Wolfgang Denk wd@denx.de wrote on 2010/12/06 23:36:21:
Dear Joakim Tjernlund,
In message OFEE7ADEEA.47556DD1-ONC12577F1.006FAF8D-C12577F1.0071C097@transmode.se you wrote:
This is indeed much uglier. What exactly is your problem with debugging the existing code?
Whenever I screw up so that one of the init funcs crashes, often without any trace on the RS232 port you don't know which one. Single stepping
Well, we turn on the serial ports as early as possible, so ther eis only a pretty small collection that is candidate for such a situation.
But that won't help much. When you don't have any output it can be any reason. I really want to see where it breaks down. That will tell me more about what I did wrong earlier. It is easy to guess where when you a playing with these function, but when you are trying to understand why your latest relocation fixes doesn't work, it is a different game.
though the loop is cumbersome and not easy as BDI tends to flush the cache when it stops so you loose your stack.
Does it? On which architecture / processor is this?
MPC8321 and BDI2000. I have to play games with some internal BDI command called SAP. Maybe they have fixed this in later models?
The other way is to look up one those funs and set a BP there and hope for the best. Then repeat with the next function and so on. Compare that with just setting a BP in the new init_sequence(), it is fast and easy to move around.
...but ugly to read. And if you really want to introduce this style, it has to be done for all architectures, as I want to see this code to become common across architectures.
That is the next issue, lets not stray from the path ..
I don't think you have been chasing bugs in this area for a long time, if you had, you would appreciate how easy it is with functions compared with a bunch of function ptrs.
I haven't, lately, that's true, or when I had, I always knew pretty well where I had to expect the issues. But I've been there before, many, many times.
And you never found it annoying when you hit these function ptrs?
I don't think this is much uglier, just a bit, but far more useful and I have a hard time buying into "beautiful trumps usefulness".
I'm not convinced that your code is actually a general improvement. I understand that it's useful for you, and your style of debugging. If you are really hanging in this area, then maybe a debug() is more prowerfull - and less invasive.
Now you are proposing that I instrument the code, much like Scott and his print of address idea. This is exactly what I want to avoid as it would probably have to be adapted from case to case and there is always the case when there is no output and then I am back to square zero again.
The argument that it is uglier alone is not a very good one. Image the roles were reversed, I am sure you would dismiss me as "less beautiful" is really not an argument alone.
Jocke