Assembly's too high-level :) In my school's sophomore year, CS majors are required to implement a CPU on an FPGA; the instruction set, CPU architecture, everything are up to the student groups to design and implement. At the end of the class, all the student processors are compared w.r.t. size and speed of execution of various simple programs. Our group had a guy that wrote an assembler for our ISA, but most of the groups just wrote their programs in raw machine code, using a hex editor.
Really, nothing teaches low-level programming better than starting with gates and building your way up to programs. The sequel to that class teaches how to deal with pipelining, CPU caches, and multiple-execution chips like the TI DSPs, so after you design your processor, you get to see how real CPUs work. It's really fun, actually.
The intro computer architecture course used _Computer Organization and Design_ by Patterson and Hennessy. The practical component described here was driven by handouts, I'm not sure the text follows building something like this.
Perhaps C offers a sweet spot of being able to accomplish something with minimal effort, but still gaining a deep understanding of what the computer is actually doing.
Those are actually good ideas, and if you received a computer science degree from Carnegie Mellon University, you may have witnessed first hand their actualization in CS 213, Introduction to Computer Systems:
1. Int’s are not integers, Float’s are not
reals. Our finite representations of numbers
have significant limitations, and because of
these limitations we sometimes have to think in
terms of bit-level representations.
2. You’ve got to know assembly language.
Even if you never write programs in assembly,
The behavior of a program cannot be understood
sometimes purely based on the abstraction of a
high-level language. Further, understanding the
effects of bugs requires familiarity with the
machine-level model.
3. Memory matters. Computer memory is not
unbounded. It must be allocated and managed.
Memory referencing errors are especially
pernicious. An erroneous updating of one object
can cause a change in some logically unrelated
object. Also, the combination of caching and
virtual memory provides the functionality of a
uniform unbounded address space, but not the
performance.
4. There is more to performance than
asymptotic complexity. Constant factors also
matter. There are systematic ways to evaluate
and improve program performance
5. Computers do more than execute
instructions. They also need to get data in and
out and they interact with other systems over
networks.
That's a 200 level class, as far as I know there is always a CS1 100 level class that usually starts with some EXTREMELY high level language, sometimes a metalanguage just for the class itself? I heard somewhere Python is in style for this. Here is one such course : http://www.academicearth.org/courses/introduction-to-compute...
*EDIT - I made a stupid comment... they have C assignments in that class. I'll leave it for others amusement. I have seen that style of starting with a high level language though...
