On Thu, Mar 21, 2013 at 08:04:10AM +0100, Paul Gevers wrote: > >>> Maybe I've picked the wrong example, the actual problem was with > >>> libavdevice53, the same soname is used for either 0.8.5-1 and 9.3-1, > >>> but 9.3-1 links to libavcodec54 while 0.8.5-1 links to libavcodec53, so > >>> my binary ended up importing either libavcodec53 and libavcodec54 when > >>> both were installed and did not work. > >> If this is true, and I must say I may have had the same experience, > >> isn't this hinting at a bug in libavdevice53, not being stable enough to > >> keep the same SONAME? I think you should file a bug against libavdevice53. > > Why? The problem is caused by indirect dependency on both libavcodec53 and > > libavcodec54, not by ABI breakage. > Could you please explain what the problem is then becasue I don't see it > (I really want to learn here). As I understand it, if you get > libavdevice53 in a newer version it should still work with stuff that > was build against an older version, right? If it breaks, the soname > should be updated and the packagename changed, so that the libraries can > be co-installed. So I guess you see the problem elsewhere? If your app is linked against libavcodec53 and against libavdevice53 and libavdevice53 is linked against libavcodec54 and libavcodec* doesn't use symbol versioning, your app will crash because of symbol conflicts. -- WBR, wRAR
Attachment:
signature.asc
Description: Digital signature