Various problems.

[rdebath]

I've just generated this piece of C code from a torture test of BF program interpreters. Obviously, being generated code it promises to be a bit of problem. I suspect this promise will be fulfilled.

https://gist.github.com/rdebath/ecd1454a259ece0a8ecf

The CPP define "C" can be declared, on the command line or inside, to be any integer type, signed or unsigned, for 8 bits to 128bits or more. The program, when run, should display one of a few variants of 'hello world' and then finish.

The default of integers complies and runs, but incorrectly. Any type smaller then an integer gets a 'not an lvalue' error from a "-- * X" type construct. Longs compile, but seem not to work at all, however, this appears to be a problem with 'stdint.h'.

Test run on Debian Wheezy.

[andrewchambers]

Nothing in the code looks too crazy, though large parts of the code are being rewritten atm so the issue might need to wait until after the rewrite.

I'll make a smaller test case out of your code later though, it might be useful to fill holes in the 8cc test suite. I'll add it onto a branch on my own fork so I don't forget about it.

[rdebath]

Heh, if you want crazy I can do that too ... https://gist.github.com/rdebath/2120d506bda710a76dd6

This one seems to work correctly, however, it doesn't get any constant folding done on it so the binary is a bit large (but the memory it needs to compile isn't too bad).

TODO: Constant folding of expressions like:

    x[4] += 1+1+1+1+1+1+1+1+1;
    si.flg = 4096+256+64+2+1;

[rui314]

This compiler has no serious test sets except the ones in test directory and the compiler itself. When I write new code for the compiler I sometimes found a miscompilation. So I wouldn't surprise that the code couldn't be compiled correctly with 8cc. I'll take a look later.

[andrewchambers]

@rdebath I don't think we want crazy :). Sane code with errors is more useful. Also, 8cc doesn't do constant folding, it may need to in the future to fix issue #5 though.