|
From: | Hans Aberg |
Subject: | Re: yet another C++ implementation bug to work around in Bison |
Date: | Mon, 6 Feb 2006 13:55:03 +0100 |
On 6 Feb 2006, at 09:01, Paul Eggert wrote:
I do not see why std::map would be needed as a part of the generated Bison parser, i.e., if the user code does not include it.It is not needed. The map stuff is used only by the user-code example in the Bison manual, which is tested by compiling it. Please see the Bison source code for details.
So, if not, the installation of Bison need not check it, but only as a part of a "make check".That would make sense, yes, but it will require that someone (not me!) actually implement such a change, and not just talk about it.
I just pointed it, in case somebody would be interested. :-) I guess the GNU philosophy that anything beyond GCC is extra. :-)
It is also strange if a C++ compiler, at this late date, cannot handle std::map properly.I don't find it strange at all, and I'm not a bit surprised by it.
I guess I do not know the world of contemporary C++ compilers that well. The problem was said (in the links you provided) depending on that operator++() was applied to std::map iterators, and the C++ standard from 1998 says that these iterators are bidirectional, and such iterators should admit this operator. So it has been some seven years plus time to fix the compiler.
Hans Aberg
[Prev in Thread] | Current Thread | [Next in Thread] |