qi::lit(<char>) parses into a '\0' in 1.47 and 1.48

[mathstuf@…]

In 1.46.0, qi::lit(<char>) would end up in the result. This changed in 1.47.0 such that they were not parsed into the result (qi::char_ is to be used instead). However, NUL characters now show up in the resulting token. Example code and output attached.

Synopsis:

Parser: +(qi::alnum | qi::lit('-'))

Input: "a-b-c"

1.46.0: "a-b-c"

1.47.0: "a\000b\000c"

1.48.0: "a\000b\000c"

The '\000' characters can be seen in gdb; std::ostream just shows "abc" instead.

[mathstuf@…]

Small example code

[mathstuf@…]

Output from 1.46, 1.47, and 1.48

[Joel de Guzman]

Use char_ instead (as you hinted):

tok %=

+( qi::alnum

| qi::char_('-') | qi::char_('_') );

lit has an "unused" attribute ( ​http://tinyurl.com/6lu2stx ) while char gives you the character type of the enclosing character encoding namespace (e.g. ascii, unicode, etc.). The way "unused" attribute is synthesized is undefined.

The behavior you see is an upshot of the optimizations and refinement we did for 1.47 and higher. As such, for alternatives, we only synthesize the advertized attribute in cases like this where the client passed in a non-variant (std::string).

[mathstuf@…]

Right. That's fine. The bug is that an "unused" attribute should not insert NUL characters in the resulting match.

[Joel de Guzman]

Fair enough. Let's open this for discussion then. There might be a "fix", but I'd like to discuss this in the Spirit mailing list. I'll re-open this ticket. Could you please post a message to the list here: ​http://boost-spirit.com/home/feedback-and-support/

Thanks!