(ITS#8192) liblmdb confuses Windows error code with errno values

[h.b.furuseth@usit.uio.no]

Full_Name: Hallvard B Furuseth
Version: LMDB_0.9.15
OS: Windows 7 with cygwin
URL: 
Submission from: (NULL) (81.191.45.5)
Submitted by: hallvard


mdb_env_write_meta() loops to retry_write if ErrCode() == EINTR.
That's ERROR_TOO_MANY_OPEN_FILES = 4 on Windows.  New in 0.9.15.

==

This one is older, in mdb_env_setup_locks().  Also wrong on Windows:
  rc = ErrCode();
  if (rc && rc != EACCES && rc != EAGAIN)

I don't know what that code is doing. I think that belongs under
  if ((rc = mdb_env_excl_lock(env, excl))) goto fail;
where it would read as
  if (*excl <= 0 && (rc = ErrCode()) && rc != EACCES && rc != EAGAIN)

==D%D

ENOENT confusion:

mdb_env_read_header() can explicitly return ENOENT, but only
as an internal code.  So mdb_strerror() should drop ENOENT.
We can rename/renumber it to avoid confusion:
  /* Internal result codes, not exposed outside liblmdb */
  #define	MDB_FILE_NEW	(MDB_LAST_ERRCODE + 9)
(above instead of MDB_NO_ROOT for easier cherry-picking.)

OTOH lmdb.h lists
 "ENOENT - the directory specified by the path parameter doesn't exist."
That's not related to the above.  It should be
"ENOENT (Unix) or ERROR_PATH_NOT_FOUND (Windows) - ..."

Or maybe the doc for this code should say something more
general, or simply be omitted.  I found the Windows code
by testing @ Windows.  Reading Windows doc did not help.

==

Finally, I suggest we replace the explicit EIO/ENOSPC returns
with either a new MDB code or #define MDB_EIO EIO on Unix
and something else on Windows.  (I think ENOSPC should be EIO
too, there could be other reasons for a short write.)
Then we can drop those from mdb_strerror() too.