Hi, i'd like to ask, is it safe to clear the root flag for every object in system. I'm artificially creating a new image, and i'd like to know, what objects should be marked as roots initially, or VM can care about it itself?
The root bit must mark all objects, and the root table must contain all objects thata) are in old space and b) refer to objects in
i.e. to find all roots into youngSpace the GC must be able to find all candidates in the root table. The root bit is a flag that saves searching the root table when adding a new root to the root set.
So yes, when creating a new image, all objects are in oldSpace and so no objects need to be in the rot table or have their root bit set.
This raises another point of ignorance for me. I would have thought that it is OK for the root bit to be set or the root table to contain objects that are not in old space or do not/no longer refer to objects in youngSpace because the GC will remove these from the root table on GC. But I don't see where objects get cleared from the rootTable. I only see the RootTable being voided when it gets full (rather drastic). Can anyone say if this happens or not, and if so, where it happens?
Looks to me like either markAndTrace: or a specialized version, say markAndTraceRootObject:, should maintain a flag, refersToYoungObject, and clear the object from the rootTable (and clear the root bit) if refersToYoungObject is false at the end of the scan. But with the pointer reversal algorithm it is hard to see where to maintain this flag.
Can those who know enlighten this ignorant soul?
TIA
On Sat, Feb 21, 2009 at 4:42 AM, Igor Stasenko siguctua@gmail.com wrote:
Hi, i'd like to ask, is it safe to clear the root flag for every object in system. I'm artificially creating a new image, and i'd like to know, what objects should be marked as roots initially, or VM can care about it itself?
-- Best regards, Igor Stasenko AKA sig.
Eliot Miranda wrote:
So yes, when creating a new image, all objects are in oldSpace and so no objects need to be in the rot table or have their root bit set.
Correct.
This raises another point of ignorance for me. I would have thought that it is OK for the root bit to be set or the root table to contain objects that are not in old space or do not/no longer refer to objects in youngSpace because the GC will remove these from the root table on GC. But I don't see where objects get cleared from the rootTable. I only see the RootTable being voided when it gets full (rather drastic). Can anyone say if this happens or not, and if so, where it happens?
I think you're right. Short of a full GC the root table may never get cleared.
Cheers, - Andreas
Looks to me like either markAndTrace: or a specialized version, say markAndTraceRootObject:, should maintain a flag, refersToYoungObject, and clear the object from the rootTable (and clear the root bit) if refersToYoungObject is false at the end of the scan. But with the pointer reversal algorithm it is hard to see where to maintain this flag.
Can those who know enlighten this ignorant soul?
TIA
On Sat, Feb 21, 2009 at 4:42 AM, Igor Stasenko <siguctua@gmail.com mailto:siguctua@gmail.com> wrote:
Hi, i'd like to ask, is it safe to clear the root flag for every object in system. I'm artificially creating a new image, and i'd like to know, what objects should be marked as roots initially, or VM can care about it itself? -- Best regards, Igor Stasenko AKA sig.
On Sat, Feb 21, 2009 at 11:01 AM, Andreas Raab andreas.raab@gmx.de wrote:
Eliot Miranda wrote:
So yes, when creating a new image, all objects are in oldSpace and so no objects need to be in the rot table or have their root bit set.
Correct.
This raises another point of ignorance for me. I would have thought that
it is OK for the root bit to be set or the root table to contain objects that are not in old space or do not/no longer refer to objects in youngSpace because the GC will remove these from the root table on GC. But I don't see where objects get cleared from the rootTable. I only see the RootTable being voided when it gets full (rather drastic). Can anyone say if this happens or not, and if so, where it happens?
I think you're right. Short of a full GC the root table may never get cleared.
I think next week I'll instrument the VM and we should put it underneath a server image which we should run for some time ad then interrupt and print the occupancy of the root table. It could be that it is filling up with objects that are no loner roots and incremental collection is slowing down.
Cheers,
- Andreas
Looks to me like either markAndTrace: or a specialized version, say
markAndTraceRootObject:, should maintain a flag, refersToYoungObject, and clear the object from the rootTable (and clear the root bit) if refersToYoungObject is false at the end of the scan. But with the pointer reversal algorithm it is hard to see where to maintain this flag.
Can those who know enlighten this ignorant soul?
TIA
On Sat, Feb 21, 2009 at 4:42 AM, Igor Stasenko <siguctua@gmail.commailto: siguctua@gmail.com> wrote:
Hi, i'd like to ask, is it safe to clear the root flag for every object in system. I'm artificially creating a new image, and i'd like to know, what objects should be marked as roots initially, or VM can care about it itself?
-- Best regards, Igor Stasenko AKA sig.
Thanks for response guys! :) I'm happy to see that my humble question drawn attention to the ways how GC operating with root objects, and potentially may lead to some improvements :)
2009/2/21 Eliot Miranda eliot.miranda@gmail.com:
On Sat, Feb 21, 2009 at 11:01 AM, Andreas Raab andreas.raab@gmx.de wrote:
Eliot Miranda wrote:
So yes, when creating a new image, all objects are in oldSpace and so no objects need to be in the rot table or have their root bit set.
Correct.
This raises another point of ignorance for me. I would have thought that it is OK for the root bit to be set or the root table to contain objects that are not in old space or do not/no longer refer to objects in youngSpace because the GC will remove these from the root table on GC. But I don't see where objects get cleared from the rootTable. I only see the RootTable being voided when it gets full (rather drastic). Can anyone say if this happens or not, and if so, where it happens?
I think you're right. Short of a full GC the root table may never get cleared.
I think next week I'll instrument the VM and we should put it underneath a server image which we should run for some time ad then interrupt and print the occupancy of the root table. It could be that it is filling up with objects that are no loner roots and incremental collection is slowing down.
Cheers,
- Andreas
Looks to me like either markAndTrace: or a specialized version, say markAndTraceRootObject:, should maintain a flag, refersToYoungObject, and clear the object from the rootTable (and clear the root bit) if refersToYoungObject is false at the end of the scan. But with the pointer reversal algorithm it is hard to see where to maintain this flag.
Can those who know enlighten this ignorant soul?
TIA
On Sat, Feb 21, 2009 at 4:42 AM, Igor Stasenko <siguctua@gmail.com mailto:siguctua@gmail.com> wrote:
Hi, i'd like to ask, is it safe to clear the root flag for every object in system. I'm artificially creating a new image, and i'd like to know, what objects should be marked as roots initially, or VM can care about it itself?
-- Best regards, Igor Stasenko AKA sig.
vm-dev@lists.squeakfoundation.org
-
Andreas Raab -
Eliot Miranda -
Igor Stasenko