Simon Josefsson wrote: > > I think that the proper fix is to change the prototypes of the 4 functions > > in crc.h from 'const char *buf' to 'const void *buf'. Just like memset(), > > memcmp(), etc. take 'void *', not 'char *', parameters. > > > > What do you think? > > Ouch. Ideally maybe uint8_t should have been used, but both uint8_t and > void* is an API change compared to current char*.
While the crc _implementation_ uses uint8_t*, from the caller's perspective: the caller passes the bounds of some region of memory. Whether these are 'unsigned char' values, 'char' values, 'bool' values, or even a struct that represents a network packet, the crc functions are made for all of these, right? What is the problem with changing the 'const char *' to 'const void *'? All callers that call the functions directly will be fine. Only code that uses function pointers will need change. For crc32_update, crc32_no_xor, crc32_update_no_xor, no such uses exist; for crc32, my guess is that no such uses exist either. > How about patching > test-crc.c to pass a char* to these functions, as they should receive? To me, that would be a not-so-good fix. Bruno [1] https://codesearch.debian.net/search?q=%5Cbcrc32_update%5Cb%5B%5E+%28%5D+-path%3Acrc.h+-path%3Acrc.c+-path%3AChangeLog&literal=0 [2] https://codesearch.debian.net/search?q=crc32_no_xor+-path%3Acrc.h+-path%3Acrc.c&literal=1 [3] https://codesearch.debian.net/search?q=crc32_update_no_xor+-path%3Acrc.h+-path%3Acrc.c&literal=1
