Sorry, Russell, for feeding the flames, but IMO this is the most common urban myth of embedded programmers.

I speak from experience - having ported successfully one quite big project twice, first changing the processor completely, second changing the major peripheral component, and I can reassure you that most of the work went into the "hardware dependent" part of it, which would have to be done in C or whatever language, too. Of course there was some work in the "rough translation", but that was mostly fun and went quickly and smoothly (I considered writing some sort of "compiler" but I estimated the gain in speed to be less than the "expense" to write it).

My view in this topic is, that

• there is no point to port smaller projects, it is faster to rewrite it from scratch
  
• the hardware dependent parts would need rewriting in C anyway; and the rest is mostly "mechanical translation" anyway
  
• the programming style makes perhaps the biggest impact on porting "speed" (but nasty asm code is a bad thing generally)
  
• porting from a feature-rich architecture (e.g. a 32-bitter) to a more restricted environment (e.g. a 8-bitter) would be really hard, but this is most probably language-independent

With the rest of your post, I completely agree.

Jan Waclawek