in file-posix.c's tr_sys_path_rename(), split assert(a && b) into assert(a)
and assert(b) so the crash will tell us which failed
[done]

in file-posix.c, several minor dead assigns of ret when it's declared. 
examples: tr_sys_path_exists(), tr_sys_path_get_info(), tr_sys_path_is_same(),
tr_sys_path_basename(), tr_sys_path_dirname()
[done]

in file-posix.c's tr_sys_path_get_info() and tr_sys_file_get_info()
functions, set_system_error() is never  called if the user passes a NULL
info arg. Also, is there any reason to not assert (info != NULL)?
[done]

file*.[ch]'s tr_sys_file_get_std() and tr_std_sys_file_t enum don't
seem to be used or tested, let's remove them or #if 0 them until
needed
[done]

torrent-magnet.c's IO in tr_torrentGetMetadataPiece() should use tr_sys_file
[done]

file-posix.c's tr_sys_file_seek() should use the off_t type for my_new_offset
and compare against (off_t) -1 ... off_t is signed
[done]

file-posix.c's tr_sys_file_read() and tr_sys_file_read_at() should use an
ssize_t type for my_bytes_read, and compare it against (ssize_t)-1.
Same for my_bytes_wirtten in tr_sys_file_write() and tr_sys_file_write_at()
[done]

file-posix.c's tr_sys_dir_get_current(), use tr_renew() instead of using
realloc() directly so that we don't mix & match tr_* and stdlib memory
management funcs
[done]

file-posix.c's tr_sys_dir_create(), dead assign of "my_error = NULL;" after
tr_error_propagate()
[done]

file-posix.c, maybe static asserts of TR_BAD_SYS_DIR in tr_sys_dir_open()?
[done]

file-win32.c, nice cleanup of the leaks in path_is_same()
[?]

file-win32.c, resolve_path... damn.
[?]

file-win32.c, more minor dead assigns of ret, and also of wide_path
[done]

error.c, tr_error_free(): might be good to set >message to NULL or DEADBEEF
after freeing? Not sure if necessary. these pointers can be dangling after
a call to error_propogate()
[done]

error.c, let's remove or comment out the unused functions tr_error_prefix(),
and tr_error_propagate_prefixed(), and error_prefix_valist() until we need them
[done]