The 8008 697

Ah, the compiler bootstrap problem.

Say, you have a compiler, 'A' you don't trust; but you have carefully crafted an alternative compiler. It has conditionals in the code to be a lot more conservative when not being compiled by itself.

So you compile C with A, and get C1 as a binary. This binary has all the code generated by A, and it generates very conservative code, because only a bootstrap backend has been generated. It also has workarounds for compiler bugs in A, further limiting what is generated. C1 does not generate optimized code.

Then you compile C again with C1, and get C2. Now, C2 is generated with a bootstrap backend, but it has the full functionality of the compiler; but operates inefficiently because the code in C2 is generated by a limited compiler, and is itself not optimized. But it generates optimized code.

So you compile C with C2, which is a full function compiler, but not with optimized code, and get C3 which is finally an optimizing compiler built with an optimizing compiler.

Then you compile C with C3 and get C4, and compare C3 and C4 just to be sure they are alike.

Now, there are compiler "features" that can affect this, and lead to further oscillations. In some instances you may not trust C1 enough for optimized code, so you get an extra step. Code generation from parse threes can also be affected, where you get functional, but different results depending on which compiler the compiler was built with.

I have heard stories of people going as far as C7 because of such effects.

-- mrr