Re: Widl Fixes

[Ove Kaaven]

fre, 2003-03-07 kl. 19:49 skrev Jon:
> This patch addresses some issues with widl. After applying the patch all .idl 
> files should be regenerated. These fixes are needed before my ocidl.idl can 
> be added. The last two lines of the changelog aren't seen in the wild; I 
> found them only when adding ocidl to my tree.

Could you explain the reason behind the last line, "Don't dump full
struct/enum definitions in prototypes", since this shouldn't happen in
any case. When a prototype is emitted, the parameter type is supposed to
be defined, or not; if it's not defined, t->defined is FALSE, and if
it's defined, t->written should be TRUE, either of which should lead to
the "else" clause which doesn't dump the full struct/enum definition.
If t->written isn't TRUE, why isn't it? (And if there's a good reason
for it after all, it would be cleaner to say "if (t->defined &&
!t->written && structs)" instead of adding a new "if (structs)" inside)
> There is one remaining bug in widl which I haven't yet tracked down; someone 
> more knowledgable with widl may want to look at it:
>
> If objidl.idl is compiled with '-b' but without '-h', the resulting header 
> contains garbage ("[][][][][][][][][]" - generated by write_array() just 
> after IPropertyStorage_Stat_Stub) in the generated header. With both flags 
> (the default for Wine) however, the output is correct.

I thought I had fixed such occurrences. If your widl is properly
recompiled, then I'm not sure why it happens.
> Also, it would be nice if widl handled function pointers as method args...

I didn't think RPC supported this. The MS IDL I've seen just uses that
cpp_quote #if 0 ... else trick to define structures with these, and I
saw no need for widl to be smarter about it than MIDL is, considering
that function pointers doesn't map cleanly to RPC anyway. Besides, it
would be nice if our .idl files stayed compatible with MIDL.

Widl Fixes

[Jon Griffiths]

Hi,

--- Ove Kaaven  wrote:
> Could you explain the reason behind the last line, "Don't dump
> full struct/enum definitions in prototypes", since this shouldn't
> happen in any case.

The following idl file, "foo.idl", cut down from my oleidl.idl for
example purposes:

---------------------------------
import "objidl.idl";

interface IOleClientSite;
interface IOleObject;

typedef LPCRECT LPCBORDERWIDTHS;

[
  object,
  uuid(00000112-0000-0000-C000-000000000046)
]
interface IOleObject : IUnknown
{
  typedef IOleObject* LPOLEOBJECT;

HRESULT DoVerb(
    [in] LONG iVerb,
    [in] struct tagMSG* lpmsg,
    [in] IOleClientSite* pActiveSite,
    [in] LONG lindex,
    [in] HWND hwndParent,
    [in] LPCRECT lprcPosRect);
}
---------------------------------

Produces as its output when run with -b -h:

struct IOleObject: IUnknown {
    virtual HRESULT CALLBACK DoVerb(
        LONG iVerb,
        struct tagMSG {
            HWND hwnd;
            UINT message;
            WPARAM wParam;
            LPARAM lParam;
            DWORD time;
            POINT pt;
        }* lpmsg,
        IOleClientSite* pActiveSite,
        LONG lindex,
        HWND hwndParent,
        LPCRECT lprcPosRect) = 0;

} ICOM_COM_INTERFACE_ATTRIBUTE;

The patch I sent fixes this problem; if you think theres a better way
to fix it then let me know, I'll strip that change out and re-send
just the other fixes.
> I didn't think RPC supported this.

Fair enough, probably best left as it is then.

