Hi,On Mar 25, 2016, at 9:45 AM, Esteban Lorenzano <estebanlm@gmail.com> wrote:
On 25 Mar 2016, at 07:43, Cl��ment Bera <bera.clement@gmail.com> wrote:
Yeah this is very well-known. Note that here we are talking about the pi B with Spur + the JIT, 2 years ago I remember needing 35 second on the pie A to open Nautilus on the stack vm while the Squeak UI was usable.
The Pharo UI has been focusing on IDE features at the cost of performance, such as auto-completion, better RB integration, Nautilus, new debugging tools... None of the current tools (Nautilus, GTInspector, GTPlayground, ...) are fast enough for slower devices, some of them are barely fast enough for a laptop (for example at some point there was no auto-refresh on GTInspector because of performance). It's not clear if it's due to the new tools themselves as Morphic is drastically slower too on Pharo.
The Squeak UI has been focusing on performance and stability. It is very convenient for slow devices but also for slow VM like for example the debug VM.
The official answer to this problem is that Block/Brick will be faster so it will solve all this problems. We will see.
yes, this is the ���semi-official��� answer and that���s an error. Yes ideally it will solve some problems (the ones related to Morphic), but current lack of performance on Pharo tools is not because of Morphic but because of some non optimised choices.
For example, months ago I got pissed because our current list implementation was not fast enough for our tools and that���s why I made FastTable: because it was evident we needed a better design for it. It speeded the startup of Nautilus by an order of magnitude. This is a widget developed with Morphic, but with a not-naive approach, so is fast.
As lists, some others widgets (notably tree) could be optimised���
But this will not be the solution of the problem either: we need to spend time on optimisation. Nautilus is not a good tool (we all know it has design problems all around) but with the time had improve and not using it is almost enjoyable��� but for example, the new debugger (which I���m not criticising as a tool, I like it a lot) makes me hurt��� is slower than eclipse debugger and that���s a lot to say.
That is a good one :). There is an open issue with the blinking problem when you accept code which makes it appear slower than it actually is, but we are stuck with it and we could benefit from help.Also new inspector: is good, but can���t we optimise it? maybe we need to make some concession, but is the moment of doing it.
The optimization we should do now is to replace the appearances of MorphTreeMorph with FastTable. This would imply that the Raw presentation would not longer have a tree (because it is not supported in FastTable). I would be very happy if we could do that. Do we agree on this?
actually there is a tree implementation based on fast table, but I���m not sure is production ready (Cyril can say something about).
But anyway since I never understood why we have a tree there (since we will ���open��� attribute in next panel), I do agree, strongly :)
Recently I���m noticing a slowdown in Spotter: I believe is because one of his categories is refreshing to the network but I���m not sure: this slowdown is not production acceptable, for me is a Show Stopper because it kills the usability of the tool (of a tool who is great, but it HAS to be fast, otherwise is useless).
Could you be more specific about the Spotter problem that you see?
not sure, I open and it takes some seconds to actually show anything. I thought it was caused because of the catalog refresh to populate projects, but I���m not sure.
In any case, load catalog projects in spotter is cool, but we need to find a better implementation. One that does not freezes spotter until it refresh (remember, Pharo is also used in places with crapy connection or not connection at all).
Also, in the last couple of days Alex Syrel improved the speed of Spotter significantly. It is not yet integrated. It would be useful to get feedback on this because maybe we missed something.Mea culpa: more than once I have not protested even knowing than solution X was ���10% slower than the older one, but that���s ok, we can live with it������ guess what? once you accumulate enough of those 10% penalties, you have a barely usable system.
We need to concentrate efforts on performance. There will not be magical solution: if we do not spend time on enhance tools for performance, they will be not good enough.
Can I ask you guys to spend some time optimising tools, before we release Pharo 5? Remember, ���First impressions counts���! (we even have a category for it in FogBugz).
It would indeed be useful to get more eyes spent on this.
yes!
if you want to suffer, do this experiment:
1. open a Pharo2 and open debugger on anything
2. then Pharo3, 4 and 5��� feel the incremental slowdown
3. cry :���(
Esteban