Thread: Adminpack build fix

Attached patch removes a couple of extern definitions from adminpack,
replacing some of them with a #include. (Cam eup with this because we
got a duplicate definition of DataDir when building with Visual C++).
Seems much more correct to pull in the header file that has the
definition, and only do that.

//Magnus

"Magnus Hagander" <mha@sollentuna.net> writes:
> Attached patch removes a couple of extern definitions from adminpack,
> replacing some of them with a #include. (Cam eup with this because we
> got a duplicate definition of DataDir when building with Visual C++).

That isn't going to work unless we put DLLIMPORT into the main headers.
The reason for redeclaring these in the contrib files is to get
DLLIMPORT onto them...

            regards, tom lane

> > Attached patch removes a couple of extern definitions from
> adminpack,
> > replacing some of them with a #include. (Cam eup with this
> because we
> > got a duplicate definition of DataDir when building with
> Visual C++).
>
> That isn't going to work unless we put DLLIMPORT into the
> main headers.
> The reason for redeclaring these in the contrib files is to
> get DLLIMPORT onto them...

Interedting - it builds on MSVC without it :-O

Anyway. That certainly explains why MSVC is complaining - it's getting
completely different definitions of these variables from the header and
from the actual file.

What do you think about simply adding DLLIMPORT to the variables in the
main header?

//Magnus

"Magnus Hagander" <mha@sollentuna.net> writes:
>> The reason for redeclaring these in the contrib files is to
>> get DLLIMPORT onto them...

> Interedting - it builds on MSVC without it :-O

> Anyway. That certainly explains why MSVC is complaining - it's getting
> completely different definitions of these variables from the header and
> from the actual file.

The same redeclaration technique is being used elsewhere (pg_buffercache
and pg_freespacemap it looks like).  Aren't you getting warnings there
too?

            regards, tom lane

> >> The reason for redeclaring these in the contrib files is to get
> >> DLLIMPORT onto them...
>
> > Interedting - it builds on MSVC without it :-O
>
> > Anyway. That certainly explains why MSVC is complaining -
> it's getting
> > completely different definitions of these variables from the header
> > and from the actual file.
>
> The same redeclaration technique is being used elsewhere
> (pg_buffercache and pg_freespacemap it looks like).  Aren't
> you getting warnings there too?

I am - I just started working on getting those done as well. (they were
also living on my list of fix-later ones for some other reasons - the
other reasons now having been fixed)

//Magnus

"Magnus Hagander" <mha@sollentuna.net> writes:
>> The same redeclaration technique is being used elsewhere
>> (pg_buffercache and pg_freespacemap it looks like).  Aren't
>> you getting warnings there too?

> I am - I just started working on getting those done as well.

OK, I guess we gotta play them all honestly.  Will fix.

            regards, tom lane

"Magnus Hagander" <mha@sollentuna.net> writes:
> Attached patch removes a couple of extern definitions from adminpack,
> replacing some of them with a #include.

I've now removed all the local DLLIMPORT-redeclarations I could find
in favor of marking the relevant variables in the main header files.

            regards, tom lane

> > Attached patch removes a couple of extern definitions from
> adminpack,
> > replacing some of them with a #include.
>
> I've now removed all the local DLLIMPORT-redeclarations I
> could find in favor of marking the relevant variables in the
> main header files.

Thanks. All affected projects build with MSVC now - haven't tested if
they work, but they pass the compiler...

//Magnus