I have attached a dialog we are getting in our VW 3.0 image. It seems to
me to be the typical out of memory dialog that the VM generates. However, I have been asked to confirm that this is what it is. It also appears that sometimes the dialog is not generated but the image is blown out nonetheless. Would that be also consistent with an out of memory condition? I say yes, just so low that it was not able to pull off the warning. The memory policy is basically vanilla although whatever the settings are I have been told that they are setup to favor growth over reclamation. I'm not handling the 3.0 product. I guess part of the consternation here on the part of 3.0 developer is why if the memory policy is set to favor growth do we get out of memory conditions in a box that seems to still have plenty of memory. I guess I don't know the answer. My approach in the past has been to not try to answer that question but instead to fiddle with memory policies and even look at subclassing memory policy if necessary. I appreciate any feedback. I also hope that the image actually gets embedded into this email as oppossed to forcing folks to open the attachment. I use Opera and although it has many advantages they still don't have html formatting available for outgoing email. thanks -- Charles A. Monteiro image001.jpg (11K) Download Attachment |
Hi Charles:
The 3.x VMs had headroom issues that are*not* mitigated by memory policy. I know that on Linux, the dialog you see is often caused by a large number of fonts loaded. Since I don't use Windows, I can only guess that the cause may be the same. Good luck!! On Mar 21, 2006, at 6:33 AM, Charles A. Monteiro wrote: > I have attached a dialog we are getting in our VW 3.0 image. It > seems to me to be the typical out of memory dialog that the VM > generates. However, I have been asked to confirm that this is what > it is. It also appears that sometimes the dialog is not generated > but the image is blown out nonetheless. Would that be also > consistent with an out of memory condition? I say yes, just so low > that it was not able to pull off the warning. > > The memory policy is basically vanilla although whatever the > settings are I have been told that they are setup to favor growth > over reclamation. > > I'm not handling the 3.0 product. I guess part of the consternation > here on the part of 3.0 developer is why if the memory policy is > set to favor growth do we get out of memory conditions in a box > that seems to still have plenty of memory. I guess I don't know the > answer. My approach in the past has been to not try to answer that > question but instead to fiddle with memory policies and even look > at subclassing memory policy if necessary. > > I appreciate any feedback. I also hope that the image actually gets > embedded into this email as oppossed to forcing folks to open the > attachment. I use Opera and although it has many advantages they > still don't have html formatting available for outgoing email. > > thanks > > > > -- > Charles A. Monteiro Thanks!! Joseph Bacanskas [|] --- I use Smalltalk. My amp goes to eleven. image001.jpg (11K) Download Attachment |
Charles,
Judging by the dialog picture you are using VM version 3.1. Did you try other releases of the 3.1 VM -- such as 3.1c or 3.1d? It may as well fix your problem. My 2c. --Mark Pirogovsky Joseph Bacanskas wrote: > Hi Charles: > > The 3.x VMs had headroom issues that are*not* mitigated by memory > policy. I know that on Linux, the dialog you see is often caused by a > large number of fonts loaded. Since I don't use Windows, I can only > guess that the cause may be the same. > > Good luck!! > > > On Mar 21, 2006, at 6:33 AM, Charles A. Monteiro wrote: > >> I have attached a dialog we are getting in our VW 3.0 image. It seems >> to me to be the typical out of memory dialog that the VM generates. >> However, I have been asked to confirm that this is what it is. It >> also appears that sometimes the dialog is not generated but the image >> is blown out nonetheless. Would that be also consistent with an out >> of memory condition? I say yes, just