Re: [boost] undefined behaviour in tuple_io.hpp (bug)(patch)

From: Rupert Kittinger (rkit_at_[hidden])
Date: 2005-08-25 13:20:55

Jonathan Wakely wrote:
> Rupert Kittinger wrote:
> 
> 
>>In the tuple_io.hpp header, there seems to be a slight problem in
>>extract_and_check_delimiter(). The following snippet is problematic:
>>
>>char c;
>>if (is_delimiter) {
>>    is >> c;
>>    if (c!=d) {
>>      is.setstate(std::ios::failbit);
>>    }
>>}
>>
>>If c cannot be read because is.eof(), the result of c!=d is undefined.
> 
> 
> IMHO the fix is neater as:
> 
>      if (is>>c && c!=d)
> 
> basic_ios::operator void*() exists to support this syntax rather than
> explicitly testing the state with good(), fail() etc.
> 
> jon
> 

Using is.good() is consistent with the rest of the file.

Another fine point that is not visible from the context diff is that the
code works as expected, because the only visible effect of the function
is the change in stream state. If is >> c failed because of is.eof(),
the failbit is already set, so it is irrelevant whether the if block is
executed. But I get false positives from valgrind in my unit tests, and
in this kind of code, the stack trace looks somewhat confusing :-)

cheers,
Rupert

-- 
Rupert Kittinger <rkit_at_[hidden]>