Improper key passed to nsHttpChannel::DoInvalidateCacheEntry()?
During the cache2 review in #13035 (moved), mcs noticed that an empty key was being passed to nsHttpChannel::DoInvalidateCacheEntry().
nsHttpChannel::DoInvalidateCacheEntry() to use our modified (isolated) cache keys. That would involve passing a non-empty string as the second parameter to cacheStorage->AsyncDoomURI() within that method. This is not new code and not something we patched in the past... and Kathy and I do not understand the implications of not patching it. But it seems like the wrong key is being used there.
I have not dug through all of the eviction code (there sure are a lot of codepaths involved there), but my initial take is that since Mozilla has been using this same extension key to isolate caching for POST requests, it probably is not a serious issue to omit it, since the original code would have been experiencing similar problems even before our isolation made further use of this key...
We should ask Mozilla for an opinion. This may be a bug in their code, too.