On Wednesday, 1994-07-27  Stacy Smith wrote to Eric Erickson about
"Thunking, I Think..." as follows:

SS> EE| No thunking is the process of converting between 16:16 (selector:offset)
SS>   | and 32 bit flat pointers. This is required under OS/2 2.x since some
SS>   | function calls   want 16 Bit Pointers and others need 32 bit.
SS> 
SS> Ah...that makes sense.  Are there functions/macros in OS/2
SS> compilers to do this, or do I need to calculate it myself? ---

Hi Stacy,

Actually, thunking is a more general process that has been around for
over 30 years. It was first used in public by a number of ALGOL 60
compilers that implemented the full, 1960 standard for the IFIP
Algorithmic Language (ALGOL).

A thunk is a compiler-generated (as opposed to programmer-written)
subroutine that provides parameter values to a programmer-written
subroutine when the parameter supplied by the calling program or
subroutine does not match the specifications of the formal parameter in
the called subroutine's definition. It is theoretically a two-way
mechanism, unlike (say) FORTRAN 77 that would automatically pass a
converted value -- typically an INTEGER passed as a REAL -- but would
not update the original variable if the parameter were updated. Thus,
the input parameter would be 'thunked' in and the result 'thunked' out.

This concept is used extensively under OS/2 to provide pointer values
when a subroutine expects a segmented-model pointer and is passed a
flat-model pointer, or vice-versa. But thunking is not limited to this.
Also, conversion of pointers other than in the parameter passing
mechanism is NOT correctly called 'thunking'; this is a misuse of the
word.

If you want to call subroutines that use segmented-model pointers from
programs that use the flat memory model, you can do so quite liberally
and a good compiler will take care of the conversion for you. However,
there are 2 API calls, DosFlatToSel() and DosSelToFlat(), that will
allow you to convert pointers almost-manually; note that these API
calls use Optlink calling convention.

Regards

Dave

___
 X KWQ/2 1.2e X I thunk therefore I am ..... 32-bit compiler

--- Maximus/2 2.01wb