Release Notes:
______________


    Mn_Fit version 4.07/36 is hereby released.

    Given the Y2K problems with the CERN libraries, I only check
Mn_Fit for CERN library version 99 and later.

    There is one install script for all Unix machines - install.unix.
I need to know the value of the environment variable SYSTEM on the
relevant unix machine. The installation script for VMS systems is
called install.vms.

    Mn_Fit can be installed using patchy rather than cmz. The cmz
installation works with version 1.47 of cmz (95a Cernlib). The patchy
installation uses nypatchy if possible. Again I make no guarantee that
it works properly with ypatchy.

    The default version of Mn_Fit is now X11 for all platforms.

    readline version 2.2 is now available. If you install Mn_Fit with
patchy, you have to create the readline libraries yourself, if they
have not been installed for you. Note that if you want to change the
location of the readline library that is used, you should create a
link in the $MN_FIT/lib directory to the readline include files. If
neither a link nor a directory already exist, the install script will
prompt you for them.

    The C. Rippich version of MINUIT file, minuit.cmz (or .car) has
been renamed to cgr_minuit.cmz. You should delete any minuit.cmz,
minuit.kumac file in you cmz directory.

    Mn_Fit is available via my home page on www:
http://www-zeus.physik.uni-bonn.de/~brock/mn_fit.html. This has
replaced the previous standard homepage at CERN:
http://home.cern.ch/~brock/mn_fit.html. If you try to access the CERN
homepage you will automatically be switched to the Bonn page.

    Some machine dependent information is given below.

ALPHA:
______

When using COMIS on alpha machines running Unix, I often get the error:

Unaligned access pid=32259 <mn_fit_x.exe> va=5859c6c pc=2075a9c ra=2075a98 type=ldt

This does not appear to have any effect on the result, but writes an
awful lot onto the screen!


HP:
___

There appears to be a bug in the optimizer using the -O option unless
patch http://wsspinfo.cern.ch/patch/PHSS_6311 has been
included. It seems to blow up on lego plot transformations when doing
calculations such as (-2.0 - -2.0)!!
I have therefore changed the default optimization to +O1 for
this version of Mn_Fit.


Linux:
______

It assumes that you have gcc available and that g77 is also available
(f77 is often a link to g77).  In this case it all seems to work well.
Note that on the Bonn system the readline library is in /usr/lib and
the termcap library is in /usr/lib/termcap. I therefore have to link
with the options: -r /usr/lib -rt /usr/lib/termcap

The 99 and 2000 version of the CERN libraries need the libraries libXm
and libXt (which should both be in /usr/X11R6/lib or
/usr/local/lib. libXm (Motif) is sometimes not installed by
default. It should be taken from LessTif (which is also where CERN
gets it from). If this turns out to be a common problem I can also
link a static version of Mn_Fit.


SGI:
____

With the upgrade to IRIX version 6, there are a number of incompatible
compiler options available. The 97a (and earlier) release of the CERN
libraries uses the -mips2 flag. If you make the readline library
yourself for use with these old versions of the CERN libraries, you
should use the following options for configure:

CFLAGS='-g -O -mips2' ./configure

If you get a segmentation violation while loading it probably means
that incompatible mips options have been used somewhere.

The default options used in this version of Mn_Fit include -n32. This
is valid for CERN library version 99 and later (I do not know which
option the 98 version was compiled with). If you have to use an
earlier version off the CERN libraries then you have to look through
mn_util.car (or mn_util.cmz) and change any occurences of -n32 to -o32
and -mips3 to -mips2.

Compiling readline with the CC command used to cause an unresolved
external alloca. This is fixed with readline version 1.2 that is
distributed with Mn_Fit. Compiling with gcc is OK.


Sun:
____

The cc command gives an error in hkuip_h.h. This error does not occur
with gcc. If gcc is in /usr/local/lib, this directory is not put in
the default path when compiling from cmz so you must make a link from
your bin directory. e.g. cd ~/bin; ln -s /usr/local/bin/gcc gcc.
Alternatively, when you are prompted for the c compiler command you can
change it to: /usr/local/bin/gcc ...

On Solaris machines you also often have to give the full path of the
FORTRAN compiler. In addition, unless you are using the 96a version of
the CERN libraires (not yet tested), you should also give the compiler
option -DSolaris2 to avoid problmes with vfork.h.


VMS machines:
_____________

The standard way to define the cmz command on the VMS is:

cmz == "@cern:[pro.exe]cmz"

    This did not work with cmz/batch. You therefore had to redefine cmz to be
something like:

cmz == "$cern:[cmz.pro.exe]cmz.exe"

With the 95a release of the Cernlib the standard definition appears to
work properly, so thee change of the cmz command is no longer necessary.

Apollo:
_______

This version of Mn_Fit has not been tested on an Apollo.
