https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87243

Eric Gallager <egallager at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|FSF needs to use xcrun on   |FSF GCC needs to do
                   |darwin18 to find system     |something special (like
                   |headers in SDK              |using xcrun) on darwin18 to
                   |                            |find system headers in SDK

--- Comment #5 from Eric Gallager <egallager at gcc dot gnu.org> ---
(In reply to Iain Sandoe from comment #4)
> All I'm asking is that that we describe the real bug (i.e. that we need to
> be able to find the headers when they are not installed in some Well Known
> Place).  It should also work for -mmacosz-version-min != current system
> revision (and therefore, might need to deal with finding libraries too - or
> else, for example, we won't be able to build m32 exes).
> 
> -----
> 
> As it happens, I had been giving this problem some thought - a couple of
> years ago.
> 
> What we don't want is to cripple Darwin's toolchain by making it call more
> and more executables for each invocation.
> 
> IFF you want to configure to use the current installed Xcode - you could
> just put 
> --with-sysroot='xcrun --sdk ... etc` on the configure line.
> 
> That doesn't solve the issue of the Xcode package being moved or the user
> issuing a --xcode-select ...
> 
> ... so ....
> let's state the problem we want to solve and the cases we want to cover -
> and then figure out how to do it.  The bug is not "we must use xcrun" the
> bug is "we need to find the headers under the following circumstances".

OK, retitling then.

> 
> I had in mind, for the record, some scheme using symlinks in the user's home
> directory (since using schemes writing into the /xxx dirs require elevated
> permissions).  There's still a penalty in looking up through a link c.f.
> direct access to things - but it would me much less than running a second
> process and parsing the output.
> 
> ====
> 
> JFTR: To support non-Apple-native clang builds I have to patch the LLVM
> cmake files to *avoid* the xcrun calls and use the SDk paths I want!.

Reply via email to