Windows doesn't reliably offer *asprintf() (sprintf functions that
automatically malloc as much memory as they need) so hand-roll the
equivalent on all platforms.
These functions are complicated because they need to be marked
"noreturn" for the compiler. There are several fundamentally
incompatible ways of doing that. Since we have only a few "noreturn"
functions, let's just list all of those in one header file of which
there is a completely different version for each compiler.
I planned to put just the file system routines into these files, but
there are starting to be so many of them that I don't want to
duplicate the same set for networking, subprocesses etc.
Stage 0 is supposed to emit compiled versions of system.scm and
compiler.scm. The compiler.scm bytecode wasn't emitted properly
because the emitter put the two files' bytecode in two separate C
variables instead of concatenating them into the same variable as they
should be.
Previously the boot image was kept in the separate file "flisp.boot".
This caused all kinds of problems, as bootstrapping is wont to do. The
biggest problem was finding a portable and convenient pathname for it.
Instead of dealing with all that stuff about how to find the file,
just store the image in a huge C string constant that goes into the
main executable. This necessitates adding a C-string hex dumper into
the Lisp system, and making build.sh slightly more complicated, but
that's a small price to pay for not having to carry a separate file
everywhere we go. Also, flmain.c is a lot simpler now and we don't
have to play around with symlinks.
Apparently in addition to "float", "double" and "long double" there
are now standard types in <math.h> called "float_t" and "double_t".
Those types don't need to be equivalent to "float" and "double". Gah!
MINIX libc headers define the ctype.h toupper() etc. functions as
preprocessor macros that don't cast their argument to int! So a char
argument causes clang to say "warning: array subscript is of type
'char'".
Lassi Kortela