Re: [PATCH] Improve behavior of AC_CHECK_TOOL; introduce AC_CHECK_TARGET_TOOL.

[Nathanael Nerode]

Peter Eisentraut wrote:
> Nathanael Nerode writes:
>
>
> > It also introduces an AC_CHECK_TARGET_TOOL macro, which checks for tools
> > normally native to the target, an oddity needed in gcc and probably useful
> > in other compiler projects which include runtime support libraries.
>
>
> I will not question why you need to know about a target tool during your
> build process, but it indeed seems to be an oddity, because a target tool
> would presumably run on the "host" platform which is in the general case
> not available during the build.  As such, I don't think this feature is
> suitable for general consumption.
You missed the point here.  Suppose we're building a Canadian Cross of 
GCC, for instance:
build=i686-pc-linux-gnu
host=i386-cygwin
target=powerpc-eabi

To build the compiler proper you need tools running on the build machine 
and targeting cygwin, which will be found under names like 
i386-cygwin-cc.  But to build the runtime libraries, you need tools 
running on the build machine and targeting powerpc, which will be found 
under names like powerpc-eabi-cc.

This is not to be confused with the version of gcc being *created*, 
which will run on the host machine and produce code for the target.

Also, consider a crossback:
build=i686-pc-linux-gnu
host=i386-cygwin
target=i686-pc-linux-gnu

Here, the tools to build the runtime libraries are in fact the native 
tools, and it's useful to *know* that.

> > The whole logic for that part would be much simpler, and would work better
> > in some corner cases, if I made AC_CHECK_TOOL & company depend on
> > AC_CANONICAL_HOST and AC_CANONICAL_BUILD, so I could just check whether
> > $host = $build.  However, I didn't want to add that dependency unless it
> > was considered desirable.
>
>
> Well, if you need to evaluate the host and build identity then you need
> to call AC_CANONICAL_HOST and AC_CANONICAL_BUILD.  That's the deal.

I'll be happy to submit that version of the patch, but I will not fix 
the testsuite for it, since I don't understand it well enough.

--Nathanael