Provide public sdb_ref sdb_unref

[ret2libc]

I think we should introduce new APIs to increment/decrement the ref value of a Sdb. This can be very useful in radare2 where sdb are passed around and sometimes it's not clear which component will be the first to release a given Sdb.

This seems the case with RBinFile, which has a ->sdb field which is not freed in r_bin_file_free because it is already freed by someone else. This leads to memory leaks in some cases and in general I think it is not good practice to have some fields of a data structure owned by something else.

[radare]

Agree. Same is hapoening in anal with all the bb and fcn info that can be shared across miltiple structures and result in crash or leaks. Debugging reference counting is a pain in the ass but at least it will allow us to handle those chaotic meshed uses of multiple sdbs being referenced as namespaces or between multiple places. I see refcounting also helpful for multithread to avoid races when using unowned structs

…

On 24 Oct 2019, at 12:41, Riccardo Schirone ***@***.***> wrote:  I think we should introduce new APIs to increment/decrement the ref value of a Sdb. This can be very useful in radare2 where sdb are passed around and sometimes it's not clear which component will be the first to release a given Sdb. This seems the case with RBinFile, which has a ->sdb field which is not freed in r_bin_file_free because it is already freed by someone else. This leads to memory leaks in some cases and in general I think it is not good practice to have some fields of a data structure owned by something else. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.