On Friday 27 December 2002 01:49 pm, Gerald Pfeifer wrote:
On Fri, 27 Dec 2002, Greg Turner wrote:
If we really want to use off_t in this file (instead of cab_off_t) we need to #include sys/types.h, or we'll break non-Linux systems.
Short to medium term, do you agree with my fix (#include <sys/types.h>) so that Wine also builds on non-Linux systems?
Maybe. I don't object, since it should be harmless to add this, but I seem to be missing the point of why sys/types.h is needed at all. cab_off_t is just typedef'ed to UINT32, which should compile just fine with only winnt.h, I think. Is there a particular scenario you are encountering that fails? If so, what is the compile error?
The following line is the problem:
fol->offset[0] = base_offset + (off_t) EndGetI32(buf+cffold_DataOffset); ^^^^^
Here we use off_t (which requires sys/types.h), that's why I wrote "If we really want to use off_t in this file. Should this simply become cab_off_t?
Bingo. Although I never formally grepped for it, I made every attempt to convert all off_t's to cab_off_t's or signed integers as appropriate. So, yes, this is most certainly an error on my part, and your fix is likely the correct one. Out of curiosity, what is your platform? I am slightly surprised that off_t compiles at all without sys/types.h, even on linux...
Would you be so kind as to change that to cab_off_t, and then run the compile again to make sure that we've really eliminated all the off_t's in there? Let me know your results and I will submit a patch including this, and the renaming I discussed before.
A random thought: It's notable, that cab_off_t is unsigned, whereas off_t is signed. I began, but did not finish, the process of pouring over the source to ensure this is a nonissue. Perhaps, I might sleep easier if I just turned it back to a signed INT32, although if I recall, I used to think I had a good reason not to do that. Before I submit I'll take a look at that once more, as well.
Thanks for spotting this,