error: revert mapping EHOSTUNREACH/ENETDOWN

Earlier commits

  - 45449e0 ('Add EHOSTUNREACH to nl_syserr2nlerr')
  - fe74608 ('Add ENETDOWN to nl_syserr2nlerr')

introduced NLE_HOSTUNREACH and NLE_NETDOWN. So now failures with those
error codes would no longer return NLE_FAILURE. This is a desirable
change (in principle), but also an API break.

As it causes problems ([1]), gut the change for now. We keep the error
defines (and the assigned NLE error numbers), but we no longer map the
errno.h codes, which in practice reverts those two changes.

This is only an intermediate solution, to revert the break. We should
bring this change back, but somehow "better". It's not clear how exactly
we can improve the API without breaking it. For example, we could
identify a few old functions which must stick to the old behavior, while
new functions get the new error codes. Or maybe better, selected
functions should guarantee that when a function returns NLE_FAILURE that
errno is set to a value that is relevant to the underlying.  failure.

[1] #433 (comment)

Author: thom311

lib/error.c

@@ -106,8 +106,21 @@ int nl_syserr2nlerr(int error)
 	case EBUSY:		return NLE_BUSY;
 	case ERANGE:		return NLE_RANGE;
 	case ENODEV:		return NLE_NODEV;
-	case EHOSTUNREACH:	return NLE_HOSTUNREACH;
-	case ENETDOWN:		return NLE_NETDOWN;
+
+	/* Historically, NLE_HOSTUNREACH and NLE_NETDOWN did not exit (it was
+	 * all NLE_FAILURE). Introducing a new error code later on was an API
+	 * change, which caused problems:
+	 *
+	 *   https://github.com/thom311/libnl/pull/433#issuecomment-3625794675
+	 *
+	 * For now, this change is undone by still map EHOSTUNREACH/ENETDOWN
+	 * to NLE_FAILURE.
+	 *
+	 * We should bring the change back in the future, but do it somehow(?)
+	 * smart to not break existing users. */
+	case EHOSTUNREACH:	return NLE_FAILURE;
+	case ENETDOWN:		return NLE_FAILURE;
+
 	default:		return NLE_FAILURE;
 	}
 }