Marco: thank you for this. A few weeks ago when I was trying to find the "lock"
problem I saw this method but I did not dare to change it; and just as you said
I been using OminBase with no problem, but never runned the Garbage Collector,
Surely I was going to have a bigger problem in the future for not running the
garbageCollect.
I have just update this method on my project and tried both the garbageCollect
and the reorganize method , both worked fine (win 7 32bits, later I am going to
test on Win 10 64 bits).
Thanks again.
Matías.
________________________________
De: Pharo-users <pharo-users-boun...@lists.pharo.org> en nombre de marco
<ma...@omeleon.de>
Enviado: miércoles, 25 de julio de 2018 10:23 a. m.
Para: pharo-users@lists.pharo.org
Asunto: Re: [Pharo-users] OmniBase for Pharo 6
One serious problem left: OmniBaseTest>>testGC fails.
Have a look at the 2 methods
/ODBContainer>>lockDataFile
"Dec, 2010 sas: this code was relying in evaluating a block that was
happening only in win32
but not in *nix. So now I'm making it wait for the lock and then execute
the code normally
because it seems the intention was the same."
| currentSpace defaultSpace |
activeDataFile waitForAddingLock.
currentSpace := activeDataFile == dataFileA ifTrue: [0] ifFalse: [1].
defaultSpace := objectManager defaultObjectSpace.
currentSpace == defaultSpace
ifFalse:
[defaultSpace == 0
ifTrue:
[dataFileA isNil ifTrue: [dataFileA := ODBObjectStorage
openOn: self dataFileNameA].
activeDataFile := dataFileA]
ifFalse:
[dataFileB isNil ifTrue: [dataFileB := ODBObjectStorage
openOn: self dataFileNameB].
activeDataFile := dataFileB].
^self lockDataFile]
ODBFile>>waitForLockAt: pos length: length whileWaitingDo: aBlock
"Wait for lock at given position.
Evaluate aBlock in each iteration."
"Dec, 2010. sas: evaluating code in each iteration while waiting sounds
weird but whatever.
I think this was designed on the wrong assumption that the OS always
will provide a false
in the first try but what actually happens is that in win32 you always
will
get at least one false but in *nixes you don't do see that happening
(which is totally reasonable).
BEWARE of stupid code relying on this stupid assumption (I've already
patched #lockDataFile because of this)."
| startTime currentTime |
(stream lockAt: pos length: length) ifFalse: [
startTime := Time totalSeconds.
[stream lockAt: pos length: length] whileFalse: [
aBlock value.
currentTime := Time totalSeconds.
currentTime - startTime < self timeOutLength
ifFalse: [
ODBLockNotification signal
ifTrue: [startTime := currentTime]
ifFalse: [^ODBCannotLock signal]]]]
/
So "sas" did a refactoring in 2010 because he/she thought it is stupid code.
This code change breaks the GarbageCollection of Omnibase.
I reverted the first method to the original code of the Omnibase author,
David Gorisek:
/lockDataFile
| currentSpace defaultSpace |
activeDataFile waitForAddingLockWhileWaitingDo:
[currentSpace := activeDataFile == dataFileA ifTrue: [0] ifFalse: [1].
defaultSpace := objectManager defaultObjectSpace.
currentSpace == defaultSpace
ifFalse:
[defaultSpace == 0
ifTrue:
[dataFileA isNil ifTrue: [dataFileA := ODBObjectStorage
openOn: self dataFileNameA].
activeDataFile := dataFileA]
ifFalse:
[dataFileB isNil ifTrue: [dataFileB := ODBObjectStorage
openOn: self dataFileNameB].
activeDataFile := dataFileB].
^self lockDataFile]]
/
I removed the comment of "sas" from the second method. It is wrong.
The garbage collection works again and the test is green. GC is an important
feature of Omnibase.
It creates a new repository/database file and copies the data into the new
file, compressing the data.
It keeps your database compact and clean. If you don't run it (frequently)
your db files keep on growing...
Matias, your fixes are absolutely valid and should be published - together
with the one above - in Esteban's repository.
--
Sent from: http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html
Smalltalk - Pharo Smalltalk Users | Mailing List
Archive<http://forum.world.st/Pharo-Smalltalk-Users-f1310670.html>
forum.world.st
Pharo Smalltalk Users forum and mailing list archive. Pharo User Forum
