Re: mktemp + openat and related questions

[Mike Frysinger <vapier at gentoo dot org>]

On Thursday 22 March 2012 16:02:31 D. V. Wiebe wrote:
> * The linker scolds me for using mktemp(3) and suggests I use mkstemp(3)
>   instead.  I'd be happy to do so but, AFAICT, it's not possible in this
>   situation (since I don't know the path to the directory and I need to
>   /at least/ make the temporary file on the same device as the directory
>   whose descriptor I have).  Am I right in disregarding this warning?

you could fchdir(dir) first if the side effects aren't a problem.  but there 
probably should be a mkstempat() func ...

> * The compiler scolds me for not checking the return value of mktemp(3),
>   but the mktemp(3) manual page says mktemp() always returns its input
>   (and sets it to the empty string on error).  This seems to jibe with
>   my reading of the mktemp() implementation in glibc-2.14.1.
> 
>   Am I right in concluding that the __wur attribute on this function
>   declaration in stdlib.h (and the accompanying comment "Returns
>   TEMPLATE, or a null pointer if it cannot get a unique file name" in
>   that header) is incorrect and can be disregarded?

the __wur probably should be dropped since POSIX too says that it always 
returns template.  as for the code, misc/mktemp.c in the latest git looks fine 
to me.  it says "null string" and not "null pointer".
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.