Missing unsigned char libc interface semantics

[sortie]

If you have these interfaces, you should audit them for whether they use unsigned char semantics (rather than signed char) as required by the standard: strcasecmp, memset, strncmp, memrchr, strchrnul, strcmp, stresep, strrchr, memchr, strncasecmp, memccpy, memcmp, fputc family, fgetc family, btowc, and wctob. Note also the <ctype.h> functions such as isspace has the domain of EOF and values representable as unsigned char, anything else is undefined behavior.

For instance, the sort order in strcmp is wrong.

[thepowersgang]

Audited and fixed list (commit 7d1c355)

• memset
• strcmp / strncmp
• strcasecmp / strncasecmp
• memchr
• memcmp
• fgetc / fputc

Absent:

• strchrnul - GNU Extension
• strsep - BSDisim
• memccpy - POSIX, may implement in posix library eventually
• btowc and wctob - No wide string support

[sortie]

I commented a bunch on the commit, found a bunch of issues. These kind of bugs will give weird third party software failures in the future. I can't comment on strncpy it seems, not part of the commit, but it has a bug where it accesses src before checking num. For instance, strncpy(dest, NULL, 0) might dereference NULL. Additonally, it doesn't zero out the rest of the buffer. Yeah, this strncpy implementation is entirely wrong. strncpy copies a string into a fixed width buffer, and if the input string is long enough it doesn't get nul-terminated, and if it's too short, it gets padded with nuls. It's an awful interface. Consider adding strlcpy and strlcat if you don't have them.

[sortie]

strncat is also wrong for that matter.

[sortie]

strstr looks wrong, it doesn't backtrack after a failed partial match. It's possible to do this correctly in O(N) time efficiently, might want to look up knuth morris pratt or what it's called.

[sortie]

I suggest adding abort or something to memcpy in case of overlapping copies, at least ind ebug builds, it'll help catch undefined behavior cases.

[sortie]

strndup doesn't check for malloc failure.

[sortie]

strtok_r is suspicious. The use of strchr there is very inefficient. I suggest you use strspn.

[ghost]

I think this should be fixed