Some assumptions: (correct me if wrong)
Ignoring 16-bit stuff, VBA can be run on 32 or 64-bit Office hosts. 64-bit Office can only be run on a 64-bit OS, whereas you can run 32-Bit office on a 32 or 64-bit version of Windows/macOS/other Operating System.
As of VBA7, we have the LongPtr
type, which becomes a Long
on 32-bit Office (#Win64 = False
), LongLong
on 64-bit Office (#Win64 = True
), regardless of the OS bitness
My question: In APIs that deal with pointers (addresses in memory), does the OS bitness ever matter, or is it only the application running the code that we care about (32/64-bit Office host/VBA)?
Current understanding:
On the one hand I can see why it might not matter:
- VBA running in 32-bit Excel will have a 32-bit address space (regardless of OS)
- Therefore any pointers to its own memory should be 32-bits (i.e.
Long
s -LongPtr
gives us this) - Similarly 64-bit VBA (necessarily in 64-bit OS, but really could be anywhere) has a 64 bit address space (each pointer is itself 64-bits long and refers to a 64-bit wide block of memory)
- Therefore any pointers to its own memory should be 64-bits (i.e.
LongLong
s -LongPtr
gives us this) - So
LongPtr
accurately represents a pointer to my code's own memory, regardless of OS bitness
However I can imagine times where the OS bitness could matter
- Running 32-bit VBA on 64-bit OS, but wanting to refer to areas in memory of a different 64-bit application, the
LongPtr
type would evaluate toLong
and would be too short to represent the maximum pointer size available in that OS. - Similarly a 64-bit VBA accessing memory of a 32-bit application would find its pointer type too large to hold the address
- Now the much trusted
LongPtr
type is actually the wrong length to represent a pointer to an address in memory outside VBA's own address space!
Problems can now arise depending on OS bitness, not Office/VBA bitness. If we're running VBA7 on 32-bit OS, then LongPtr
will be the right length for any chunk of memory you could want to throw at it; it is a suitable pointer datatype in 99.9% of cases since everything the OS does is 32-bit (ignoring 16 bit).
However the same 32-bit VBA7 code instead running on a 64-bit OS would run into difficulties when trying to use LongPtr
to hold 32-bit addresses. I feel it's quite common to mix 32-bit and 64-bit applications on a 64-bit OS so this could be a real issue.
32-bit VBA6 and earlier applications running on 32 vs 64 bit Operating Systems could face similar issues, only without the help of LongPtr
Now I appreciate that's a pretty contrived situation, who would ever want to access another application's memory, right? In fact it's so contrived that I'm not sure I could ever come up with a situation where this is important and worth worrying about. Could an application ever obtain an address to the memory of another application with different bitness? Maybe there are some read-write protections preventing this.
Perhaps accessing another application's window handle would be one such occasion; that's public memory and perhaps the bitness of the window handle reflects the bitness of the application or the Operating system, in which case it would be feasible for my 32-bit VBA to want to hold a reference to a 64-bit Hwnd
, I'm not sure...
PS I'm aware there are situations other than pointer length where OS bitness may be important. I know of one; the SetWindowLong
function required different declarations on 64-bit vs 32-bit Windows - although IIUC that's now been solved with the SetWindowLongPtr
function which is identical for both. But any other similar quirks would be useful to know about, I'm only focussing on pointer length here because I have a problem that requires that specific info.
PPS come to think of it, can you even get OS bitness at compile time; I think you can infer it from MAC_OFFICE_VERSION
, and ofc Win64 = True
means 64-bit Office and de-facto 64-bit OS. But I'm not sure if there's a way to tell whether 32-bit VBA is running on 64-bit Windows...
LongPtr
as you pointed out. This is probably only really impacts those who have 32 bit office on 64 bit systems and the API uses 8 bytes for a param. I guess, just use the correct type given the API specs being awareLongPtr
can have differing values? – HicLongPtr
doesn't turn out to be as long as they thought. But firstly, that's not really the problem I'm concerned about. Suppose the API function isQueryPerformanceFrequency
. I would argue that in your example the user is using the wrong data-type - the API asks for aLARGE_INTEGER
, not a pointer, so whileLongPtr
happens to work out for them in 64 bit and doesn't in 32 bit office, it's the wrong choice either way. – TillfourdLongPtr
is suitable as a pointer in all situations, or whether the bitness of the OS can sometimes change the semantic meaning of an address in memory, and change how long a pointer data type needs to be. Secondly, I don't think OS bitness matters in the situation you describe. – Tillfourd