Don’t use the global variable ‘skip’ to decode into, as this causes a data race, when decoding keys from inside multiple different goroutines. Instead we now create a new string variable for every decode function, which will be created on the heap and ignored when the function returns.
It is now possible to specify a function to manipulate each values stored in a data object, when retrieving that value. This enables storing compressed obejcts as strings, which are expanded when accessed. This is used to store *sql.Thing values, which when requested can access embedded fields by fetching and returning the full record when the *sql.Thing is requested.
Only store read-write statement status in the database for statements which can be either read or write. For those statements which are always write statements (Live, Kill, Create, Update, Delete, Relate, Insert, Upsert, Define, Remove….), then hardcode this on the statement itself.
If a type only defines one-half of the symmetry (e.g. it implements MarshalJSON() but not UnmarshalJSON() ), then that type doesn't satisfy the codec encoding/decoding check and instead codec will not encode or decode the item using the special interface methods.
If a type only defines one-half of the symmetry (e.g. it implements MarshalJSON() but not UnmarshalJSON() ), then that type doesn't satisfy the codec encoding/decoding check and instead codec will not encode or decode the item using the special interface methods.
The executor was being returned to the pool twice, which was resulting in nil interface{} errors when creating new executors concurrently, as the transaction would not be set properly.
Deadline timeouts are not useful, as they do not give any meaningful http error responses to the end user. Instead we now make use of contexts throught the system to detect query timeouts, and send error messages accordingly.
