This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] Simplify MinGW canadian crosses

On Tue, Aug 29, 2006 at 05:35:40PM +0200, Corinna Vinschen wrote:
>On Aug 29 11:09, Daniel Jacobowitz wrote:
>> On Tue, Aug 29, 2006 at 10:59:27AM -0400, DJ Delorie wrote:
>> > 
>> > > If you want to build some code that runs on mingw, I don't think
>> > > that having mingw tools installed is an unreasonable requirement.
>> > 
>> > This is how you *get* mingw tools installed.  The same logic that
>> > gives you a canadian (worst case) also gives you host-x-host.
>> Not so, unless I'm vastly confused.
>> Corinna is trying to generate --host=i686-mingw32 tools, with a
>> different --target.  This requires at least a --target=i686-mingw32
>> compiler coming from elsewhere.  That compiler can build the
>> --host=i686-mingw32 libraries, and usually should.
>> This is more like adding support for using the in tree newlib with an
>> arm-linux compiler so that you could build GCC to run on arm-linux,
>> without having to install an arm-linux C library first.  That's why
>> I'm dubious about the value.  But maybe Corinna has some good example
>> of when you want to do this?
>Sorry, but that's not the deal.  Using my patches, you can install a
>standard source tree, including gcc, gdb, binutils, [...], and last but
>not least the winsup directory on, say, a Linux machine, and then build
>a complete three stage canadian cross on *Linux*, which generates a
>i686-pc-mingw32-x-arm-elf toolchain.  You don't have to install the MinGW
>libraries and header files somewhere on the Linux machine and tweak the
>build process to find them.  Everything comes out of the same source
>tree.  From my point of view this simplifies stuff, it doesn't make it
>more complicated.

To be devil's advocate - I think the question is "Why should MinGW be
treated specially?"

I know that Cygwin is already treated specially in this regard (i.e.,
you can do an in-tree build to generate a cygwin cross-compiler without
necessarily having cygwin installed), and I believe that it is possible
to build newlib versions of gcc, but it apparently hasn't been a goal to
build MinGW without first installing a MinGW toolchain.

I think the answer is - since we already handle Cygwin and gcc's
configure already understands the winsup directory, why not just extend
configure to handle MinGW as well as Cygwin?  I don't think this
suggestion would make sense if we were talking about accommodating a
*new* directory in the build tree but since the MinGW directory is
almost handled now, I don't see any harm.

Btw, I agree with Daniel's suggestion of using
../config/no-executables.m4 if that's possible.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]