On Sun, Jul 29, 2012 at 9:55 AM, Marek Vasut marex@denx.de wrote:
Dear Otavio Salvador,
On Sun, Jul 29, 2012 at 4:00 AM, Stefano Babic sbabic@denx.de wrote:
Personally I prefer that the function names are the same and the implementation itself of the function hides the SOC details. In this way, we provide the same interface API to the user (=the board maintainer) and to the drivers that are surely shared between the MX28 and MX23.
Sure but the accessing structure is the same for MX233 and MX28 so makes sense to have it with SOC name. If we have some divertion here a ifdef will be need to handle.
And fill the files with gazilions of ifdefs, making them unreadable.
No; if we have too much difference we can move the structs to another header and keep one to "include the right" providing the layer and hidding it.
Am I missing something?
I also think we ought to try to split function implementation when it diverts much (as code of spl_mem_init does
spl_mem_init() does not. How?
The are code there would need many ifdefs to get working fine for both; this could be split on spl_mem_mx233.c and spl_mem_mx28.c and spl_mem.c doing the generic part and calling the specifics.