Re: [MiNT] C++ Stuff

[Mark Duckworth <mduckworth@atari-source.org>]

Miro Kropacek wrote:
>     Everything is fine except when I use vectors so far.  It seems if
>     I use vector<T>.push_back() there is a call to an FPU function.
>      In gdb this results in an EMTTRAP emulation trap and the 
>
> Hmm and that is T? No matter, integer or float number, this always happen?

T is a buddy class.  The vector is defined in the buddylist class.  The 
failure happens when constructing a buddylist.

class buddy

{

   public:

       string screenname;
       unsigned short int selected;
       unsigned short int highlighted;
       unsigned short int online;
       unsigned short int away;

       struct aim_userinfo_s *userinfo;

       string prof;
       string info;
       int info_display_offset;
      
       void is_now_online();
       void is_now_offline();
       void is_now_away();
       void is_now_back();
};

class buddylist
{
   public:
       vector<buddy> thelist;       
       buddy* find_by_screenname(string in_sn);
       void add(string in_sn);
       buddylist(bool fromdisk);
       void remove(string in_sn);
       void print();
       void save_to_disk();
       string oscar_string();
       class AnError { };
};


The buddylist constructor opens a local file and reads it line by line 
to add buddies to the list.  This is a snippet:

               buddy mybuddy;               
               mybuddy.online=0;
               mybuddy.selected=0;
               mybuddy.away=0;
               mybuddy.highlighted=0;
               mybuddy.screenname=string(line).substr(1, strlen(line)-2);
               mybuddy.userinfo=NULL;
               mybuddy.prof="";
               thelist.push_back(mybuddy);

It dies on the push_back call.

>      It works when I compile with -m68020-60.
>
> This would explain if your T is some float number -- i.e. it would 
> mean there's a bug in floating-point emulation for 68000.
>
>     Does anyone else think it's strange that something as simple as
>     vector<T> would need FPU functions?
>
> since push_back is nothing more than adding variable to some (mostly 
> prealloced) space, why not. Imagine you keep it in FP register and you 
> want to add it to this array -- something like fmove.s fp0,(a3)+ is 
> quite normal. Did you try to disassemble it?
I did not.  But I will. 

> --
> MiKRO / Mystic Bytes
> http://mikro.atari.org