cintre: A C repository from naclsn

Running a REPL against a C lib

Situation: you have a C lib (header/s + obj/ar/shared)

Idea: automatically make a program that is a REPL linked against this lib, you enter C code, can interact with everything exposed in the header (declare variables with lib types, call lib functions, ...)

`example/`

Example/PoC with example/mylib:

$ cd example
$ make -f mylib.makefile && ./mylib
Type `;help` for a list of command
(*^^),u~~ sayhi; ty
Expression is of type: fun(times: int) -> void
(*^^),u~~ sayhi(2);
hi
hi
Result:
_: void
   = ()
(*^^),u~~

If you really don't want to use make (why?):

$ make -Bnf mylib.makefile
cc -O2   -c -o mylib.o mylib.c
cc ../cintre/preparer.c -o preparer.exe -O2
./preparer.exe ../cintre/standard.h -Pno-emit-decl -Pno-emit-incl -o a-standard.h
./preparer.exe mylib.h -o a-mylib.h -O2
./preparer.exe -m ./a-standard.h ./a-mylib.h -o c-mylib.c
cc -c c-mylib.c -o c-mylib.o -O2 -I. -I../cintre
cc -c ../cintre/cintre.c -o cintre.o -O2 -DUSE_READLINE -D_DEFAULT_SOURCE -D_XOPEN_SOURCE=600
cc mylib.o c-mylib.o cintre.o -o mylib -O2  -lc -lm -lreadline

Bluntly the steps are:

build mylib.o
build pr (one time)
pr standard.h make an interface (or "adapter") to stdlib/stdio/string... (one time)
pr mylib.h make an adapter with mylib.o
pr -m merge and list namespaces in a way cintre.o can use
build c-mylib.c
build cintre.o, the main() of the REPL (here with readline) (one time)
link mylib.o c-mylib.o and cintre.o

The driver.Makefile does exactly that, see example/Makefile.

Bugs

Assumes platform:

has 8 bits bytes (ie. CHAR_BIT == 8);
is little endian (ie. (char[4]){0xd6, 0xff, 0x00, 0x00} == (int)0xff2d);
respects sizeof(size_t) == sizeof(void*) (which is "not guarenteed", but like..).

It also collapses long and long long into the largest (ie. at runtime sizeof(long) == sizeof(long long) and usually == 2*sizeof(int)).

(wip and such)

breaking

fix precedence in a = 1 ? 2 : 3
pr: bitfields here and there
pr: array size/align

progress

pr: macros, like Err codes
more conversions in _fit_expr_to_slot
handle bitfields in runtime types

opti/cleanup

maybe reverse the sp and vsp to have sane zero init (// xxx: sizeof stack)
limit avoidable capture-copying
go through tokn(...) to reduce expansion in favor of locals
more coverage
pr: "ptr tail" in emit_forward => is a hack on fun/arr declarator, so do it proper
pr: is getting weird again (eg. emit_forward for ptr to x, emit_adpt_type_val for param[n], size/align situation for arrays, ..)
n -> logn ns search

the GitHub "Need inspiration?" bit was "animated-spoon"

naclsn/cintre