Re: compilation problems with ecj.jar

[Mohan Embar <gnustuff at thisiscool dot com>]

Hi Tom,

>Mohan> I disagree with the assessment that the current version is
>Mohan> wrong.  The current version defaults host_exeext to $ac_exeext,
>Mohan> but then unconditionally sets it to $cross_host_exeext if test
>Mohan> -n "${with_cross_host}", namely when the real host isn't the
>Mohan> same as the target.
>
>Hmm.  I think I am being confused by the comment about crossed-native
>compilers in the case where with_cross_host is empty:
>
>        # Crossed-native build (host=target and host!=build)
>        which_gcj=cross
>
>If it is possible to hit this case, then I think my original post was
>correct.  But perhaps it is not possible.  Ugh.

I think that it is both possible to hit this case and that your original
post (i.e. your assertion that the current version is wrong) was incorrect.

A concrete example of the aforementioned case is a
(i686-pc-linux-gnu,i686-pc-mingw32) cross being used to build a
(i686-pc-mingw32,i686-pc-mingw32) native compiler:

1. which_gcj=cross, meaning that the (i686-pc-linux-gnu,i686-pc-mingw32)
   cross compiler (named i686-pc-mingw32-gcj) should be used

2. host_exeext=${ac_exeext}

...since host=target (even though host!=build)

In this example, 2. says that we want an ecj1.exe (using the executable
suffix for i686-pc-mingw32) to be produced as part of the
(i686-pc-mingw32,i686-pc-mingw32) native compiler we're building.
This is true despite the fact that we're building on i686-pc-linux-gnu.

-- Mohan
http://www.thisiscool.com/
http://www.animalsong.org/