I'll repost and expand upon my comment from above so that this question can have an accepted answer.
I recently came across a case similar to this when building my GPUImage framework. It appears that the build process is slightly different when archiving than when building and directly installing an application on the device via Xcode.
This can be exposed by building and archiving the application, then choosing to distribute it for ad hoc or enterprise distribution. Take the .ipa and place it in iTunes and manually load it on one of your test devices that way. The behavior of an application prepared in this manner may differ from one built and installed through Xcode, and should be closer to how a build submitted for review will act.
In my case, the problem was due to a lack of proper weak linking. On SDKs newer than iOS 4.3, you should no longer need to weak link whole frameworks if you wish to conditionally use classes and functions that are present in newer SDKs but missing in older ones. If you target 4.0 and higher, the linker should now perform class- and function-level weak linking.
However, this was failing for people using my static library, which does runtime checks for the presence of the new texture cache functions in iOS 5.0, but it only failed in these archived builds. I never saw it in all my testing against 4.x devices, because that was done by installing via Xcode. In the end, I needed to have users explicitly weak link the entire Core Video framework in order to get applications using this framework to run properly when archived and installed via iTunes.
Note that this wasn't due to the Release vs. Debug build configurations, because I tried switching between these in my build schemes when deploying to the device through Xcode and it made no difference there. Something else is different in the way that archived applications are built and linked.
与恶龙缠斗过久,自身亦成为恶龙；凝视深渊过久,深渊将回以凝视…
