Re: Possible problem with AC_CHECK_TYPES

[Assar Westerlund]

Peter Eisentraut <address@hidden> writes:
> Using "Intel(R) C++ Compiler for 32-bit applications, Version 7.1" (which
> also serves as a C compiler), this only draws a warning, not an error, if
> 'union semun' doesn't exist:
> 
> config.c(8): warning #279: controlling expression is constant
>       if ((union semun *) 0)
>           ^
> 
> config.c(11): warning #70: incomplete type is not allowed
>       if (sizeof (union semun))

> Is this compiler justified in doing so and/or should the Autoconf test be
> adjusted to cover this problem?

I would argue that's un-useful behaviour by the compiler.  The
standard seems to say that "you shall not" do sizeof(incomplete type)
and the compiler is required to issue a "diagnostic" for this.  The
standard doesn't really say that it should fail to compile this
program, but I cannot see how that would be useful.

Have you/can you ask intel about this?

It might make sense to have autoconf work-around this problem.  Do you
have any suggestion for a way to do so?