kern/164674: vsprintf/vswprintf return error (EOF) on success if __SERR flag is already set on file

[Matthew Story]

On Tue, Jan 31, 2012 at 6:06 PM, Matthew Story <matthewstory at gmail.com>wrote:

>
> >Number:         164674
> >Category:       kern
> >Synopsis:       vfprintf/vfwprintf return error (EOF) on success if
> __SERR flag is already set on file
>

Apologies for cross-posting, but I think that standards might be a more
appropriate responsible party for this than -bugs or kern.  See description
for more info, but the basic issue is that C99 and C11 stipulate that
fprintf should return -1 "if an output or encoding error occurred."
Currently, printf is encoding and outputting successfully (on line or fully
buffered FILEs), but returning -1 if the FILE has an error.  The C99/C11
specifications make no mention of FILE state in fprintf return conditions,
so this functionality seems to not conform to the specification, attached
patch resolves that issue.


> >Confidential:   no
> >Severity:       non-critical
> >Priority:       low
> >Responsible:    freebsd-bugs
> >State:          open
> [...snip]
> >Description:
> The printf family of functions behaves unpredictably if the file passed to
> the function has the __SERR flag set in _flags.  The one consistency across
> all of the cases detailed below is that regardless of the number of bytes
> written to the file, and the success/failure of that operation, the printf
> functions (printf, fprintf, wprintf, fwprintf) will always return -1 (EOF).
>
>  * For the case of an unbuffered file,  the operation currently fails
> transmitting no bytes, and returns -1.
>  * For the case of a buffered file, the operation transmits all bytes and
> the function returns -1.
>
> The problem is that the behavior here is inconsistent, and should not be.
>   The question is wether all should be made to fail consistently and
> transmit no bytes if __SERR is set on _flags, or if failure should be
> determined based on the success of byte transmission, discounting file
> state.
>
> Per the ISO/IEC 9899:201x (C11) Section 7.21.6.1, 14:
>
> The fprintf function returns the number of characters transmitted, or a
> negative value if an output or encoding error occurred.
>
> My reading of this specification is that success should be determined
> based on byte-transmission, and should not factor-in file state.  In
> addition to the ISO standard, the glibc implementation will reliably
> succeed with any error flag set if bytes are successfully transmitted
> (although it will transmit partial messages prior to successful conversion,
> which is unfortunate).
>
> The attached patch makes the operation on buffered and unbuffered files
> consistent across the affected printf/wprintf functions, determines
> success/failure based on successful byte-transmission alone, and preserves
> _flags state for __SERR as passed in.
>
> [...snip]


-- 
regards,
matt