#7195 BLOC Never A: Corrupted multiple downloads at the same time

[Zarro Boogs per Child]

#7195: Corrupted multiple downloads at the same time
------------------------+---------------------------------------------------
  Reporter:  cjb        |       Owner:  erikos        
      Type:  defect     |      Status:  new           
  Priority:  blocker    |   Milestone:  Never Assigned
 Component:  datastore  |     Version:  Update.1      
Resolution:             |    Keywords:                
  Verified:  0          |    Blocking:                
 Blockedby:             |  
------------------------+---------------------------------------------------

Comment(by tomeu):

 Replying to [comment:1 mstone]:
 > After reviewing the Browse source code, I think I see two problems in
 downloadmanager.py:
 >
 > 1. According to the log, we're creating two DownloadManagers. Why on
 earth are we creating two DownloadManagers?

 DownloadManager should have been renamed to HelperAppLauncherDialog or
 something similar. We are not implementing our own nsIDownloadManager
 anymore.

 http://www.xulplanet.com/references/xpcomref/ifaces/nsIHelperAppLauncherDialog.html
 http://www.xulplanet.com/references/xpcomref/ifaces/nsIDownloadManager.html

 Mozilla thinks that the user wants to download the same file twice, so
 asks us twice where each file should be saved to. And we seem to tell it
 to save to the same location.

 > 2. Empirically, we're seeing interleaved writes. However, all data
 transfer is done by xulrunner; we just supply it a target path.
 Consequently, I infer that we're creating two Downloads (Transfers) to the
 same path. We should never ever have two transfers to the same target path
 if that target path is a local file.

 Right.

 > Next question: did I miss something? Is there some reason why the two
 DownloadManagers we create are going to return distinct file paths?

 Nope.

 > Final question: what should we do? My preference at the moment is to use
 a synchronization mechanism (e.g. a lock) to deduplicate Transfer target
 paths. Comments?

 Any problem with just using mkstemp?

-- 
Ticket URL: <http://dev.laptop.org/ticket/7195#comment:3>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system