so low that it was not able to >> pull off the warning. >> >> The memory policy is basically vanilla although whatever the settings >> are I have been told that they are setup to favor growth over >> reclamation. >> >> I'm not handling the 3.0 product. I guess part of the consternation >> here on the part of 3.0 developer is why if the memory policy is set >> to favor growth do we get out of memory conditions in a box that >> seems to still have plenty of memory. I guess I don't know the >> answer. My approach in the past has been to not try to answer that >> question but instead to fiddle with memory policies and even look at >> subclassing memory policy if necessary. >> >> I appreciate any feedback. I also hope that the image actually gets >> embedded into this email as oppossed to forcing folks to open the >> attachment. I use Opera and although it has many advantages they >> still don't have html formatting available for outgoing email. >> >> thanks >> >> >> >> -- >> Charles A. Monteiro > > > ------------------------------------------------------------------------ > > > Thanks!! > Joseph Bacanskas [|] > --- I use Smalltalk. My amp goes to eleven. > |
Mark Pirogovsky wrote:
> Charles, > Judging by the dialog picture you are using VM version 3.1. Did you try > other releases of the 3.1 VM -- such as 3.1c or 3.1d? It may as well > fix your problem. There was a thread on VW-dev recently ('fixed space has issues') that had the same error exit. Eliot posted a fix for that particular issue on 3/3/2006 (I would point you to the archive but http://parcplace.net seems to have disappeared). We have an application that shows this error exit so we ported it forward from 3.1 to 7.4 which didn't fix the problem. Applying Eliot's patch didn't help either. We got the frequency of occurrence down a *lot* by incrementing some headroom value in the MemoryPolicy, but I'll have to prod my colleagues tomorrow to get the details. R - |
In reply to this post by Charles A. Monteiro-2
One possibility: there was a problem in 3.0 with the memory policy not
reserving enough space to do a garbage collect. This led to crashes under heavy memory loading. The following solves it: | mp | mp := MemoryPolicy new setDefaults contiguousSpaceSafetyMargin: 500000; availableSpaceSafetyMargin: 500000. ObjectMemory installMemoryPolicy: mp! As Mark mentioned, you should make sure you are using the 3.1c VM on Windows and 3.1d on Linux. HTH, Steve > -----Original Message----- > From: Charles A. Monteiro [mailto:[hidden email]] > Sent: 21 March 2006 16:33 > To: [hidden email] > Subject: 3.0 Out of memory > > > I have attached a dialog we are getting in our VW 3.0 image. > It seems to > me to be the typical out of memory dialog that the VM > generates. However, > I have been asked to confirm that this is what it is. It also > appears that > sometimes the dialog is not generated but the image is blown out > nonetheless. Would that be also consistent with an out of memory > condition? I say yes, just so low that it was not able to > pull off the > warning. > > The memory policy is basically vanilla although whatever the > settings are > I have been told that they are setup to favor growth over reclamation. > > I'm not handling the 3.0 product. I guess part of the > consternation here > on the part of 3.0 developer is why if the memory policy is > set to favor > growth do we get out of memory conditions in a box that seems > to still > have plenty of memory. I guess I don't know the answer. My > approach in the > past has been to not try to answer that question but instead > to fiddle > with memory policies and even look at subclassing memory policy if > necessary. > > I appreciate any feedback. I also hope that the image actually gets > embedded into this email as oppossed to forcing folks to open the > attachment. I use Opera and although it has many advantages > they still > don't have html formatting available for outgoing email. > > thanks > > > > -- > Charles A. Monteiro > |
In reply to this post by Reinout Heeck
There are a few things I've been having trouble with using Pollock.
