Typo in original report:
#define MDB_FILE_NEW (MDB_LAST_ERRCODE + 9) (above instead of MDB_NO_ROOT for easier cherry-picking.)
"above instead of below MDB_NO_ROOT".
On 08/07/15 22:36, Howard Chu wrote:
When there is a platform-specific error code that describes the situation, we should use it. In particular, if the error code comes from an underlying system call and accurately describes that failure, we should pass it through.
Yes. I only meant to talk about where mdb.c explicitly picks errno codes, like it does with EIO/ENOSPC.
(And where it compares codes from system calls with errno codes.)
If the failure is LMDB-specific, and not related to any particular system call, then we should define an MDB_* code for it.