correctly, reimplementing all
the VMs (or even VM/VI boundary) to make them work like a Mac would
result in increased expenditure per unit change on Windows and Linux
(but decreased on Mac). It would of course also be a major project,
which at best would leave the situation unchanged for Windows users. It
could of course have the best long-term effect, but it's hard to see how
that could be the case if supporting Windows and Linux would become
harder work for Cincom.

As for being able to achieve the same appearance and performance on
Windows, whether the fundamentals are Mac-based or the current ones, I
don't know. Certainly it seems to be the case that you feel pain on the
Mac because VW's current fundamentals are different from the Mac's. Even
if the Mac way is more conceptually sound, translating to a 'stupid'
Windows graphics world is going to result in problems of speed and
accuracy, compared with working natively in the Windows world.

Maybe part of the confusion is between which things are just suboptimal
in the current VW graphics approach, and which are not Mac-friendly. I'm
all for moving away from direct writes to graphic contexts, gc copying
etc. (as VW seems to be doing).

Steve

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc

Eliot Miranda wrote 30 November 2009 22:51:

> implementing cross-platform abstractions in the VM is a
> really bad idea because the VM, being C, and having to
> provide backward-compatibility with older images simply
> can't evolve fast enough and instead one should expend
> resources on the VM's FFI so that the FFI does not
> prevent one from accessing platform APIs.  

A major change in VM graphics support would be a new major number, so there would be no expectation of backwards compatibility. Some of the 7.* VMs have broken backward compatibility anyway in some way.

A faster FFI would of course be useful in any case - although looking forward on Windows the speed needs are more in calling .NET code or SOAP than DLLs.

> Second, when choosing a cross-platform abstraction it is
> better to aim at implementing a superset rather tan a
> common subset because the common subset is too limiting.

Of course. I suppose we need to be more concrete here about what we are discussing. Graphics and UI have been the topics so far, and that covers say three levels:

1) low level - e.g. can pixels have alpha
2) 'advanced' functionality - e.g. gradient fills
3) GUI widgets

IIRC VW hasn't moved an inch on the first two levels in all the time it's existed(!). Many basic bugs are still uncorrected (e.g. ellipse outlines with width 2 being seriously ugly, ellipse fills being off by one pixel).

Level 3 'chameleon switching' used to be one of the big plus points of VW, but nowadays (as Mark Plas pointed out) we don't even know what a widget should look like on a given platform, because native apps from the platform provider don't use them. 10 years ago users demanded pixel-faithful emulation of Windows widgets, and Cincom (OS/PPD/PP) could make many people very happy with a new L&F for a new Windows version. Nowadays, I still want Vista and Win7 L&Fs, but I doubt they're going to have such an effect on user satisfaction.

I think a working grid or ribbon, using platform colours but otherwise the same on all platforms, would have a bigger effect. Other than that, to make the users happy these days, each application needs to add its own cool effects. That's really bad news for reuse, but there's little we can do. Worse, cool effects all require new facilities from levels 1 and 2, so in the current VW we either do without or each spend massive amounts of our own effort on achieving them in some roundabout way.

I'm delighted to see the work on Cairo and OpenGL, and hope functionality like that makes its way into the base graphics system of VW soon. IMHO having bolt-on DLLs etc. doesn't work for the cross platform story: downloading and maybe even compiling 3rd party DLLs for several different platforms is acceptable for preview code, but not in production - at least not for a core component of the system like graphics.

Steve

_______________________________________________
vwnc mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/vwnc