I'm currently up to date with Pollock 7.4 (thanks Sam and Michael!). Running using the non-x11 Mac vm, the arrow keys do not do anything in the TextEdit control. Other platforms seem fine. Sluggishness. Pollock in general isn't snappy. I do a lot of my work on an older Pentium 3 so I really feel it. There are still a lot of these machines out there. A lot of my customers are running on even older machines than that. I'm sure that Sam is aware of the situation, but I'm just providing the feedback that I think I should. In particular the TextEdit control is unusably slow. On my 1.5GHz PowerBook G4 the TextEdit control lags behind my typing, spitting out what I type at less than a character per second. :-/ The InputField control is almost as slow. Performance on my 500MHz X20 Thinkpad is only slightly better. It would be great to have a fix for this ASAP, as I am demoing my software to the NYC ST group in a couple of weeks. Placing a TextEdit control inside a tabbed form causes the Enter keypress to be mishandled. I have to press Shift+Enter to cause the action to register. For Windows 98/XP looks: 1) The sizing for tabs isn't correct in the Windows XP look and feel. The longer the label is, the more extra whitespace added to the right side of the tab. 2) There is no focus indicator for listboxes (also for Motif look). For the Motif look: 1) The button controls aren't as tall as they should be. 2) Scrollbars have an ugly black line on the top and left borders. Thanks in advance, -Carl Gundel, author of Liberty BASIC http://www.libertybasic.com |
Hello Carl, All,
Sluggishness. Pollock in general isn't snappy. I do a lot of my work Yes, I agree with you that in some cases, this is very slow. I do my work in a Pentium IV 2.4 GHz and the lack of reaction in panes (get focus, reaction of panes about clicks, change state of checkButtons and radioButtons, unfold the drop down list, etc) is evident. I observed that when you put panes inside a Form this situation grow exponentialy and the flicking is too much. I'm sure that Sam and Pollock team is aware of this too, but I wonder if this will be solved sooner. Like Carl, this is only my feedback to the Pollock team that, by the way, I think that are making a good work. Best Regards! |
The Pollock team :)
On Thu, 23 Mar 2006 08:00:13 -0500, Guillermo Sapaya <[hidden email]> wrote: > Hello Carl, All, > > Sluggishness. Pollock in general isn't snappy. I do a lot of my work >> on an older Pentium 3 so I really feel it. There are still a lot of >> these machines out there. A lot of my customers are running on even >> older machines than that. I'm sure that Sam is aware of the situation, >> but I'm just providing the feedback that I think I should. > > > Yes, I agree with you that in some cases, this is very slow. I do my > work in > a Pentium IV 2.4 GHz and the lack of reaction in panes (get focus, > reaction > of panes about clicks, change state of checkButtons and radioButtons, > unfold the drop down list, etc) is evident. I observed that when you put > panes inside a Form this situation grow exponentialy and the flicking is > too > much. I'm sure that Sam and Pollock team is aware of this too, but I > wonder > if this will be solved sooner. Like Carl, this is only my feedback to the > Pollock team that, by the way, I think that are making a good work. > > Best Regards! > Guillermo Sapaya -- Charles A. Monteiro |
In reply to this post by Carl Gundel
Carll,
>There are a few things I've been having trouble with using Pollock. >I'm currently up to date with Pollock 7.4 (thanks Sam and Michael!). > >Running using the non-x11 Mac vm, the arrow keys do not do anything in >the TextEdit control. Other platforms seem fine. That's a Mac VM problem but I'll add an AR for it. (AR50549) >Sluggishness. Pollock in general isn't snappy. I do a lot of my work >on an older Pentium 3 so I really feel it. There are still a lot of >these machines out there. A lot of my customers are running on even >older machines than that. I'm sure that Sam is aware of the situation, >but I'm just providing the feedback that I think I should. > >In particular the TextEdit control is unusably slow. On my 1.5GHz >PowerBook G4 the TextEdit control lags behind my typing, spitting out >what I type at less than a character per second. :-/ The InputField >control is almost as slow. Performance on my 500MHz X20 Thinkpad is >only slightly better. It would be great to have a fix for this ASAP, >as I am demoing my software to the NYC ST group in a couple of weeks. I wish it was faster too. We have a HUGE step planned for next winter dealing with speed issues. Nothing sooner