On 10/09/2015 02:57 PM, Roland McGrath wrote: >> However, I'm afraid to change signedness, if only because -1 is used in >> error conditions, e.g. returned from elf_update(). Someone checking <0 >> will be in for a bad time. > > Yeah, that sounds like a wise fear. But maybe Elf64_Off is OK (and > preferable even) for the cases that are not return values?
In practice we're limited to 2^63-1 off_t anyway, so the only benefit would be enforcing that they can't be negative, right? I dunno, seems like a small benefit that still has a chance to perturb users. Even just say Elf_Data.d_off comparisons may lead to "comparing signed and unsigned" warnings if we change our part, and I know I like -Werror in my builds. >> How about just int64_t? We already pull stdint.h for elf.h, so it seems >> fine to include and use that in libelf.h too. > > Sold! We already have <sys/types.h> which covers <stdint.h> too. But we > only ever had that for off_t/loff_t, so just change it to <stdint.h> and > use int64_t. I'll just do this for now. Thanks!
