compiling d code without phobos ?

[Frits van Bommel]

Wiz wrote:
> I'm currently working on a little OS, and my goal is to use the D 
> language as the main one.

I'm working on a similar project, on and off.

> I had no problem to build a running very very primitive version, but 
> when I tried to define (not even using it), I have seen some troubles : 
> gdc tells me there is a few reference that are undefined.
> Right, theses references are the phobos ones : I have no need of since 
> phobos can't work on my system yet (memory allocs and lot of things are 
> going to be rebuild by myself later).
> I then tried to make my own object.d file, I put it into my kernel 
> folder and added it to the makefile, I still have some problems...

 From what I remember, object.d doesn't need much work to be included in 
a kernel. It doesn't allocate any memory anywhere that I can tell.
In fact, I just did a quick diff and the main differences between my 
version and the one with DMD were substituting some C functions for 
self-written ones, as well as a local version of some std.string 
functions. The only printf() uses write strings without further 
formatting, so you don't even need to implement it using varargs yet.

The best way to work is probably to remove all imports, 
rewrite/copy/remove use of the functions the compiler complains about, 
then try to link.
For every missing symbol the linker complains about, try to find it in 
the phobos sources (I use grep) and understand what it does. Then either 
reimplement it, copy it to your source tree, or edit object.d so it 
isn't used anymore.
For instance, you can either implement _d_assert (called when an 
assertion without message fails) or comment out all such assertions.

Also, don't forget to use phobos/internal/object.d as a basis instead of 
phobos/object.d.

> When compiling my version of object.d containing just a class definition 
> 'object', gdc seg faults. It seems to me that gdc imports 
> mingw/include/d/3.4.5/object.d before compiling mine, and don't know 
> what to do in the case of these two.
> I also before got a error telling me the class was already defined (but 
> don't remember how, from what I know there is no notable difference 
> between the yesterdays context, and todays one), that's wy I ask if 
> anybody knows any way of not using the phobos object.d, just my own ?
 > Tried the -nostdlib of gcc, without any positive results.

Make sure you -I the directory containing your version of object.d.
Also, I invoke ld directly to link to make sure the standard library 
isn't linked, just my own code.

> ps: Using windows version of gdc (GCC) 3.4.5 (mingw special) (gdc 0.19, 
> using dmd 0.162)

I use DMD for Linux, so there may of course be some differences between 
our setups.


A general tip:
Assertion support is one of the first things you want to implement after 
basic text output is implemented. It's invaluable.
Make sure it relies on as little other code as possible though. If your 
output code has any chance of asserting, implement a simplified copy 
without assertions specifically for use by error reporting code to 
prevent infinite loops (yes, I've been bitten by this :( ).
My code at first just copied a text message to the screen and halted the 
system.
Later, I extended it to halt only the current thread, provide register 
dumps and stack traces (with demangled function names), as well as 
sending all output to a serial port for debugging convenience (it's 
easy). But first you need to get the basics working, the rest can come 
later.