I'm afraid. >Placing a TextEdit control inside a tabbed form causes the Enter >keypress to be mishandled. I have to press Shift+Enter to cause the >action to register. AR50550 for this one. >For Windows 98/XP looks: 1) The sizing for tabs isn't correct in the >Windows XP look and feel. The longer the label is, the more extra >whitespace added to the right side of the tab. AR50551 for this one >2) There is no focus indicator for listboxes (also for Motif look). Do you mean items don't highlight? There is no other focus indicator that I'm aware of. >1) Motif: The button controls aren't as tall as they should be. I don't get what you mean here. They're as big as you make 'em. Motif Buttons do have a 7 pixel border around the "button" part, that cut into the inner bounds. That's the way it is... However, I've been thinking of getting rid of that all together and just copying the Win9x look for buttons in Motif. Haven't decided yet. >2) Motif: Scrollbars have an ugly black line on the top and left borders. AR50552 for this one. And So It Goes Sames ______________________________________________________________________ Samuel S. Shuster [|] VisualWorks Engineering, GUI Project Smalltalk Enables Success -- What Are YOU Using? |
In reply to this post by Charles A. Monteiro-2
Charles,
> The Pollock team :) Me, Myself and I, and sometimes Kevin :) And So It Goes Sames ______________________________________________________________________ Samuel S. Shuster [|] VisualWorks Engineering, GUI Project Smalltalk Enables Success -- What Are YOU Using? |
but doesn't Vassili sometimes splash around with you and the Greek :)
On Thu, 23 Mar 2006 14:41:40 -0500, Samuel S. Shuster <[hidden email]> <[hidden email]> wrote: > Charles, > >> The Pollock team :) > > Me, Myself and I, and sometimes Kevin :) > > And So It Goes > Sames > ______________________________________________________________________ > > Samuel S. Shuster [|] > VisualWorks Engineering, GUI Project > Smalltalk Enables Success -- What Are YOU Using? -- Charles A. Monteiro |
In reply to this post by Samuel S. Shuster <sames@interaccess.com>
On Mar 23, 2006, at 2:40 PM, Samuel S. Shuster <[hidden email]>
wrote: >> There are a few things I've been having trouble with using Pollock. >> I'm currently up to date with Pollock 7.4 (thanks Sam and Michael!). >> >> Running using the non-x11 Mac vm, the arrow keys do not do anything in >> the TextEdit control. Other platforms seem fine. > > That's a Mac VM problem but I'll add an AR for it. (AR50549) No workaround? The arrow keys work fine for Wrapper on that VM. >> In particular the TextEdit control is unusably slow. On my 1.5GHz >> PowerBook G4 the TextEdit control lags behind my typing, spitting out >> what I type at less than a character per second. :-/ The InputField >> control is almost as slow. Performance on my 500MHz X20 Thinkpad is >> only slightly better. It would be great to have a fix for this ASAP, >> as I am demoing my software to the NYC ST group in a couple of weeks. > > I wish it was faster too. We have a HUGE step planned for next winter > dealing > with speed issues. Nothing sooner I'm afraid. Can't you do anything at all? I have to ship this by late August, or this will have serious repercussions for my business. :-/ Perhaps you could make some suggestions about where the speed issue is, and I could take a stab at it myself? This is really important to me. -Carl Gundel, author of Liberty BASIC http://www.libertybasic.com |
Carl,
> No workaround? The arrow keys work fine for Wrapper on that VM. The AR is there, and I'll look into it. No workaround at this time. >>> In particular the TextEdit control is unusably slow. On my 1.5GHz >>> PowerBook G4 the TextEdit control lags behind my typing, spitting >>> out >>> what I type at less than a character per second. :-/ The >>> InputField >>> control is almost as slow. Performance on my 500MHz X20 Thinkpad is >>> only slightly better. It would be great to have a fix for this >>> ASAP, >>> as I am demoing my software to the NYC ST group in a couple of >>> weeks. >> >> I wish it was faster too. We have a HUGE step planned for next >> winter dealing >> with speed issues. Nothing sooner I'm afraid. > > Can't you do anything at all? I have to ship this by late August, > or this will have serious repercussions for my business. :-/ > Perhaps you could make some suggestions about where the speed issue > is, and I could take a stab at it myself? This is really important > to me. Carl, Pollock is NOT optimized. It is also not Production quality. One of the MUST DO tasks is optimization. I will not release Pollock for Production