On Oct 21, 2011, at 2:10 PM, Sebastian Pop wrote: > First, my patches conserve the semantics of configure if you do not > explicitly set the --target flag. If you want the current behavior, then > you can omit --target and the configure script will infer the value of > target from the host. > > My patches make --target useful, as the current semantics of --target > is that of nop. The target backed should default to the one specified > by the user at configure time. > > Also, without my patches I do not see how clang configured with > ../configure --enable-targets=arm --target=arm > would work on an x86 host machine if you explicitly disable the x86 > backend.
Some thinking out loud to open up the discussion: I understand this, but I'm just not sure _why_ we want to do such a thing. Smaller shipping binaries I suppose? Is it worth the headache of having clang understand the cross target but not the native target? I'm trying to avoid the "ship a host-x-target for every target you really want to support" headache that comes out of using --target. This would solve the need for setting a default target in the "canadian cross" style build (or constructing a small shell script that does the same), but mostly this feels like a "where do we draw the line" question and my only answer is "blue". Also, James's points should be answered as well. :) -eric
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
