Re: [Boost-Users] lexical_cast regression

From: Terje Slettebų (tslettebo_at_[hidden])
Date: 2003-04-18 16:19:03

>From: "Tom Matelich" <tmatelich_at_[hidden]>

>> From: Terje Slettebų
>>
>> >From: "Tom Matelich" <tmatelich_at_[hidden]>
>>
>> > I haven't gone and looked at the boost ml archives to track
>> > the reason for
>> > this down, but the addition of stream.unsetf(std::ios::skipws) in
>> > lexical_cast caused some of our code to fail.  I've commented it out
>> > locally, and was wondering why it was decided this is more
>> > appropriate default behavior.
>>
>> It was done to avoid the problems with throwing exceptions on
>> whitespace.
>> For example, without it, the following would throw a bad_lexical_cast:
>>
>> lexical_cast<std::string>(' '); // A space character
>>
>> What kind of code is the current version failing at?

>We have some legacy files that are converted to xml through some process (I
>didn't write it) and it ends up with space-padded strings for ints.  So the
>test case was dying converting " 86" to an int.

Ah, yes. After I read a similar report just now at the Boost list (reply
copied here), I guessed it might be about the same thing, which it was.

>I can see its purpose, except that example wouldn't happen because of the
>std::string overload, right?

Right. Apparently, the whitespace skipping only happens for reading _out_ of
the stringstream, not into it. However, if the above example is turned
around, then ir won't work (without whitespace skipping turned off):

lexical_cast<char>(" ");

Ironically, cases like the above example was a recurring error report,
leading to this change.

Regards,

Terje