Installing SML/NJ

These are instructions for installing recent versions of SML/NJ on Unix-like systems, including MacOX X with the developer tools (Xcode) installed.

If you are running MacOS X but don't have the developer tools, you can install the system starting from the disk image that we provide. See MACOSINSTALL for details.

If you are running Windows, see WININSTALL.html instead.

1. If you are in a hurry (and you have network connectivity) ...

WARNING
The 64-bit version of SML/NJ is new as of 110.94 and has not been as thoroughly tested as the 32-bit version. If the installer complains about not being able to fetch those tarballs, try step 2.1.

2. If you have a bit more time...

2.1. If you do not have network connectivity...

Use some other means of downloading the necessary tarballs. Store them in directory $d. Then run the installer (config/install.sh).

Here is a table that shows which packages you will need to download depending on what you select in config/targets. Take the union of all the packages corresponding to your selection. Don't forget to include the "(always)" entry. (<arch> and <os> should match your host system.)

Keep in mind that making selections in config/targets there are dependencies between packages. For example, eXene requires cml and cml-lib. The installer script "knows" about these dependencies and will automatically calculate the transitive closure of the "requires" relation for you. Still, if you do not have network connectivity, then you should have downloaded tarballs for the entire transitive closure. (If you didn't, the installer will kindly remind you.)

If you sellect ... ... you need
anything 
boot.<arch>-<os>.tgz
runtime.tgz
src-smlnj 
MLRISC.tgz
cm.tgz
compiler.tgz
smlnj-lib.tgz
ml-yacc.tgz
system.tgz
old-basis old-basis.tgz
ml-yacc ml-yacc.tgz
ml-lex ml-lex.tgz 
ml-burg
  requires: ml-lex
ml-burg.tgz
smlnj-lib smlnj-lib.tgz
pgraph-util cm.tgz 
cml
  requires: smlnj-lib
cml.tgz
cml-lib cml.tgz 
eXene
  requires: cml, cml-lib
eXene.tgz
ckit ckit.tgz
ml-nlffi-lib ml-nlffi-lib.tgz 
ml-nlffigen
  requires: smlnj-lib, ckit
ml-nlffigen.tgz
mlrisc-tools MLRISC.tgz 
nowhere
  requires: mlrisc-tools
doc doc.tgz

2.2. Tweaking "preloads"...

Before running the installer (but after unpacking the config directory), you can customize the set of libraries that are pre-loaded (either directly or via CM's "autoload" mechanism). For this, edit config/preloads before running config/install.sh.

2.3. Using a different ftp server (for a step-1.-like installation)

If you have a local mirror for smlnj.cs.uchicago.edu/dist, then you can speed up installation by pointing the automatic URL fetch mechanism there. The URL of the repository is stored in config/srcarchiveurl. Edit this file before running config/install.sh.

2.4. Using locally stored tarballs

This is like step 2.3. -- except the URL in config/srcarchiveurl should read something like 
          file:///usr/src/packages/smlnj/working/110.94
(In other words, the URL should spell out the name of the local directory that happens to contain the tarballs. Notice that this method does not seem to work with certain versions of wget which do not understand the file:// scheme.)

2.5. What needs to be on my PATH?

You should include $d/bin into your PATH, where $d is the install directory.

2.6. Which of all those files do I really need?

After the installer completes, directory $d will be full of files and subdirectories. Not all of them are actually required if you want to use the system. In particular, if you are not interested in any of the source code, you can DELETE EVERYTHING EXCEPT: 
      $d/bin
      $d/lib

2.7. Can I move files to another directory after the fact?

Yes. Here is the procedure: Suppose you have installed everything under $d and now you want to move things to some other directory $e (e.g., e=/usr/share/smlnj):

2.8 Forcing the installer to use a particular URL getter program

By default, config/install.sh will attempt to use wget, curl, and lynx (in that order). If you already know which of these programs you want the installer to use, then set the environment variable URLGETTER to "wget", "curl", or "lynx", respectively. You can also set URLGETTER to any other command "foo" as long as it can be invoked as 
       $ foo
where <source> is the URL you want to fetch and <destination> is the file name where the result will be stored.

2.9 Where are those "CM" directories?

CM metadata (source skeletons, compiled files, stable libraries, GUIDs, ...) are now stored under a directory called ".cm". We changed this from the original "CM" because of name clashes on case-insensitive file systems. However, SML/NJ itself can work with any (reasonable) choice of name for the metadata directory -- but for consistency the choice has to be made at installation time. To choose a different name for metadata directories, run config/install.sh in the presence of an environment variable CM_DIR_ARC. For example, if you want to have all these directories be named "FOOBAR", then run the installer (assuming a Bourne-ish shell) as 
      $ CM_DIR_ARC=FOOBAR config/install.sh

3. Troubleshooting

3.1. You think you have network connectivity, but the installer fails

You probably do not have any of the following commands available on your system: wget, lynx, curl. Failsafe remedy: Proceed according to step 2.1., using whatever other FTP client you have. Alternative remedy: If you have some other command-line tool to fetch URLs, make a wrapper for it (if necessary) so that it can be invoked with precisely two arguments: and . Then set the environment variable URLGETTER to the name of this (wrapped) tool and try running the installer. Example: Suppose you don't have wget but a wget-lookalike named "xget". You could then make a shell script that wraps xget, e.g. using a shell script "myxget" consisting of the single line: 
         $ exec xget -nv -O $2 $1
Once you have myxget, simply run: 
         $ URLGETTER=myxget config/install.sh

3.2. The install fails with a message about 32-bit support

By default, SML/NJ is a 32-bit system. We have recently (as of version 110.94) added beta support for 64-bit Linux and macOS systems. See Section 1 above for more information. To install the 32-bit version on a 64-bit architecture, you must have support for running 32-bit binaries. On macOS (prior to macOS 10.15 Catalena), this support is standard, but some 64-bit Linux systems are not configured with 32-bit support. In such a case, you will ge the message 
        !!! SML/NJ requires support for 32-bit executables
during the install process. To fix this problem, you will need to acquire the 32-bit emulation libraries for your particular Linux distribution. For Debian (7.0 Wheezy and later) and recent version of Ubuntu, you will need to enable multiarch support. See https://wiki.ubuntu.com/MultiarchSpec for details, or try the following commands: 
        dpkg --add-architecture i386
        apt-get update
        apt-get install libc6:i386
You may also need to install the two following packages: 
        apt-get install gcc-multilib g++-multilib
For Red Hat Fedora (at least Fedora 16), you will need to install glibc-devel(i686): 
        yum install glibc-devel.i686
For Red Hat Enterprise Linux (or CentOS), you may also have to install the rpm package libgcc-multilib, e.g.: 
        yum groupinstall "Development tools"
        yum install libgcc.i686
        yum install glibc-devel.i686
Older instructions for RHEL and CentOS: download the libgcc-multilib package from rpmseek.com and then run 
        rpm -ivh libgcc-multilib-xxx.x86_64.rpm
where the libgcc-multilib package is the one you downloaded. For openSUSE use the YaST administration tool to install the gcc-32bit package.
David MacQueen, 2019.11.5