August 2006 - Darwin-dev - lists.apple.com

Re: MACFORGE request for a project
by Shawn Erickson 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 28, 2006, at 9:12 AM, almisr wrote: i would like to open a " K&R removal " project ( Source Clean up ) aka Normalisé where/who ? Which source are you planning to touch? For this moment all old BSD base part, system_cmd ecetera -Shawn _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… -Shawn I don't see the point of doing that... since it quickly becomes a merge/diff nightmare with existing code branches. It would likely have to be Apple that took such steps to decide to lock down development on a segment of the code base to do reformatting (based on their corporate code standards). Don't forget that the code Apple is working of for Leopard, etc. is likely different in many places from what we have in the 10.4.7 source drop... making formatting changes against this code and attempting to merge that back into what Apple has can be difficult and error prone. This email sent to site_archiver(a)lists.apple.com

1 0

Re: MACFORGE request for a project
by almisr 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com i would like to open a " K&R removal " project ( Source Clean up ) aka Normalisé where/who ? Which source are you planning to touch? For this moment all old BSD base part, system_cmd ecetera Best Regards , {|} _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… -Shawn This email sent to site_archiver(a)lists.apple.com

1 0

Re: MACFORGE request for a project
by Shawn Erickson 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 28, 2006, at 5:34 AM, almisr wrote: i would like to open a " K&R removal " project ( Source Clean up ) aka Normalisé Which source are you planning to touch? -Shawn _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… where/who ? This email sent to site_archiver(a)lists.apple.com

1 0

MACFORGE request for a project
by almisr 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com i would like to open a " K&R removal " project ( Source Clean up ) aka Normalisé where/who ? Best Regards {|} _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: SQLite optimization
by Sjoerd Tieleman 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi Mark, On 28-aug-2006, at 12:48, Mark Gilbert wrote: Regards, Sjoerd. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… My app accesses the same SQlite database from several threads, and in order to accommodate this I currently open and close the database on each access, and the various threads will wait when the database is busy on another thread. This all works OK, but I am finding performance issues, which I think may be related to the opening and closing. My idea is to limit access to the database to a single intermediate thread, with each other thread passing their requests through the 'owner' thread for the SQLite database. My hope is that this thread can leave the database permanently open with a large cache size and consequently speed things up. Would anyone with experience in SQLite performance care to comment if this is a good strategy to improve performance ? Currently an open/access/close cycle is taking around 250mSecs, even though its a tiny database and fairly simple SQLite requests. I may need to do several database accesses per second, so currently its all stacking up. First of all, I don't think this is the correct place to ask that question. Secondly, this sounds an awful lot like you're trying to put a concurrency layer upon SQLite. The developers themselves note that applications requiring high concurrency should probably use a "real" RDBMS in favor of SQLite (http://www.sqlite.org/ whentouse.html). But if that is not an option, then this might just do the trick for you (by using some queueing mechanism that basically serializes database access). As you may know SQLite has a very pessimistic locking mechanism: any access to the database (read and/ or write) locks the entire database from other processess (read and/ or write). This email sent to site_archiver(a)lists.apple.com

1 0

SQLite optimization
by Mark Gilbert 28 Aug '06

28 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi There. Any comments ? Thanks -- mark(a)gallery.co.uk Tel: +44 208 340 5677 fax: +44 870 055 7790 http://www.gallery.co.uk _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… My app accesses the same SQlite database from several threads, and in order to accommodate this I currently open and close the database on each access, and the various threads will wait when the database is busy on another thread. This all works OK, but I am finding performance issues, which I think may be related to the opening and closing. My idea is to limit access to the database to a single intermediate thread, with each other thread passing their requests through the 'owner' thread for the SQLite database. My hope is that this thread can leave the database permanently open with a large cache size and consequently speed things up. Would anyone with experience in SQLite performance care to comment if this is a good strategy to improve performance ? Currently an open/access/close cycle is taking around 250mSecs, even though its a tiny database and fairly simple SQLite requests. I may need to do several database accesses per second, so currently its all stacking up. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Mount Points
by Kevin Elliott 27 Aug '06

