dMZX Forums: config.sh doesn't support PPC or other non-x86/amd64/arm architectures on linux -> Archived MegaZeux Bugs -> Tracker

I could swear MZX used to run fine on PPC Linux, and it certainly runs fine on PPC OSX last I checked.

config.sh however spits out "Add a friendly MACH to config.sh" when I try to run it on Linux PPC. I have absolutely 0 idea what this means, but I looked into config.sh and found this section that seems to be the culrpit:

elif [ "$PLATFORM" = "unix" -o "$PLATFORM" = "unix-devel" ]; then
        OS="`uname -s`"
        MACH="`uname -m`"
[...]
        if [ "$MACH" = "x86_64" -o "$MACH" = "amd64" ]; then
                ARCHNAME=amd64
                LIBDIR=lib64
                # FIXME: FreeBSD amd64 hack
                [ "$UNIX" = "freebsd" ] && LIBDIR=lib
                if [ "$MODULAR" = "true" ]; then
                        echo "ARCH_CFLAGS+=-fPIC" >> platform.inc
                        echo "ARCH_CXXFLAGS+=-fPIC" >> platform.inc
                fi
        elif [ "`echo $MACH | sed 's,i.86,x86,'`" = "x86" ]; then
                ARCHNAME=x86
                LIBDIR=lib
        elif [ "`echo $MACH | sed 's,^arm.*,arm,'`" = "arm" ]; then
                ARCHNAME=arm
                LIBDIR=lib
        else
                echo "Add a friendly MACH to config.sh."
                exit 1
        fi

It seems a case of simply looking at uname -m (which is just "ppc"), and adding a line along the lines of

        elif [ "$MACH" == "ppc" ]; then
                ARCHNAME=ppc
                LIBDIR=lib

above the else, but if this is the case, why is there a hardline "Oops, you're compiling on an unsupported platform, stopping now" bit with an inscrutable error message in the first place? Its not as though MegaZeux generally seems to have all that many unfixable issues compiling on random architectures that all run linux.

As in, why isn't the else simply this:

        else
                ARCHNAME=$MACH
                LIBDIR=lib # The default for most systems
                echo "WARNING: Compiling on an unsupported architecture. Add a friendly MACH to config.sh."

So, yeah, I guess I'm confused by the design decisions here mostly, but submitting this as a bug.

Your suggestion of a liberal fall-through sounds fine to me.

The code block is there for two reasons. Firstly, to get LIBDIR right, but you're right that on non-multi-arch Linux machines that's going to be 'lib'. Secondly, ARCHNAME is used for the updater; not really a problem since it's disabled by default for Linux, but it was developed and does work on Linux. We can't have a million different uname -m variants handled in the updater, so we have to compress this down to the true ELF/ABI class.

Obviously with your 'guesswork' fallback you run the risk of enabling the updater and getting undefined results, but that's better than just flat failing to compile.

--ajs.

Thanks for the explanation, I understand the purpose of that section much better now.

And yeah, I think having a liberal fall-through like the above is a good idea, and have implemented the bit I suggested above in my own copy that I'm using with my VMs.

Updating status to: Proposed Fix
Updating severity to: 3 - Medium

Hey, I hate to be That Guy, but since this was reported around the same time as the recently fixed armhf linker bug, I'm kind of bumping this to see if my proposed fix was ever actually accepted into the code. Was it?

Not certain—I'll look into it...

Updating status to: Fixed
Issue fixed in: GIT

The answer is no, it wasn't. Added in GIT 572c534.

Moving to: Archived MegaZeux Bugs