[CDG5] MPW and Trampoline

[Max Poliakovski]

Hi Elliot,

thank you for your helpful suggestions!

After hours of downloading of historical development disks and searching
them for MPW packages, I finally managed to find a MrC compiler that
produces binary code very close to the code found in the Apple binary. That
compiler is resided on the E.T.O #21 disk from 1996 and is a part of the
MPW 3.4.2b3 pre-release. It identifies itself as "MrC C Compiler 2.0.2d1c6".

After playing a bit with this compiler and my code, I was able to produce
near byte-perfect output for string.c and fully byte-perfect output for
ofci.c!

The former contains seven functions; six of them are byte-identical to the
Apple binary; the 7th - memchr() -contains two instructions where operands
are swapped compared to Apple's code. I wasn't able to figure out why it
happens.

When I say "byte-identical", I actually mean anything except TOC references
and cross-module call offsets. These will be changed by PPCLink during the
linkage of the final binary. Tweaking PPCLink to generate byte-identical
output is a task on its own. I'm not sure I'll even succeed on it.

Anyway, round-tripping MrC has become feasible. Even if the 100% match
cannot be achieved, it's still a good validation of the correctness of the
decompiled code.

Stay tuned.

Cheers
Max


Am Mi., 9. Jan. 2019 um 05:22 Uhr schrieb Elliot Nunn <
elliotnunn at fastmail.com>:

> Happy new year!
>
> MrC round-tripping sounds like a very hard slog. After all, RISC is
> sometimes said to stand for "Relegate the Interesting Stuff to the
> Compiler"! Mind you, with your intimate familiarity with the
> architecture, you could well enjoy some breakthroughs.
>
> I can't seem to find MPW 3.4 in my archives. Is it old enough even to
> have a RISC compiler? I think that the early ones were distributed as a
> separate SDK. I suggest downloading all the E.T.O. images that you can
> find. They are the best library of MPW tools that I have found.
>
> Have you built any bootable Trampolines? That would be a very powerful
> validation of your reversal.
>
> > On 8 Jan 2019, at 11:05 pm, Max Poliakovski <
> maximumspatium at googlemail.com> wrote:
> >
> > Hi guys,
> >
> > happy new year!
> >
> > > Quoting Elliot's mail from 05-02-2018:
> > > I have had some success getting Apple's 68k C compiler to produce
> byte-accurate output. (A few things were in my favour here: simple C code,
> a simple C compiler and some sources for a previous version of the binary.)
> I always aim for byte-perfection after building my reversed sources
> (admittedly mostly assembly) because it lets me know when I've messed up.
> However, my approach is not always practical, and your focus on high-level
> hacking is likely to be more rewarding.
> >
> > I've played a couple of days with EMPW with the goal to get a
> byte-perfect output for the string.c module that consists of seven short
> functions (3-7 statements).
> >
> > As expected, I wasn't able to obtain byte-perfect output with MrC
> regardless different optimizations and even compiler versions.
> >
> > Below a short report:
> >
> > MrC 4.1.0 from MPW 3.5 GM emits code that is far from the code located
> in the Trampoline binary.
> >
> > I therefore downgraded to MrC 3.0.1 from 02-20-1997 that can be obtained
> here:
> https://www.staticky.com/mirrors/ftp.apple.com/developer/Tool_Chest/Core_Mac_OS_Tools/MPW_Website_Downloads/
> >
> > With MrC 3.0.1, I was able to perfectly reconstruct strlen() and
> memset() - both are very, very simple functions. After playing with
> optimization settings abit, the code for strcpy() got very close to the
> Apple's binary except a small difference in the register allocation ($r11
> and $r12 are swapped).
> >
> > I also tried MrC 4.1a1 but it seems to go in the wrong direction.
> >
> > I suspect that Apple might have used even an older compiler. The only
> possibility to check that out is to get a copy of MPW 3.4 (1995) which
> seems to be one of the first MPW versions including PowerPC compilers.
> >
> > Does somebody own MPW 3.4? I've seen only 3.3, 3.5 and PR so far...
> >
> > Cheers
> > Max
> > _______________________________________________
> > cdg5 mailing list
> > cdg5 at ucc.asn.au
> > https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
>
> _______________________________________________
> cdg5 mailing list
> cdg5 at ucc.asn.au
> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.ucc.gu.uwa.edu.au/pipermail/cdg5/attachments/20190205/a41676b6/attachment.htm