Leonid wrote:
On Wednesday, May 16, 2007 5:26 AM Wolfgang Denk:
Because command identification are 64bit value and U-BOOT has 63 commands now. We need extend unsigned long long value.
Unfortunately the C standard does not define any longer integer data types that can be used directly in the C preprocessor.
What proper solution is?
Get rid of the existing coee and come up with a completely new implementation. But this is a non-trivial task.
There is actually an approach in place, virtually converting single u-boot command into root of command tree. See for example how "nand" command (or shall I say "nand" tree?) is implemented. Wolfgang, do you approve this way of doing or it had sneaked into u-boot "illegally"?
Of course, this is only half-solution which can provide temporary relief. As a matter of fact, I'm using it myself for my proprietary commands.
In my mind, there are two (not mutually excluding) generic ways to proceed (backward compatibility will be preserved of course meaning old scripts will be still working).
- Strictly hierarchical "CISCO-like" CLI instead of "flat" u-boot
scheme. There are several existing implementations of such CLI which can be used.
- Further advance of "bash"-like shell (hush is used in u-boot right
now). Existing shell lacks many features one is used to see in Linux/Unix shells. It's highly questionable though whether it makes sense to add u-boot specific commands in there, making u-boot shell scripts incompatible with Linux ones.
Best regards,
Leonid.
Hi Leonid,
I think you are missing the fundamental part of the problem: u-boot uses a #defined bitmap to enable and disable commands at compile time. The #defined bitmap can hold, at most, 64 bits and 63 of those bits are used. The fundamental limitation stems from the desire to enable/disable commands at compile time in conjunction with how many bits gcc (actually, the preprocessor) supports for #if conditional compilation. There are also implicit desires to use & and | to combine the #defined bit flags.
This has come up a couple of times, but no good solution has shaken out.
At one time I proposed simply creating a second set of 64 bit command enable/disable #defines, but Wolfgang wasn't too keen on that solution. I really cannot think of any other way to maintain the current method of compile-time selection and add expansion. Discussion thread here: http://thread.gmane.org/gmane.comp.boot-loaders.u-boot/24647
I personally sidestepped the issue by enabling and disabling my new "fdt" command based on whether CONFIG_OF_LIBFDT was defined or not, thereby not needing to use the last command control bit in the #define.
gvb