Re: [Mingw-users] gcc cc1plus bug

[Hugh M. <das...@gm...>]

Hi Keith,

On 04/07/07, Keith MARSHALL wrote:
> Under Cygwin, `gcc -mno-cygwin' is sufficient to ensure that the
> MinGW compiler is used, in place of Cygwin's own; for C++ code,
> you will also need CXX='g++ -mno-cygwin', and I guess for FORTRAN,
> F77='g77 -mno-cygwin'; (I'm far from certain of the latter).

I thought that gcc would normally invoke g++ with '-mno-cygwin'
atuomatically, like gcc invokes g++ when needed.  Is that correct?  As
for Fortran77, I'm not sure about that at all.

> `--host=x --build=y' is the standard autoconf way to initiate a
> cross-compile.  The purist in me says that, if you use this method
> of specifying the host compiler, then you should *not* need to use
> `-mno-cygwin' at all.  I did try it once upon a time[1], and IIRC,
> this is in fact the case.
>
> Conversely, if you use CC='gcc -mno-cygwin', and friends, you should
> not need `--host', (nor `host_alias'), nor `--build', IIRC.

I would have thought that using the CC='gcc -mno-cygwin' would invoke
MinGW, but isn't the '--host' configure option necessary to tell the
compiler that it would be building for 23/64 bit Windows?

> `host_alias=i686-pc-mingw32' is completely redundant here; configure
> sets it internally anyway, to identically the same value as you give
> with `--host'.

I wasn't aware of that.  Now I know.

> An important point to note: if you do adopt the `--host/--build'
> invocation syntax, then the value you specify with `--host' must
> *exactly* match the prefix by which your MinGW tool chain is
> identified, within your Cygwin file system; in Hugh's example
> above, this would imply that the MinGW gcc is installed into a
> PATH directory, with the name `i686-pc-mingw32-gcc.exe', (and
> similarly for `i686-pc-mingw32-g++.exe', `i686-pc-mingw32-ld.exe',
> etc.); on my Cygwin installation, no such tool chain exists, yet
> `-mno-cygwin' does invoke the MinGW compile mode[2].

Yes.  That is true.  Those files with the toolchain do not exist, but
Cygwin seems to default to its own set, while still invoking MinGW.

Hugh