Cheers,
Jon

=====
"Don't wait for the seas to part, or messiahs to come;
 Don't you sit around and waste this chance..." - Live

[e-mail]

__________________________________________________
Do you Yahoo!?
Yahoo! Web Hosting - establish your business online
http://webhosting.yahoo.com

Widl Fixes

[Ove Kaaven]

lør, 2003-03-15 kl. 13:35 skrev Jon Griffiths:
> Hi,
>
> --- Ove Kaaven  wrote:
> > Could you explain the reason behind the last line, "Don't dump
> > full struct/enum definitions in prototypes", since this shouldn't
> > happen in any case.
>
> The following idl file, "foo.idl", cut down from my oleidl.idl for
> example purposes:
>
> import "objidl.idl";
>
> interface IOleClientSite;
> interface IOleObject;
>
> typedef LPCRECT LPCBORDERWIDTHS;
>
> [
>   object,
>   uuid(00000112-0000-0000-C000-000000000046)
> ]
> interface IOleObject : IUnknown
> {
>   typedef IOleObject* LPOLEOBJECT;
>
> HRESULT DoVerb(
>     [in] LONG iVerb,
>     [in] struct tagMSG* lpmsg,
>     [in] IOleClientSite* pActiveSite,
>     [in] LONG lindex,
>     [in] HWND hwndParent,
>     [in] LPCRECT lprcPosRect);
> }
>
> Produces as its output when run with -b -h:
>
> struct IOleObject: IUnknown {
>     virtual HRESULT CALLBACK DoVerb(
>         LONG iVerb,
>         struct tagMSG {
>             HWND hwnd;
>             UINT message;
>             WPARAM wParam;
>             LPARAM lParam;
>             DWORD time;
>             POINT pt;
>         }* lpmsg,
>         IOleClientSite* pActiveSite,
>         LONG lindex,
>         HWND hwndParent,
>         LPCRECT lprcPosRect) = 0;
>
> } ICOM_COM_INTERFACE_ATTRIBUTE;

Oh I see now... struct tagMSG is defined in an imported idl file so it
won't be written out while processing ocidl, and thus t->written remains
FALSE even though it's properly defined. The best way to solve this
would still be to ensure that t->written is TRUE, by changing the
parser.y line

if (!parse_only) write_typedef($$, $4);

into something that will still set the "written" field of the type (and
its embedded structures) even if parse_only is true (and similarly with
the enumdefs, structdefs, uniondefs, and interfacedefs, probably). A new
parameter to write_type/write_typedef which inhibits writing anything to
the header but still sets "written", for example... would that work?

Widl Fixes

[Jon Griffiths]

Hi,
> > --- Ove Kaaven  wrote:
>
> A new parameter to write_type/write_typedef which inhibits writing
> anything to the header but still sets "written", for example...
> would that work?

Err, probably, but I'm not convinced its a better solution without
going back through the code again (I made those changes over a month
ago). I'll resubmit the other two fixes and look at it again when my
current batch of oleaut32 things are wrapped up. I only came accross
this bug trying to add IRecordInfo for the SafeArray/Variant
functions I wanted to implement, so I've wandered quite far from my
original path already...

Cheers,
Jon

=====
"Don't wait for the seas to part, or messiahs to come;
 Don't you sit around and waste this chance..." - Live

[e-mail]

__________________________________________________
Do you Yahoo!?
Yahoo! Platinum - Watch CBS' NCAA March Madness, live on your desktop!
http://platinum.yahoo.com

Widl Fixes

[Ove Kaaven]

tir, 2003-03-18 kl. 08:39 skrev Jon Griffiths:
> Hi,
> > > --- Ove Kaaven  wrote:
> >
> > A new parameter to write_type/write_typedef which inhibits writing
> > anything to the header but still sets "written", for example...
> > would that work?
>
> Err, probably, but I'm not convinced its a better solution without
> going back through the code again (I made those changes over a month
> ago).

I think it is. Consider:

file1.idl:
typedef struct tagSTRCT { int a; } STRCT;

file2.idl:
import "file1.idl"
typedef struct tagSTRCT *LPSTRCT;

Without my solution, we'd get

file2.h:
#include "file1.h"
typedef struct tagSTRCT { int a; } *LPSTRCT;

which would be a struct redefinition.
> I'll resubmit the other two fixes and look at it again when my
> current batch of oleaut32 things are wrapped up. I only came accross
> this bug trying to add IRecordInfo for the SafeArray/Variant
> functions I wanted to implement, so I've wandered quite far from my
> original path already...

OK.