Wolfgang Denk wrote at Thursday, October 20, 2011 3:47 PM:
In message 74CDBE0F657A3D45AFBB94109FB122FF173D51C286@HQMAIL01.nvidia.com you wrote:
DOn;t give up to oearly. My opinion is that #ifdef's are bad, and increased code size is bad either. If we can trade off one for the other, we should know exactly what we can win here and how much we have to pay for that there. That's all I want to know.
Do you mean the total without /any/ ifdefs? If so, please see where I answered this question earlier in the thread:
http://lists.denx.de/pipermail/u-boot/2011-October/104427.html
The first time I submitted this patch, there were no ifdefs, and you stated that the additional 576 bytes text plus 136 bytes rodata was unacceptable, so I rewrote the patch to add ifdefs everywhere.
The statement that triggered me was "Maybe relax that by a few 10s of bytes?", and my question was then ``how much is "a few 10s of bytes"''?
Obviously this must be something inbetween the 700+ bytes you state here, and the zero bytes you state elsewhere.
I wonder what exactly it was, then?
It could be anything.
There are many places in the patch where ifdefs are added. I think Simon's comment was a general one that /some/ of those ifdefs (say the ones that just add extra values in an if that checks an image header type) might be candidates for not ifdefing. However, I don't believe it was targeted at any one individual ifdef, or even a /specific/ set of ifdefs.
Hence, to answer the question of how many bytes we're talking about in that case, either:
* Someone would have define exactly which specific set of ifdefs we want to measure removing. It seems best for you to look at the patch and define that set, since neither I nor Simon are currently pushing removing any of the ifdefs.
Or:
* I'd have to build the code with every possible combination of each new ifdef that was added being present or removed so that one of the builds I do answers the undefined question you're asking; that's more building than I'm willing to do.