Re: PATCH: --with-build-sysroot, et. al.

[Mark Mitchell <mark at codesourcery dot com>]

DJ Delorie wrote:
> The logic looks OK.  But, thinking ahead, given that we currently
> support this:
>
> 	./configure --prefix=...
> 	make prefix=...
>
> Yet this:
>
> 	./configure --with-build-sysroot=...
> 	make SYSROOT_CFLAGS_FOR_TARGET=...
>
> It seems clumsy.  I wonder if SYSROOT_FOR_TARGET is a bettern choice,
> or even just "sysroot" if this is going to become popular enough.

We already have a "--sysroot" configure option, so we can't use that. 
The reason I chose "build" is that this is a property of the build 
machine; it's the location of the (target) sysroot on the build machine. 
 The "--sysroot" option is what I would call the host sysroot; it's the 
location of the (target) sysroot on the host machine.  And, yes, I 
concede that all of thse sysroots actually contain target files, so it's 
confusing!

> And if this sets the sysroot where gcc assumes things are installed,

It sets the location of the sysroot where gcc assumes things are 
installed *during the build process*; after that point, GCC uses 
whatever was provided with --sysroot.

> Another thought is... can we specify sysroot as a relative directory
> off prefix?  I mean, --with-sysroot='$(prefix)/...', so that changing
> prefix also changes sysroot?  I expect the logic for that would be
> more complex though.

Yes.  That would be nice, but it's, as you say, another level of hair. 
And, it wouldn't particular help me, because the issue at hand is that 
$prefix isn't accurate; we're configuring with one prefix and installing 
with another.

> I'm OK with you checking it in and changing the names later, though,
> if this is blocking anyone.

Thanks!  I'll give you a few hours to reply to this mail, in case I've 
clarified things in some scary way, and then check in.  And, of course, 
if you want me to change something after that point, I will.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304