tcc_simple

UPDATE: see experiments/m2min_v3 for a working bootstrap path from a minimal
C/JS compiler all the way up to a bootstrappable version of tcc.

Original README (TODO update).


This repo is essentially the hypothetical v2 version of:

https://github.com/cosinusoidally/tcc_bootstrap_alt

To build tcc_simple, in general you also need tcc_bootstrap_alt checked out in
the parent directory eg:

$ ls
tcc_bootstrap_alt
tcc_simple
$ cd tcc_simple
$ ./mk_coverage_simple_c

The above should in theory work on many Linux systems, but you will need to
have the 32 bit libc installed, plus a copy of gcc that supports -m32. Works
fine on a i386 Ubuntu Bionic 18.04 chroot (except the gcovr step) or an Ubuntu
Jammy 22.04 64 bit system with 32 bit libs installed.

Note some of the mk scripts depend on the exitence of a prebuilt copy of a
stock tcc-0.9.27 (specifically the release tarball from 2017). The coverage mk
scripts depend on gcovr (they also try to start firefox from a specific path).
You might need to tweak things slightly.

The code in this repo is not yet complete, but the general plan is as follows:

Strip down The Tiny C Compiler (tcc-0.9.27) into 2 versions:

tcc_27_refactor - a cut down version of tcc-0.9.27 that is complete enough to
                  compile tcc-0.9.27
tcc_simple_c    - a cut down version of tcc_27_refactor that is only capable of
                  compiling a cut down version of the C language. Specifically
                  the language supported by M2_simple_asm.c (also included in
                  this repo).

Once tcc_simple_c and tcc_27_refactor are complete the next step is to create
2 bootstrap compilers:

tcc_boot_min - tcc_simple_c ported to the M2_simple_asm.c dialect. Will also be
               self hosted.
tcc_boot_max - tcc_27_refactor ported to the M2_simple_asm.c dialect. Will also
               be self hosted.

The bootstrap process would then be as follows:

hex0          - A "compiler" for commented hex dumps. See tcc_bootstrap_alt,
                the base bootstrap tools are from stage0-posix
hex1          - A "compiler" for commented hex dumps with support for single
                character labels.
hex2          - similar to hex1 but has support for larger labels, plus
                relative and absolute addressing modes.
M0            - A simple macro language used to implement a simple assembler
cc_x86        - a C compiler written in assembly (the version from
                tcc_bootstrap_alt). cc_x86_min.M1 or M2_simple_asm.M1 could
                also be used.
M2_simple_asm - a C compiler written in the "simple_c" dialect.
tcc_boot_min  - cut down version of tcc-0.9.27 implemented in the simple_c
                dialect. Implements the simple_c dialect.
tcc_boot_max  - cut down version of tcc-0.9.27 implemented in the simple_c
                dialect. Implements a large enough subset of C in order to
                to compile tcc-0.9.27.
tcc_27        - a nearly stock tcc-0.9.27 from tcc_bootstrap_alt . This version
                is capable of compiling a stock tcc-0.9.27. More importantly
                this version of tcc can also compile the GNU C compiler.
                Further, this version can even be plugged into live-bootstrap
                to replace the mescc steps.

For reference, fork of live-bootstrap that uses tcc_bootstrap_alt instead of mescc:
https://github.com/cosinusoidally/live-bootstrap/tree/tcc_bootstrap_alt-refactor_nov24

FAQs

Why not go from cc_x86 straight to tcc_boot_max?

It will be a bigger job to port tcc_boot_max to the simple_c dialect. By
starting with tcc_boot_min I can prove out the process. In theory once
tcc_boot_max is done it can then replace tcc_boot_min, but if there there are
any issues then tcc_boot_min would serve as a useful intermediate stepping
stone.

Even if I fail to finish tcc_boot_max I should still be able to use
tcc_boot_min to simplify tcc_bootstrap_alt. I could even potentially use the
unfinished version of tcc_boot_max to eliminate some steps in
tcc_bootstrap_alt.