until that is done. It is a task that I expect will take up to 3 months to do. Unfortunately, Pollock isn't ready for optimization it... yet. What you can do is what I will be doing in that task. Load the profiler, and try to figure out why it's slow. Now, before I get hammered that I don't care, or that I'm not being responsive to my customers: Pollock is beta. I have no intention of letting this, or a hundred other issues, be ignored. Pollock will NOT go into production until ALL are addressed. And So It Goes Sames ______________________________________________________________________ Samuel S. Shuster [|] VisualWorks Engineering, GUI Project Smalltalk Enables Success -- What Are YOU Using? |
In reply to this post by Carl Gundel
Carl,
> No workaround? The arrow keys work fine for Wrapper on that VM. The OSX VM is adding #meta to the arrow (and other) keys! Wrapper (actually, inappropriately) ignores it. I'll make an AR for this tomorrow <sigh>... Actually two. One for the VM, the other for Pollock to deal with a workaround until which time the VM is fixed. And So It Goes Sames ______________________________________________________________________ Samuel S. Shuster [|] VisualWorks Engineering, GUI Project Smalltalk Enables Success -- What Are YOU Using? |
In reply to this post by Carl Gundel
Carl Gundel wrote:
> Can't you do anything at all? I have to ship this by late August, or > this will have serious repercussions for my business. :-/ Perhaps > you could make some suggestions about where the speed issue is, and I > could take a stab at it myself? This is really important to me. Hi, I am not sure if this is still the issue, but I have reported the Pollock slowness some time ago (about half a year ago). Its cause was in LayoutFrame machinery. Every time you click somewhere (maybe even move the mouse), the system has to determine on which widget you actually clicked. This eventually leads to computing widget bounds by asking frames to do it. And because frames do not cache them, they are computed each time. This is problem especially with FractionalFrame and RelationalFrame, which in order to compute their bounds have to ask its parents and/or even neighbours... I was trying to solve it myself, but failed, because I don't understand Pollock internals that well... But basically all you have to do is add caching of frame bounds and properly invalidate them when they change... Hope this helps, Ladislav Lenart |
In reply to this post by Charles A. Monteiro-2
Charles A. Monteiro schrieb am 21.03.2006 15:33:
> I have attached a dialog we are getting in our VW 3.0 image. It seems to > me to be the typical out of memory dialog that the VM generates. As has already been noted, there's more than one possible reason for the error. Make sure the latest 3.1 VM for the respective platform is used, and that the space safety margins are set to large enough values. From the VisualWorks Memory Management Guide, v.7.2, p.8: "Finally, a certain portion of the free contiguous data is reserved for use by the object engine to ensure that it can perform at least one scavenge in extreme low-space conditions, thereby providing the system with one final opportunity to take the appropriate action." This was true in 3.x also. From the same manual, p. 9: "The remembered table (RT) is a special table that contains one entry for each object in Old Space or Perm Space that is thought to contain a reference to an object housed in New Space." In VW 3.x these tables could overflow, resulting either in an "out of memory" or a "remembered table overflow" error from the VM (I saw the "remembered table overflow" message only once with no way to reproduce it). The probability of this condition can be reduced by making OldSpace segments and space safety margins larger (we used to set availableSpaceSafetyMargin to 2.5 MB and contiguousSpaceSafetyMargin to 1.5 MB for an application producing large numbers of temporary objects) and by enlarging Eden and SurvivorSpace. If you enlarge Eden and SurvivorSpace e.g. by a factor of 10.0 to 20.0, fewer temporary objects will be tenured from NewSpace to OldSpace, thus reducing the probability that tenured objects reference objects in NewSpace. If you force the image to return empty OldSpace segments to the operating system, there can be another type of error in 3.x and some later versions of VW. It can happen that you return all empty OldSpace segments to the OS, and that the space safety margins were located in one of these segments. In extreme cases, all the remaining OldSpace segments can be full to the brim. This will lead to an out of memory error during the