[m-users.] Installing 13.05.1 on Older Mac PPC-G5 and OS X 10.5.8

[Paul Bone]

On Mon, Apr 20, 2015 at 05:19:26PM -0700, Delmas Buckley wrote:
> Hello,
> 
> I'm trying to install mercury-13.05.1 on an older Mac PPC-G5 running OS X
> 10.5.8. Attempting to install in /usr/local. Have done the following which
> appears to have worked so far:
> 
> $ sudo sh configure --with-llds-base-grade=none
> 
> $ sudo make
> 
> My question relates to options to be used (or not) during make install.
> I've read down the line of options listed in the 'fine tuning' portion of
> INSTALL.
> 
> The set of library grades from which to choose is:
> 
> hlc.gc
> 
> hlc.gc.trseg
> 
> hlc.par.gc
> reg.gc
> reg.gc.debug
> reg.gc.decldebug
> reg.gc.memprof
> reg.gc.prof
> reg.gc.prof.deep
> reg.gc.trseg
> reg.gc.trseg.debug
> reg.par.gc.stseg
> 
> 
> I just do basic programming in Mercury, and, if possible, I'd like to be
> able to do some simple profiling. My question is: which options, if any,
> should I be using to accomplish those things?
> 
> $ sudo make install --enable-libgrades=< . . . >
> 

--enable-libgrades is a ./configure option.  There is a way to control this
at the "make install" stage by setting the LIBGRADES variable.

$ sudo make LIBGRADES=... install

I'm afraid the grades situation is complicated.  I want to improve the
documentation on this very soon.  But for now I suggest choosing either
"reg" or "none" for your LLDS base grade.  Reg may be faster but we haven't
tested it (or anything else) on a PPC machine for a long time, you're in
uncharted territory.  Then choosing grades such as:

hlc.gc,
reg.gc,
reg.gc.decldebug,
reg.gc.memprof,
reg.gc.prof,
reg.gc.profdeep,

This gives you "hlc.gc" which will probably be much faster than "reg.gc",
each of the three profiling grades and a debugging grade.  When you don't
need profiling or debugging then hlc.gc is your fastest grade.

Note that the deep profiling and possibly the debugging grades do not
support tail recursion.  If your programs run out of stack space there are
two potential solutions, use a segmented stack (.stseg), or use a very large
non-segmented stack.  The latter can be wasteful and inconvenient however in
this case I'd recommend it as I don't think the new-ish segmented stack code
has been tested on PPC.

Once again, you're in uncharted territory.  If you find something
interesting that others should know about it please let us know.

Cheers.


-- 
Paul Bone