Squeak-dev June 2005

squeak-dev@lists.squeakfoundation.org

148 participants
234 discussions

Re: SharedQueue issues
by Lex Spoon 25 Jun '05

25 Jun '05

> I believe the problem here is that flushAllSuchThat: is now waiting > on readSynch, but the signal it needs > is waiting on accessProtect (pending nextPut:) before doing the > readSynch signal. > > Classic deadlock situation. What a bear. It required an extra cup of coffee to figure out how SharedQueue is even trying to work, much less how to fix it. This message has my analysis of the situation, but the punchline is very simple: let's replace SharedQueue with a new class as described in issue #1375. Semaphore is just too hard to use correctly. The problem John's detective work has found is that flushAllSuchThat: is grabbing accessProtect, but other methods such as #next remain free to decrement readSynch even though they don't hold accessProtect. This leads to deadlocks. That's the short of it. I have written up a longer description of the current implementation's strategy down below. It might make a good addition to the class comment for SharedQueue, if we decide to keep the current implementation around. One thing that gave me pause, by the way, is that the "flush" methods do not do what I think of as "flushing" -- instead, they are "removing" items. "Flush" normally means to push data downstream, not to destroy data completely. It may be better to name these methods with "remove" instead of "flush". The problem John has found applies more widely than just flushAllSuchThat:--it also impacts nextOrNilSuchThat: and nextOrNil. I'll call these methods "the Bad 3". All of the Bad 3 share the habit of checking readPointer vs. writePointer for available items *before* waiting on readSynch. This is bad because some other process may have done the right thing and waited first on readSynch. Thus, one of the items that the Bad 3 *think* is available based on readPointer vs. writePointer, may have already been reserved by another process by waiting on readSynch. This observation leads to a simple exhibition of the problem that I have appended below. It uses only next, nextPut:, and nextOrNil. It takes advantage of the fact that nextOrNil does not decrease readSynch at all unless it clears it completely to 0; this can be used to cause readSynch to have more signals on it than there are actual items in the queue. Here's a solution strategy based on the above analysis. Whenever one of the Bad 3 wants to consider removing an element, it should first try to decrease readSynch. If the attempt succeeds, then it has full rights to the next item in the queue and can procede. If the attempt fails, then there are no more unreserved items in the queue and the method should stop its work. After doing their normal work, the Bad 3 methods should all increase readSynch by the number if items they reserved but did not actually remove, so that other callers may access those items. Unfortunately, class Semaphore does not sufficiently support "try to wait, but it's okay if there aren't any signals available". It has waitTimeoutSeconds:, which presumably you could use with a timeout of 0, but that is not quite enough because you cannot tell whether it timed out or ate a signal. This functionality is key to the solution strategy I proposed, and so I am giving up on this solution for now. Given all this analysis, I see three possible ways forward: 1. Make a new SharedQueue based on Nathanael's Monitor implementation. 2. Improve class Semaphore as described above, so that the all-semaphores approach can be made to work. 3. Come up with another solution strategy that can use the current Semaphore class. For example, it may work to have all the reading methods (the Bad 3 and also #next) poll whenever they successfully wait on readSynch but then discover that no items are available. My inclination is to try #1..... And 30 minutes later, I have done so. Check out Mantis bug #1375. I'll post this message anyway for the record, in case someone wants to perservere with #2 for some reason. -Lex CURRENT IMPLEMENTATION STRATEGY SharedQueue works like this: contentsArray: holds the actual items. nil entries are unused (and so, nil cannot be sent across a SharedQueue). the buffer is NOT circular. readPointer: the pointer into contentsArray to start reading the next element. nil's are skipped. writePointer: the pointer where the next element should be placed. readSynch: a semaphore holding, roughly, a count of the number of items available to read. To be precise, it holds a number less than or equal to the number of available items; if it holds a lower number, then the delta by which it is lower is the number of calls to #next (and similar methods) that have registered to remove an item but have not yet removed it. accessProtect: a semaphore used for mutual exclusion. it must be held by code using or modifying contentsArray, readPointer, or writePointer. ==== EXAMPLE CODE ==== "exhibits a problem where nextOrNil does not properly decrease readSynch" q := SharedQueue new. q nextPut: 1. q nextPut: 2. "readSynch has 2 signals now" q nextOrNil. "readSynch *still* has 2 signals" [ r1 _ q next ] fork. [ r2 _ q next ] fork. "the second #next will think there is an item available. BOOM." (Delay forSeconds: 1) wait. "the below code finishes the exercise, but the code has already crashed by this point..." q nextPut: 3. {r1.r2}

