Re: [ooc-compiler] H20 bugs

[Stewart G. <sgr...@ii...>]

Hi Frank,

> The glib library headers use an idiom that looks like this:
> 
>   typedef struct _GStaticMutex GStaticMutex;
>   struct _GStaticMutex
>   {
>     struct _GMutex *runtime_mutex;
>     union {
>       char   pad[24];
>       double dummy_double;
>       void  *dummy_pointer;
>       long   dummy_long;
>     } static_mutex;
>   };
> 
> As far as I can tell from the documentation, the leading underscore 
> means the _GStaticMutex type is meant to be "opaque", presumably meaning 
> its record fields should not be exported. However, this is simply a 
> matter of convention and is not the real problem.

I think its done to ensure that the names used in structs and typedefs 
don't collide. From memory, I think the C spec requires separate 
name-spaces, but there may be compilers that don't do this. There 
doesn't seem to be a consistent convention. In WIN32, structure tags are 
prefixed with "tag", rather than "_".

> The code produced by TestH2O looks like this:
> 
> TYPE
>   _GStaticMutex_tag* = RECORD
>     runtime_mutex* : POINTER TO _GMutex_tag;
>     static_mutex* : RECORD [UNION]
>       pad* : ARRAY 24 OF CHAR;
>       dummy_double* : LONGREAL;
>       dummy_pointer* : SYSTEM.PTR;
>       dummy_long* : LONGINT;
>     END;
>   END;
>   _GMutex_tag* = RECORD END;
>   _GStaticMutex_tag* = RECORD END;
>   GStaticMutex* = RECORD END;
> 
> Naturally this doesn't compile.

Yes, that looks like H2O has got something wrong. In your code, struct 
_GMutex is opaque but there is an error in the declaration of 
GStaticMutex. When GStaticMutex is declared, _GStaticMutex is opaque but 
that should not matter. It should have output just:
   GStaticMutex* = _GStaticMutex_tag;

I'll take a look at what's wrong. Thanks too for your patch, which I'll 
add to the CVS version.

Cheers,
   Stewart