Re: PATCH: cups dynamical dependency

[Dimitrie O. Paun]

On November 10, 2002 02:40 pm, Marcus Meissner wrote:
>    Do not link against -lcups directly, but dynamically load it
>    if present. (just like freetype etc.)

[...]
> +#ifdef HAVE_CUPS
> -
>       /* dynamically load CUPS if not yet loaded */
> -
>       if (!cupshandle) {
> -
>           cupshandle = wine_dlopen(CUPS_SONAME, RTLD_NOW, NULL, 0);
> -
>           if (!cupshandle) cupshandle = (void*)-1;
> -
>       }
>
> +#endif

Well, if we do this dynamically, why have this HAVE_CUPS check which is a
compile time check? IMO we should just include a copy of the CUPS headers
that we need, and drop the compile time check altogether. In fact, this
check is misleading, as it suggests that we've verified some sort of
compatibility with CUPS which we haven't. We _assume_ that a certain API
is available at runtime, so why pretend we use something that's on the
machine we compile on?

PATCH: cups dynamical dependency

[Vincent Béron]

Le dim 10/11/2002 à 23:18, Dimitrie O. Paun a écrit :
> On November 10, 2002 02:40 pm, Marcus Meissner wrote:
> >    Do not link against -lcups directly, but dynamically load it
> >    if present. (just like freetype etc.)
>
> [...]
> > +#ifdef HAVE_CUPS
> > -
> >       /* dynamically load CUPS if not yet loaded */
> > -
> >       if (!cupshandle) {
> > -
> >           cupshandle = wine_dlopen(CUPS_SONAME, RTLD_NOW, NULL, 0);
> > -
> >           if (!cupshandle) cupshandle = (void*)-1;
> > -
> >       }
> >
> > +#endif
>
> Well, if we do this dynamically, why have this HAVE_CUPS check which is a
> compile time check? IMO we should just include a copy of the CUPS headers
> that we need, and drop the compile time check altogether. In fact, this
> check is misleading, as it suggests that we've verified some sort of
> compatibility with CUPS which we haven't. We _assume_ that a certain API
> is available at runtime, so why pretend we use something that's on the
> machine we compile on?

I don't think including foreign headers is the way to go. Better
pressure packagers to have CUPS-devel installed when they compile. For
self-compilation, if CUPS is not installed at compile-time, it won't be
there at runtime (except if installed since then, or if CUPS-devel is
not installed).

Foreign headers mean you must track new versions, etc. It doesn't seem
to be the right thing to do.

Vincent

PATCH: cups dynamical dependency

[Marcus Meissner]

On Sun, Nov 10, 2002 at 11:18:27PM -0500, Dimitrie O. Paun wrote:
> On November 10, 2002 02:40 pm, Marcus Meissner wrote:
> >    Do not link against -lcups directly, but dynamically load it
> >    if present. (just like freetype etc.)
>
> [...]
> > +#ifdef HAVE_CUPS
> > -
> >       /* dynamically load CUPS if not yet loaded */
> > -
> >       if (!cupshandle) {
> > -
> >           cupshandle = wine_dlopen(CUPS_SONAME, RTLD_NOW, NULL, 0);
> > -
> >           if (!cupshandle) cupshandle = (void*)-1;
> > -
> >       }
> >
> > +#endif
>
> Well, if we do this dynamically, why have this HAVE_CUPS check which is a
> compile time check? IMO we should just include a copy of the CUPS headers
> that we need, and drop the compile time check altogether.

Actually we would just need 4 lines of function prototypes with char*,
char** and int* arguments, no structures are passed.
> In fact, this
> check is misleading, as it suggests that we've verified some sort of
> compatibility with CUPS which we haven't.

But we did check against at least one set of function prototypes.
> We _assume_ that a certain API
> is available at runtime, so why pretend we use something that's on the
> machine we compile on?

It is the same for freetype.

I can do it the other way easily. I'll send a new patch later this day.

Ciao, Marcus

PATCH: cups dynamical dependency

[Sylvain Petreolle]

> It is the same for freetype.
>
> I can do it the other way easily. I'll send a new patch later this
> day.
>
> Ciao, Marcus

Couldn't this be done for all dlls thare loaded inconditionnally by
another dlls ? For example comdlg32 loads winspool.drv, even if you
don't want to print anything. 
I loaded for example the installation of ie6 and it wanted
winspool.drv, which has no sense at all.

___________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com

PATCH: cups dynamical dependency

[Marcus Meissner]

On Mon, Nov 11, 2002 at 02:26:31PM +0100, Sylvain Petreolle wrote:
> > It is the same for freetype.
> >
> > I can do it the other way easily. I'll send a new patch later this
> > day.
> >
> > Ciao, Marcus
>
> Couldn't this be done for all dlls thare loaded inconditionnally by
> another dlls ? For example comdlg32 loads winspool.drv, even if you
> don't want to print anything.

You could use this patch:

Index: dlls/commdlg/Makefile.in
===================================================================
RCS file: /home/wine/wine/dlls/commdlg/Makefile.in,v
retrieving revision 1.24
diff -u -u -r1.24 Makefile.in
--- dlls/commdlg/Makefile.in	18 Oct 2002 23:46:29 -0000	1.24
+++ dlls/commdlg/Makefile.in	11 Nov 2002 20:24:39 -0000
@@ -4,7 +4,8 @@
 SRCDIR    = @srcdir@
 VPATH     = @srcdir@
 MODULE    = comdlg32.dll
-IMPORTS   = shell32 shlwapi comctl32 winspool.drv user32 gdi32 kernel32
+IMPORTS   = shell32 shlwapi comctl32 user32 gdi32 kernel32
+DELAYIMPORTS = winspool.drv
 ALTNAMES  = commdlg.dll
 EXTRALIBS = $(LIBUUID)

Ciao, Marcus

PATCH: cups dynamical dependency

[Alexandre Julliard]

"Dimitrie O. Paun"  writes:
> Well, if we do this dynamically, why have this HAVE_CUPS check which is a
> compile time check? IMO we should just include a copy of the CUPS headers
> that we need, and drop the compile time check altogether. In fact, this
> check is misleading, as it suggests that we've verified some sort of
> compatibility with CUPS which we haven't. We _assume_ that a certain API
> is available at runtime, so why pretend we use something that's on the
> machine we compile on?

I think the check is just fine. If you don't have the headers you
can't build CUPS support in, no matter what will be present at
runtime. The assumption is that the library that you will (maybe) find
at runtime is compatible with the headers you have built against; this
is much better than shipping our own version of the headers and
assuming all platforms are using a compatible library.