1 0

External call failed
by Himanshu Garg 25 Jun '05

25 Jun '05

Hello, I have a function fnTest which is written in a MS Windows DLL. It works fine when I call it from another exe. When I call it from my subclass of Object in the following method I get externalCallFailed error. --------------------------------------------------------------------- go "Call fnTest from PangoDll" <apicall: void 'fnTest' (void) module:'pangodll.dll'> ^super externalCallFailed --------------------------------------------------------------------- How do I figure out what the problem is. The call stack is given at the end of this mail. Thank You, Himanshu. P.S. Call stack when go is called:- Error: Unable to find function address 25 August 2005 1:21:40 pm VM: Win32 - a SmalltalkImage Image: Squeak3.7 [latest update: #5989] SecurityManager state: Restricted: false FileAccess: true SocketAccess: true Working Dir D:\squeak Trusted Dir D:\squeak\printer Untrusted Dir C:\My Squeak\printer ExternalFunction class(Object)>>error: Receiver: ExternalFunction Arguments and temporary variables: aString: 'Unable to find function address' Receiver's instance variables: superclass: ExternalObject methodDict: a MethodDictionary(#argTypes->a CompiledMethod (1544) #callingConve...etc... format: 136 instanceVariables: #('flags' 'argTypes') organization: ('initialize-release' initialize) ('accessing' argTypes flags mod...etc... subclasses: #(ExternalLibraryFunction) name: #ExternalFunction classPool: a Dictionary(#FFIErrorMessages->a Dictionary(-1->'Callout mechanism ...etc... sharedPools: an OrderedCollection(FFIConstants) environment: a SystemDictionary(lots of globals) category: nil ExternalFunction class>>externalCallFailed Receiver: ExternalFunction Arguments and temporary variables: errCode: 13 Receiver's instance variables: superclass: ExternalObject methodDict: a MethodDictionary(#argTypes->a CompiledMethod (1544) #callingConve...etc... format: 136 instanceVariables: #('flags' 'argTypes') organization: ('initialize-release' initialize) ('accessing' argTypes flags mod...etc... subclasses: #(ExternalLibraryFunction) name: #ExternalFunction classPool: a Dictionary(#FFIErrorMessages->a Dictionary(-1->'Callout mechanism ...etc... sharedPools: an OrderedCollection(FFIConstants) environment: a SystemDictionary(lots of globals) category: nil Pango(Object)>>externalCallFailed Receiver: a Pango Arguments and temporary variables: Receiver's instance variables: a Pango Pango>>go Receiver: a Pango Arguments and temporary variables: Receiver's instance variables: a Pango -- (9885420248)

2 1

SqueakSource Suggestion
by Daniel Salama 24 Jun '05

24 Jun '05

At the expense of sounding selfish, please be sure that I mean no offense nor disrespect. I don't know if anyone has requested this before, but, it would be really nice if you could "search" for a package in squeaksource. Thanks, Daniel

5 7

Syntax highlighting in 3.8 full
by Andreas Raab 24 Jun '05

24 Jun '05

Hi - How do I change the rules for syntax highlighting in 3.8 full? I really don't like the way this looks. Is there a user interface for it? Thanks, - Andreas

3 4

Beware, squeak.org repointed!
by Torsten Bergmann 24 Jun '05

24 Jun '05

>There is still work to be done. I am compiling some screenshots, for > instance. Hi Jason, maybe http://minnow.cc.gatech.edu/squeak/5715 can help you. I've found most of them using Googles image search. Bye Torsten -- Geschenkt: 3 Monate GMX ProMail gratis + 3 Ausgaben stern gratis ++ Jetzt anmelden & testen ++ http://www.gmx.net/de/go/promail ++

1 0

Re: Why doesn't Morph's drawOn get called when Main Squeak
by Himanshu Garg 23 Jun '05