27 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi Michael The problem I have is that I am using file:///Volumes/Drive/Folder/File style URLs to pass pointers to files over a network connection. When I use CFURL to resolve a URL like: file:///Volumes/G4_HD/folder/file As you can see, this is an inherently fragile way to do things. Even without the bug discussed above, I can't recommend this as a way to go. At 21:18 +0100 on 8/26/06, Mark Gilbert wrote: Thanks for the explanation about the phantom mount directory. Forcing an rm of the directory has solved the problem on this machine We need to pass around a file specification via TCP/IP between various different applications working on the same file from different computers. The file spec also has to go through an asset management system which needs to store the location as (unicode) text, then pop out the other side and be picked up again. What better mechanism would you recommend instead to identify a file which resides on a shared volume in a workgroup ?. One more note as to the problem your seeing- the precise problem your describing can easily occur in the wild if the user happens to mount two volumes with the same name. The first volume will get the normal volume, the second will have _# (some 1-x for each additional duplicate volume) appended. If robustness under lots of conditions is important to you, relying on the volume path being the same between machines is a bad idea... -- ______________________________________________________ Arguing with an engineer is like wrestling with a pig in mud. After a while, you realize the pig is enjoying it. ______________________________________________________ Kevin Elliott <mailto:kelliott@mac.com> AIM/iChatAV: kelliott(a)mac.com (video chat available) ______________________________________________________ _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Mount Points
by Michael Smith 26 Aug '06

26 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 26, 2006, at 1:18 PM, Mark Gilbert wrote: The problem I have is that I am using file:///Volumes/Drive/Folder/File style URLs to pass pointers to files over a network connection. When I use CFURL to resolve a URL like: file:///Volumes/G4_HD/folder/file = Mike _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… As you can see, this is an inherently fragile way to do things. Even without the bug discussed above, I can't recommend this as a way to go. We need to pass around a file specification via TCP/IP between various different applications working on the same file from different computers. The file spec also has to go through an asset management system which needs to store the location as (unicode) text, then pop out the other side and be picked up again. What better mechanism would you recommend instead to identify a file which resides on a shared volume in a workgroup ?. The file is uniquely identified by two things; the share identifier tuple (machine,sharename) and the share-relative path. Your current scheme makes the (unreliable) assumption that a given share will always be mounted at the same local path but (as you have discovered) offers no guarantee of that. A more robust approach would be to pass around the share identifier and the share-relative path, and then on each client system translate the share identifier into a local path at the time of use. This would also facilitate (re)connecting to the share if it's not currently mounted. One additional point to consider; the share mounted in /Volumes is nominally under the control of the user. If robustness is important to you (and user navigation of the mounted share less so) you might consider having your application explicitly mount the share somewhere else where it can guarantee the local path translation for a given network share. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Mount Points
by Mark Gilbert 26 Aug '06

26 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi Michael The problem I have is that I am using file:///Volumes/Drive/Folder/File style URLs to pass pointers to files over a network connection. When I use CFURL to resolve a URL like: file:///Volumes/G4_HD/folder/file As you can see, this is an inherently fragile way to do things. Even without the bug discussed above, I can't recommend this as a way to go. Cheers Mark -- mark(a)gallery.co.uk Tel: +44 208 340 5677 fax: +44 870 055 7790 http://www.gallery.co.uk _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Thanks for the explanation about the phantom mount directory. Forcing an rm of the directory has solved the problem on this machine We need to pass around a file specification via TCP/IP between various different applications working on the same file from different computers. The file spec also has to go through an asset management system which needs to store the location as (unicode) text, then pop out the other side and be picked up again. What better mechanism would you recommend instead to identify a file which resides on a shared volume in a workgroup ?. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Mount Points
by Michael Smith 26 Aug '06

26 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 26, 2006, at 12:02 PM, darwin-dev-request(a)lists.apple.com wrote: I am seeing a strange behaviour sometimes when mounting afp network volumes on an Intel iMac. Essentially the OS is creating more than one /Volumes representation for the drive, depending on what steps have occurred, and its creating a persistent entry in /Volumes which I can't 'unmount' For example: I have another Mac sharing its hard disc. This volume is called "G4_HD". When I try to mount it on the iMac, it ends up being referred to as G4_HD-1 by Unix (in /Volumes). This is because there is already a 'persistent' entry in /Volumes called 'G4_HD' from some earlier life. This persistent entry only contains one of the directories which belong to the volume. The fact that it contains another directory is a giveaway; what has happened is that the volume has been unmounted at the same time that an application has decided that it is going to create (e.g.) /Volumes/G4_HD/SomeFolder/SomeDocument and the application has stupidly assumed that it should create *any* intervening directories (it may even have asked the user) along the way. As the volume was unmounted, /Volumes/G4_HD ceased to exist, and so the application created it again. The problem I have is that I am using file:///Volumes/Drive/Folder/File style URLs to pass pointers to files over a network connection. When I use CFURL to resolve a URL like: file:///Volumes/G4_HD/folder/file So, I have 2 questions: 1) What is this phantom directory ? It's a directory. a) delete it b) find and either fix or avoid the application that created it = Mike _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… The problem isn't caused by you mounting them; it's an interaction between unmounting them and an application bug. The OS didn't create it, and you can't "unmount" it because it's a directory (folder). As you can see, this is an inherently fragile way to do things. Even without the bug discussed above, I can't recommend this as a way to go. 2) How do I solve my problem and avoid this in the field ? This email sent to site_archiver(a)lists.apple.com

1 0