Hello Stefan,
On 1/19/22 19:53, Stefan Rohrbacher wrote:
> To whom it may concern
>
> Dear Sir or Madam,
>
> I am writing to you because I believe I have found a typo in the
> manpage for the getline() function provided by stdio.h.
> I am using version 5.12 on Fedora 35 but it is also present in the
> current version 5.13.
> The typo is located in the "EXAMPLES" section at the bottom, there
> the variable "nread" in line 10 is of type "ssize_t" and I believe it
> is supposed to be of type "size_t":
>
> [...]
> int
> main(int argc, char *argv[])
> {
> FILE *stream;
> char *line = NULL;
> size_t len = 0;
> ssize_t nread; <-- supposed typo here!
>
> [...]
I'm not sure why you believe this. The prototype for getline(3) is:
ssize_t getline(char **restrict lineptr, size_t *restrict n,
FILE *restrict stream);
which uses 'ssize_t'. Or checking the glibc sources:
alx@ady2:~/src/gnu/glibc$ grep_glibc_prototype getline
libio/stdio.h:621:
extern __ssize_t getline (char **__restrict __lineptr,
size_t *__restrict __n,
FILE *__restrict __stream) __wur;
So, if we assign the result of getline(3) to a variable, it should be of
type 'ssize_t'. After that, we use the variable as input to fwrite(3),
which uses 'size_t', but using that type would be incorrect, since we
would be transforming the error code into a huge positive value, which
would be wrong. And since non-error 'ssize_t' values are always valid
'size_t' values, there's no problem in passing a 'ssize_t' to a 'size_t'
(after doing the proper error checking).
In general, avoid using unsigned types for things that are not bitfields
(or need modular arithmetic). Using mistakenly 'ssize_t' instead of
'size_t' is usually not dangerous. Doing the reverse, using mistakenly
'size_t' instead of 'ssize_t' is _very_ dangerous. This also applies to
other types with signed and unsigned counterparts.
Anyway, thanks for reporting potential bugs!
Cheers,
Alex
--
Alejandro Colomar
Linux man-pages maintainer; https://www.kernel.org/doc/man-pages/
http://www.alejandro-colomar.es/
