Daniel Green Wrote:

> On 3/24/2011 6:53 PM, Kagamin wrote:
> > Is the AIX problem valid for windows too?
> What is the AIX problem?

https://bitbucket.org/goshawk/gdc/wiki/UserDocumentation#!known-issues

writeln("hello world")
compiles to 3MB exe which is approx 1/3 of libgphobos2.
object file is just 3KB.

or may be it's just the datetime module...

"Kagamin" <spam@here.lot> wrote in message news:imha5f$2tke$1@digitalmars.com...
> Daniel Green Wrote:
>
>> On 3/24/2011 6:53 PM, Kagamin wrote:
>> > Is the AIX problem valid for windows too?
>> What is the AIX problem?
>
> https://bitbucket.org/goshawk/gdc/wiki/UserDocumentation#!known-issues
>
> writeln("hello world")
> compiles to 3MB exe which is approx 1/3 of libgphobos2.
> object file is just 3KB.
>
> or may be it's just the datetime module...

That poor datetime module just can't catch a break! :)

On 3/25/11, Kagamin <spam@here.lot> wrote:
> Daniel Green Wrote:
>
>> On 3/24/2011 6:53 PM, Kagamin wrote:
>> > Is the AIX problem valid for windows too?
>> What is the AIX problem?
>
> https://bitbucket.org/goshawk/gdc/wiki/UserDocumentation#!known-issues
>
> writeln("hello world")
> compiles to 3MB exe which is approx 1/3 of libgphobos2.
> object file is just 3KB.
>

The real issue here is that the GDC binaries that were just released weren't stripped. I guess Daniel forgot about that. :)

There's strip.exe that you can use on Windows. It usually comes with either TDM-GCC or MinGW binaries. Once you have it, run this batch file from GDC's main directory:

for /F %%G in ('dir /B /S *.exe;*.a;*.so') do strip --strip-unneeded %%G

Now libgphobos2 will be 3.5 megs in size, and a test project with a
writeln() will compile to 966Kb.

Andrej Mitrovic Wrote:

> for /F %%G in ('dir /B /S *.exe;*.a;*.so') do strip --strip-unneeded %%G
> 
> Now libgphobos2 will be 3.5 megs in size, and a test project with a
> writeln() will compile to 966Kb.

Forgot about that, thanks.

On 3/25/11, Kagamin<spam@here.lot>  wrote:
> Daniel Green Wrote:
>
>> On 3/24/2011 6:53 PM, Kagamin wrote:
>>> Is the AIX problem valid for windows too?
>> What is the AIX problem?
>
> https://bitbucket.org/goshawk/gdc/wiki/UserDocumentation#!known-issues
>
> writeln("hello world")
> compiles to 3MB exe which is approx 1/3 of libgphobos2.
> object file is just 3KB.
It doesn't.  The following sizes I think use shared libgcc.  Might be a couple hundred more for a static build.  A lot of the size is overhead.

856 hello.o
682,987 hello.exe
112,128 hello.exe after strip hello.exe
39,424 hello.exe upx...

On 3/25/2011 11:35 AM, Andrej Mitrovic wrote:
> The real issue here is that the GDC binaries that were just released
> weren't stripped. I guess Daniel forgot about that. :)
That was on purpose.  Unless it's an actual release, I won't strip them.

Also, stripping the GDC binaries won't have any effect on output file size.  Stripping the phobos libraries will, but I believe you'll also permanently lose any debugging information in the process.  A better solution is to strip the outputted binary when your done with it.

Does GCC have something equivalent to Microsoft's PDB file?  Essentially it's all the debug information stored in a separate file.

Apparently there is a way to strip the debuggin info and put it in another file, and then link the executable to the debug file so GDB can automatically load it. But this doesn't seem to work for me:

gdc -g foo.d -o foo.exe
objcopy --only-keep-debug foo.exe foo.debug
strip -g foo.exe
objcopy --add-gnu-debuglink=foo.debug foo.exe

It doesn't add the .gnu_debuglink section to foo.exe, I don't know why..

> > writeln("hello world")
> > compiles to 3MB exe which is approx 1/3 of libgphobos2.
> > object file is just 3KB.
> It doesn't.  The following sizes I think use shared libgcc.  Might be a couple hundred more for a static build.  A lot of the size is overhead.
> 
> 856 hello.o
> 682,987 hello.exe
> 112,128 hello.exe after strip hello.exe
> 39,424 hello.exe upx...

----
gdc -v2 test4.d -o test4.exe
----
dir test4*
----
26.03.2011  19:17                78 test4.d
26.03.2011  19:18         2 835 758 test4.exe
               2 File(s)      2 835 836 bytes
----
copy test4.d con
----
import std.stdio;
int main()
{
        writeln("hello world");
        return 0;
}
        1 file(s) copied.
----

with
gdc -v2 test4.d -o test4.exe -s
it's 454kb

On 3/26/2011 12:27 PM, Kagamin wrote:
> ----
> gdc -v2 test4.d -o test4.exe
> ----
> dir test4*
> ----
> 26.03.2011  19:17                78 test4.d
> 26.03.2011  19:18         2Â 835Â 758 test4.exe
>                 2 File(s)      2Â 835Â 836 bytes
Apparently, I used core.std.stdio.  When I retested using std.stdio I got similar numbers.

About 2 megs can be removed by stripping libgphobos2.a.  I'll start doing that now that I know it's possible to separate debug information.

nm test4.exe | grep ModuleInfo

Run for a list of included modules.  The best way to tell would be compare outputs with linux.

> ----
> copy test4.d con
> ----
> import std.stdio;
> int main()
> {
>          writeln("hello world");
>          return 0;
> }
>          1 file(s) copied.
> ----
>
> with
> gdc -v2 test4.d -o test4.exe -s
> it's 454kb
From the GCC manual.
-s
    Remove all symbol table and relocation information from the executable.

It also looks like the AIX issue was related to the linker for the platform and applied to c++ as well.

Daniel Green Wrote:

> nm test4.exe | grep ModuleInfo

lol, winsock, regex, regexp, random(?)
Well, std.datetime and std.complex can be understandable: they can be used by writeln...

> > with
> > gdc -v2 test4.d -o test4.exe -s
> > it's 454kb
>  From the GCC manual.
> -s
>      Remove all symbol table and relocation information from the
> executable.

yes, -s means strip

Daniel Green Wrote:

> nm test4.exe | grep ModuleInfo

as to std.random, in std.algorithm we have

version(unittest)
{
    import std.random, std.stdio, std.string;
    mixin(dummyRanges);
}

is phobos compiled with unittests enabled?