#8985 closed patch (fixed)
Pointer casts on LLP64 platforms lose precision
| Reported by: | dhewg | Owned by: | |
|---|---|---|---|
| Priority: | normal | Component: | Port: Win64 |
| Version: | Keywords: | ||
| Cc: | Game: |
Description
With mingw-w64 (gcc version 4.4.0 20090308 (experimental) (GCC)) a few of these errors occur:
backends/platform/sdl/graphics.cpp:1000: error: cast from ‘const byte*’ to ‘long int’ loses precision
64bit MS platforms use the LLP64 data model, so a "long" is only 32bit wide. There're a few instances where pointers get casted to long (to check for alignment), and that results in the above mentioned error.
While that can be worked around with -fpermissive, it's not really a nice solution, so I'll attach a patch that changes those casts from long to size_t. It compiled just fine afterwards (all engines enabled). Maybe it would be better idea to provide an common inline function to check for the alignment of a pointer.
Btw: It's working too ;) I don't have a win64 box, but Hkz just played BS/PSX on Vista64 with that binary.
Ticket imported from: #2685623. Ticket imported from: patches/1090.
Attachments (1)
Change History (8)
by , 16 years ago
| Attachment: | no-fpermissive.diff added |
|---|
comment:1 by , 16 years ago
comment:2 by , 16 years ago
Yes, rolling out our own PrtAsInt makes sense. Same is for IS_PTR_ALIGNED()
comment:3 by , 16 years ago
I just patched all of the source places, I found, to use the newly introduced IS_ALIGNED and fixed some other bad casts and an offset calculation related cast in the sky engine, now mingw64 compiles ScummVM just fine (check out buildbot for that).
I do not know if we really need to keep this one open now, thus I just set it to pending for now. Feel free to revert that if you have any problems with it.
comment:4 by , 16 years ago
| Resolution: | → fixed |
|---|---|
| Status: | new → pending |
comment:5 by , 16 years ago
This Tracker item was closed automatically by the system. It was previously set to a Pending status, and the original submitter did not respond within 14 days (the time period specified by the administrator of this Tracker).
comment:6 by , 16 years ago
| Status: | pending → closed |
|---|
comment:7 by , 6 years ago
| Component: | → Port: Win64 |
|---|
We should fix this, but I am not quite happy with the current patch: * size_t may not be sufficient to store a full pointer on some systems either. In stdint.h, there are type intptr_t and uintptr_t for that purpose, but stdint.h is not available everywhere, now is inttypes.h. Maybe we should provide our own PtrAsInt type or so, in scummsys.h ?
* Many of the changes affect code which just checks pointer alignment. For that it really shouldn't matter if the pointer is cast to a byte or a long long, only the 1-2 lowest bits are of interest. But if this causes errors or warnings, maybe we should wrap this into a macro / inline func: IS_PTR_ALIGNED(ptr) or so (OK, maybe with a second param which tells how many of the lowest bits shall be zero: 1, 2, 3...
* The OFFS style macros compute offsets, for that there is off_t (but also not quite portable); again, since this is about offsets within structs, an uint16 should be sufficient in most cases... But I guess casting to size_t should be OK, too.