Over time I have created GNAT compiler for a variety for of platforms and one thing I noticed is the constant encounter of Hen and Egg situation. — At least when you leave the save heaven of Linux. Below some examples:
Texinfo and libiconv
A classic one: The compiler needs Texinfo to create it's online help and you need an working compiler to create Texinfo. Ahh, and Texinfo needs libiconv and libiconv needs a working C++ compiler.
But these problems is easily solved since online help is only optional extra.
Similar to Texinfo only a little trickier as a working assembler and linker are an absolute must. And don't expect every platform to come with a sufficiently up-to-date assembler or linker — Solaris springs to my mind here.
C++ is an integral part of GCC compiler and needed by libiconv. But that is not the problem. The problem is that the C++ build expect the C header file to exist at prefix
/include. Of course they only exist there after
For us Ada users the worse one: Ada is self hosted. Being self hosted is of course very cool indeed: About 95% of the Ada compiler and runtime are written in Ada. I do pity the poor GNU-Fortran maintainer where most of the compiler and runtime is written on C — Yuck.
Of course that coolness comes for a price: You need a working Ada compiler to build the Ada compiler.
Why am I telling you all that? Now, if you are planning to build a newer compiler / tool chain for any platform apart from an up to date Linux you should – from the very onset – expect to build the compiler more then once. In fact: plan for 3 to 4 successful iterations (and an uncounted amount of unsuccessful iterations). Start off with just C and Ada, add other languages later. Don't worry if for example Texinfo is reported to old, Create a newer version later and try again. And remember that libiconv needs an up to date C++ compiler.