I have been looking at this again, with the macros

#define I2C_MSG_OP(_addr, _buf, _len, _flags) \
        { .addr = _addr, .buf = _buf, .len = _len, .flags = _flags }

#define I2C_MSG_WRITE(addr, buf, len) \
        I2C_MSG_OP(addr, buf, len, 0)
#define I2C_MSG_READ(addr, buf, len) \
        I2C_MSG_OP(addr, buf, len, I2C_M_RD)
#define I2C_MSG_WRITE_OP(addr, buf, len, op) \
        I2C_MSG_OP(addr, buf, len, 0 | op)
#define I2C_MSG_READ_OP(addr, buf, len, op) \
        I2C_MSG_OP(addr, buf, len, I2C_M_RD | op)

and the tuners files as a random first example.  This time I haven't made
any adjustments for arrays of size 1.

The following file is a bit unfortunate, because a single structure is
mostly used for writing, but sometimes used for reading, in the function
tda827xa_set_params.  That is, there are explicit assignments to
msg.flags.

drivers/media/tuners/tda827x.c

Currently, this is done in 8 files across the entire kernel.  Would it be
a problem to use a READ or WRITE macro to initialize such a structure,
give that the type is going to change?  Maybe the I2C_MSG_OP macro could
be used, with an explicit flag argument?

julia