next run of the scavenger. When freeing memory, you should force the allocation of a new OldSpace segment if there is not enough free memory in the remaining segments. Starting with 7.0 or 7.1 (I don't remember the exact version), it seems to be necessary that the newly introduced MemoryPolicy parameter freeMemoryUpperBound is set to a value higher than the preferredGrowthIncrement. It's not stated explicitly in the documentation, and I can't actually prove this, but we had occasional out of memory errors when we raised preferredGrowthIncrement to 10 MB, while leaving freeMemoryUpperBound at its default 8 MB. Raising freeMemoryUpperBound to 15 MB solved the problem. I hope this helps. If it doesn't and you must make sure that the problem is resolved, you might want to hire John McIntosh. It helped in our case, and was worth every cent invested. ;-) Of course, the right solution would be to port to VW 7.x - and hire John to tune the MemoryPolicy... ;-) Joachim Geidel |
In reply to this post by Samuel S. Shuster <sames@interaccess.com>
On Mar 23, 2006, at 2:40 PM, Samuel S. Shuster <[hidden email]>
wrote: >> 2) There is no focus indicator for listboxes (also for Motif look). > > Do you mean items don't highlight? There is no other focus indicator > that I'm > aware of. No, I mean there is no focus indication. In Windows when a listbox has input focus the selected item will have a dotted outline, and if there is no selected item the first item will still have a dotted outline. As soon as you press the space bar or arrow keys the listbox will make a selection, or you can press a character key and the first item with that character will be selected. Press that character again and the next matching item will be the selection, etc. When I try this last thing in a Pollock listbox the whole list portion of the widget (not the scrollbar) flashes black and then redisplays. >> 1) Motif: The button controls aren't as tall as they should be. > > I don't get what you mean here. They're as big as you make 'em. Motif > Buttons do > have a 7 pixel border around the "button" part, that cut into the > inner bounds. > That's the way it is... However, I've been thinking of getting rid of > that all > together and just copying the Win9x look for buttons in Motif. Haven't > decided > yet. Well, perhaps that's all it is, that the border is stealing space from the label. Thanks, -Carl |
If I select some text in a TextEdit control by dragging up or down with
the mouse, and reverse direction before letting go of the mouse button the selection doesn't get cleaned up. It leaves a mess behind. The same thing happens if I use Shift and up arrow or down arrow keys. Also, I have been unable to make the accelerators work for menu items. For example I might define the standard Ctrl+N to make a new document for my editor, but it does nothing. Here's the code that defines the menu item. Am I doing something wrong? addMenuItem: ((Panda.MenuItem label: '&New File') keystroke: (Panda.Keystroke ctrl: $n); actionID: #newFile); Thanks, -Carl Gundel, author of Liberty BASIC http://www.libertybasic.com |
Carl Gundel,
>If I select some text in a TextEdit control by dragging up or down with >the mouse, and reverse direction before letting go of the mouse button >the selection doesn't get cleaned up. It leaves a mess behind. > >The same thing happens if I use Shift and up arrow or down arrow keys. AR50562 for this one. >Also, I have been unable to make the accelerators work for menu items. >For example I might define the standard Ctrl+N to make a new document >for my editor, but it does nothing. Here's the code that defines the >menu item. Am I doing something wrong? > > addMenuItem: ((Panda.MenuItem label: '&New File') > keystroke: (Panda.Keystroke ctrl: $n); > actionID: #newFile); I assume you added an Action with the id #newFile somewhere. The issue is where. If this is a MainMenu item, then it needs to be added to the Window itself. And So It Goes Sames ______________________________________________________________________ Samuel S. Shuster [|] VisualWorks Engineering, GUI Project Smalltalk Enables Success -- What Are YOU Using? |
VW 7.4 changed the directory structure of parcels, specially 'goodies'
now 'contributed'. However, the ColorEditing parcel that used to be under 'parc' subdir in goodies, is not available anymore. Is it because there is a new clever way to obtain this? Otherwise shall I download from Cincom repository or loading a parcel from an ealier release of VW would suffice? TIA -- Cesar Rabak GNU/Linux User 52247. Get counted: http://counter.li.org/ |
Free forum by Nabble | Edit this page |