Some test programs (one-liners):

> void main() {}
$ dmd main1.d
> void main(char[][] args) {}
$ dmd main2.d
> void main(char[] string) {}
$ dmd main3.d
main3.d:1: function main3.main parameters must be main() or main(char[][] args)

So far, so good. But then comes:

> int main() { return 0; }
$ dmd main4.d
> int main(char[][] args) { return 0; }
$ dmd main5.d
> char[] main() { return "EXIT_SUCCESS"; }
$ dmd main6.d

i.e. No errors (on Mac OS X), until you run it:

> $ ./main6; echo $?
> Bus error

similar on Linux, when using the GDC compiler:

> $ ./main6; echo $?
> Segmentation fault

but, if you compile it on Linux X86: (with DMD)

> $ ./main6; echo $?
> 12


I think that main should be *forced* to return int or void,
where "void" would mean that the program would return a 0...


This would shorten small programs,

from:
> int main()
> {
>     printf("Hello, World!\n");
>     return 0;
> }

to:
> import std.stdio;
> void main()
> {
>     writefln("Hello, World!");
> }

At least some constants from C's /usr/include/stdlib.h
could be used, to make D programs more self-documenting:

> import std.stdio;
> import std.c.stdlib;
> int main()
> {
>     writefln("Hello, World!");
>     return EXIT_SUCCESS;
> }


In summary, "void main()" should be either defined or invalid
- to avoid this: http://users.aber.ac.uk/auj/voidmain.shtml ?

Currently it has different results for DMD and GDC compilers....

--anders

PS.
"12" was of course (?) the length of the string "EXIT_SUCCESS"
See my previous post, currently DMD returns register EAX (!):
http://www.digitalmars.com/drn-bin/wwwnews?digitalmars.D/13127