segmentation fault on down arrow keypress

[kspalaiologos]

segmentation fault on down arrow keypress

[doug-moen]

Thanks for the workaround Kamila. Merry Christmas and/or Newtonmas and/or Winter Solstice!

[doug-moen]

i have reproduced this in my own app.

[doug-moen]

valgrind gives a better stack trace

==182348== Conditional jump or move depends on uninitialised value(s)
==182348==    at 0x6EC301: std::vector<char32_t, std::allocator<char32_t> >::operator=(std::vector<char32_t, std::allocator<char32_t> > const&) [clone .isra.0] (vector.tcc:224)
==182348==    by 0x6F2BAB: assign (unicodestring.hxx:74)
==182348==    by 0x6F2BAB: history_move (replxx_impl.cxx:1657)
==182348==    by 0x6F2BAB: history_move (replxx_impl.cxx:1644)
==182348==    by 0x6F2BAB: replxx::Replxx::ReplxxImpl::history_next(char32_t) (replxx_impl.cxx:1636)

the code is

    UnicodeString& assign( UnicodeString const& other_ ) {
>       _data = other_._data;
        return *this;
    }
---
history_move
>   _data.assign( _history.current() );
---
Replxx::ACTION_RESULT Replxx::ReplxxImpl::history_next( char32_t ) {
>   return ( history_move( false ) );
}

_history._current._data is an uninitialized std::vector object.

std::vector of length 266338303, capacity 2 = {<error reading variable>}

[doug-moen]

_history.current() is implemented like this:

    UnicodeString const& current( void ) const {
        return ( _current->text() );
    }

So _current is an iterator (a pointer) that initially does not point at a valid object (because there is no history yet).
Here we are calling _history.current() at a time when _current doesn't point to a valid object, so current() returns garbage and then we crash.

[kspalaiologos]

Workaround: add repl.history_add(""); before any invocation to input. Might break when you log repl history somewhere, but it's not the case for me.