23 Jun '05

>Am 21.06.2005 um 18:52 schrieb Himanshu Garg: >> Hello, >> >> I have subclassed Morph and overridden its drawOn. I found that >> it doesn't get called when the main Squeak window is maximized >> after minimization or when it becomes visible after an overlapping >> window moves away. >> >> What is the function that I must override so that redrawing >> takes place even on unminimization or movement of overlapping >> window. >> >> I am drawing text using an external dll with FFI which directly >> draws in the Squeak window. Any suggestions/pointers would be a >> great >> help. It isn't necessary that I use Morphic if it is not possible with >> it. >[snipped] >If you draw yourself via FFI, you also need to get hold of the redraw >events yourself. There is no support for this in regular Squeak. >- Bert - Thanks for your reply. Could you also tell me how to get hold of the redraw events. In particular what is the event that gets generated when only some parts of an inactive window get exposed on Windows. If I do get hold of them somehow will it not affect Squeak's handling of these events. The host window stuff might not be for me because I am on Windows. Thank You, Himanshu. P.S. Sorry for starting a new thread. I get daily digests and don't know how to reply in the same thread.

2 1

install gardner
by marcelo Cortez 23 Jun '05

23 Jun '05

hi goran ,folks what is procedure to install gardner ? seaside version or smallwiki in your repository? what order of packages. is last version of seaside compatible? squeak version , etc tia. best regards mdc ___________________________________________________________ 1GB gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del mundo http://correo.yahoo.com.ar

1 0

Squeak Chat Monday June 27th 05:00 GMT
by Ken Causey 23 Jun '05

23 Jun '05

Just a reminder that the next Squeak Chat will take place between 05:00 (5 AM) and 09:00 (9 AM) GMT Monday June 27th, 2005. As usual we will meet in #squeak on irc.freenode.net. More information about the #squeak IRC channel can be found at http://people.squeakfoundation.org/article/7.html This is a good time to contribute to the development of Squeak by participating in the Harvesting Process. One of the main ways to contribute is by Reviewing submitted fixes and enhancements. For information about the Harvesting Process see http://minnow.cc.gatech.edu/squeak/3152 If you have the time please attend and work toward the future of Squeak. The next seven chats are scheduled for 09:00 (09:00 AM) GMT Friday July 01, 2005 13:00 (01:00 PM) GMT Tuesday July 05, 2005 17:00 (05:00 PM) GMT Saturday July 09, 2005 21:00 (09:00 PM) GMT Wednesday July 13, 2005 01:00 (01:00 AM) GMT Monday July 18, 2005 05:00 (05:00 AM) GMT Friday July 22, 2005 09:00 (09:00 AM) GMT Tuesday July 26, 2005

1 0

Re: Fixing the look of squeak in 3.9.
by John Pierce 23 Jun '05

23 Jun '05

Hi Romain, I've incorporated your latest suggestions to fix some of the buttons on common windows. The latest LookEnhancements (v16) is on http://squeak.saltypickle.com. I agree that many more overrides are apparently necessary to modify the look across the system. This is where in the interest of good style it will require more and more energy to update that last pixel to be consistent across all Morphs in the system, and eventually the payoff will be limited, but these updates look simple enough. For now, I'll keep taking more enhancements from the community and apply them to the look. For the record, getting an MCZ will make my life easier in the future instead of change sets. Thanks again for the updates -- they look good with square buttons. Regards, John On 6/21/05, Romain Robbes <romain.robbes(a)lu.unisi.ch> wrote: > > Then you might want to try this changeset, which modifies a few more > buttons (debugger, file list, test runner). > There's a lot of overrides in this, but it's difficult to do otherwise... > > Romain > > > On Jun 21, 2005, at 3:15 PM, John Pierce wrote: > > Thanks! The buttons look great! I will integrate these changes. > > On 6/21/05, Romain Robbes <romain.robbes(a)lu.unisi.ch > wrote: > > > > Hi John, > > > > Here's a small CS modifying buttons in the browser so that they > > fit well with your LookEnhancements package. > > If you find the changes worthwile, feel free to include them in > > your package. I also provide you with a screenshot, so that > > you can dismiss this quicker if you find it awful ;-). > > > > Cheers, > > Romain > > > > > > > > > > > > > > > -- > It's easy to have a complicated idea. It's very very hard to have a simple > idea. -- Carver Mead > > > > > -- It's easy to have a complicated idea. It's very very hard to have a simple idea. -- Carver Mead

1 0

generated 64 bit image does not work
by Martin Kuball 23 Jun '05

23 Jun '05

Hi! I tried the SystemTracer64 to convert one of my images to 64 bit format. Whe I tried to start it with a 64 bit VM I got the following message: This interpreter (vers. 68000) cannot read image file (vers. 292057776128000).

2 3

← Newer
1
2
3
4
5
6
7
8
...
24
Older →

Jump to page: