* On some NetBSD systems, the last unset command returns a
non-zero exit code
* This made the make(1) processes terminate immediately in
source-env-and-recurse, breaking the build.
Reported by: Thomas Klausner <wiz@NetBSD.org>
should be pretty safe.
The main risk if is something rely on the 'undef if empty' behavior of
the typical trap is the use of *= in dmake
a git grep of "\$(SYSTEM_" did not reveal any problematic case.
This introduce config_host.mk and config_build.mk
generated from config_host.mk.in
The intent is for that file to be both sourcable in a shell and includable
in a gnu Makefile.
Eventually all the environment set-up will be in config_host.mk.in
and so there won't be a need to source anything in shell before make.
This fixes cross-compiling from Linux to Android. My guess is that the
build system nowadays, when building the cross_toolset (for the build
platform), sources Env.Build.sh in an environment where Env.Host.sh
has already been sourced. So, if Env.Host.sh exports SYSBASE, that
stayed in force also after sourcing Env.Build.sh, with predictably
horrible results.
For some reason it didn't hurt cross-compiling from Mac OS X, which is
why I hadn't noticed.
It turns out even quite recent gdb (7.1) reportedly does not
support DWARF4, moreover while the produced binaries are smaller,
faster and make gdb initially use less memory, this quickly ceases
to the case as soon as one actually does anything in gdb.
Presumably the DWARF4 handling in current gdb is very inefficient
(http://sourceware.org/bugzilla/show_bug.cgi?id=13498).
This reverts commit 5214f7843c.
Conflicts:
solenv/gbuild/platform/WNT_INTEL_GCC.mk
solenv/gbuild/platform/unxgcc.mk
Generalized the OSX ppc workaround for ancient bison versions,
factored out for all gcc platforms. Put bison version detection
into configure accordingly, to switch on that, and not on platform.
