Re: yacc.c: Extracting yysyntax_error

[Hans Aberg]

On 21 Sep 2005, at 20:43, Paul Eggert wrote:

> > The C++ standard, 18.7:5, says:
>
> I'm afraid that doesn't resolve the issue well enough for Bison.  A
> portable ISO C signal handler can't do much of anything -- all it can
> do is set a global variable.  Most real-world signal handlers in
> Bison-using programs do more than that (e.g., longjmp, remove files,
> etc.) but they are relying on POSIX guarantees not merely ISO C
> guarantees.
>
> POSIX allows portable signal handlers to do much of the stuff that
> real-world signal handlers need to do, but POSIX does not specify how
> C++ programs behave, and C++ does not specify how POSIX signal
> handling works.

To me it looked as though signal handling must happen within the POF  
C-like code, in order to fall within the C++ standard. In the case of  
Bison, this means that if signal handling takes places within the  
actions, and one wants to jump out of them, they must be first  
converted to C++ exceptions, if the standard C++ parser is used,  
i.e., the one using C++ containers. Alternatively, if conversions to C 
++ exceptions cannot be done, one can use a compile C as C++ parser,  
even though it won't be able to do cleanup in that case either. But,  
at least, the behavior then is not undefined.

> So there's no standard for how C++ exception handling
> interacts with POSIX signals, and in practice I suspect it tends to be
> a wild and woolly area.

The C++ standard says mixing a thrown exception with longjmp results  
in undefined behavior. The situation is quite simple: both exists  
within C++, but one must make sure signal handling and longjmp is  
isolated within the C-like POF C++ subset.

  Hans Aberg