Home My Page Projects Code Snippets Project Openings SML/NJ
Summary Activity Forums Tracker Lists Tasks Docs Surveys News SCM Files

SCM Repository

[smlnj] View of /sml/trunk/src/cm/Doc/bt-02-basicusage.tex
ViewVC logotype

View of /sml/trunk/src/cm/Doc/bt-02-basicusage.tex

Parent Directory Parent Directory | Revision Log Revision Log

Revision 839 - (download) (as text) (annotate)
Thu Jun 7 20:28:44 2001 UTC (21 years, 3 months ago) by blume
File size: 8333 byte(s)
several internal changes related to C calls
% -*- latex -*-

\section{Basic usage}


To be able to use the bootstrap compiler, one must first install
SML/NJ (i.e., the interactive system that contains the compiler) as
well as both ML-Yacc and ML-Lex.

It is further necessary to have all ML source code for the system
available.  If the basic installation did not install this source
code, then one must now fetch all source archives and unpack them.
(There is an option for the SML/NJ installer that lets it install all
source trees automatically.  However, by default this feature is turned

The following list shows all required source packages.  (Path names
are shown relative to the SML/NJ installation directory.)

\item[{\bf src/system}] This archive contains the sources for the SML
Standard Basis library as well as lots of ``glue'' code.  The glue is
used for assembling a complete system from all its other parts.
Directory {\tt src/system} must be the current working directory at
the time the bootstrap compiler is started.
\item[{\bf src/MLRISC}] This is the implementation of the MLRISC
backend (the low-level optimizer and code generator) used by SML/NJ.
\item[{\bf src/cm}] This source tree contains most of CM's (and CMB's)
\item[{\bf src/compiler}] This is the implementation of the frontend
(parser, type checker, etc.) of the compiler as well as its high-level
optimizer (FLINT).
\item[{\bf src/ml-yacc}] This source tree contains the implementation
of ML-Yacc and its library ({\tt \$/ml-yacc-lib.cm}).
Technically, the sources of ML-Yacc itself are not needed (provided a
working executable for ML-Yacc has been installed), but the library
sources are.
\item[{\bf src/smlnj-lib}] This source tree hosts several sub-trees,
each of which implements one of the libraries in the SML/NJ library
collection.  For bootstrap compilation, the following sub-trees are
\item[{\bf src/smlnj-lib/Util}] This directory holds the sources for
{\tt \$/smlnj-lib.cm}.
\item[{\bf src/smlnj-lib/PP}] This directory holds the sourcse for
{\tt \$/pp-lib.cm}, i.e., the pretty-printing library.
\item[{\bf src/smlnj-lib/HTML}] This directory holds the sources for
{\tt \$/html-lib.cm}, a library for handling HTML files.  The need for
this library arises from the fact that {\tt \$/pp-lib.cm} statically
depends on it.  (The compiler does not actually use any services from
this library.)

\subsection{Invoking the bootstrap compiler}

To invoke the bootstrap compiler, first one has to change the current
working directory to {\tt src/system}:

$ cd src/system
%$emacs unconfuse

The next step is to start the interactive system and load the
bootstrap compiler.  This can be done in one of two ways:

\item Start the interactive system and then issue a {\tt CM.autoload}
command that causes the bootstrap compiler to be loaded. The resulting
session could look like this:
$ sml
Standard ML of New Jersey ...
- CM.autoload "$smlnj/cmb.cm";
val it = true : bool
\item Start the interactive system and specify {\tt \$smlnj/cmb.cm} on
the command line:
$ sml '$smlnj/cmb.cm'
Standard ML of New Jersey ...

\noindent{\it Note for frequent compiler hackers:} The {\tt makeml}
script (see below) builds the the interactive system in such a way
that {\tt \$smlnj/cmb.cm} is already pre-registered for autoloading.
Therefore, when using an interactive system built by {\tt makeml}
(as opposed to the original config/install.sh) there is no need
for loading the bootstrap compiler explicitly.

At this point one can invoke the bootstrap compiler by simply issuing
the command {\tt CMB.make()}:

- CMB.make ();

If {\tt CMB.make()} does not run to successful completion, you do not
have to start from the beginning.  Instead, fix the problem at hand
and re-issue {\tt CMB.make()} without terminating the interactive
session in between.  This tends to be a lot faster than starting over.

This process can be repeated arbitrarily many times until {\tt
CMB.make()} is successful.

A successful run looks like this:

- CMB.make ();
New boot directory has been built.
val it = true : bool

The return value of {\tt true} indicates success.  This means that (as
indicated by the message above the return value) a directory with
stable libraries and some other special files that are needed for
``booting'' a new interactive system has been created.

The name of the boot directory depends on circumstances.  A part of it
can be chosen freely, other parts depend on current architecture and
operating system.  For example, on a Sparc system running some version
of Unix, the default name of the directory is {\tt

There is also a similarly-named {em binfile directory} which is used
by {\tt CMB.make()} itself but which is not required for the purpose
of subsequent ``boot'' steps.

\subsection{Booting a new interactive system}

Once {\tt CMB.make()} has completed its work successfully (indicated
by its return value of {\tt true}), the next step is to build a new
heap image.  To do so, issue the command {\tt makeml} at the shell

$ ./makeml
Standard ML of New Jersey ...
./makeml: Heap image generated.
%$emacs unconfuse

The {\tt makeml} command generates a new heap image and also prepares
stable libraries to be used by this image.  Neither the heap image nor
the libraries will at this time be installed for permanent use
though.  This means that invoking {\tt sml} still starts the old

Again, the name of the generated heap image is variable and depends on
programmer choice, current architecture, and current operating
system.  For example, on an Intel x86 Linux machine, the default would
be {\tt sml.x86-linux}.\footnote{The operating-system dependent
portion of the name of a heap image is more discriminating than the
operating-system dependent portion of boot- or binfile directories.
On the same Intel x86 Linux machine, the name of the boot directory is
{\tt sml.boot.x86-unix}.}

\subsection{Testing the newly booted system}

To test-drive a newly booted system without installing it, issue the
{\tt testml} command at the shell prompt:

$ ./testml
%$emacs unconfuse

This starts an interactive system in a way very similar to {\tt sml},
but it uses the heap image and libraries from a previous run of {\tt

\subsection{Installing a newly booted system}

Once one is sure that a newly booted system is good enough to replace
the old system, issuing the {\tt installml} command will replace old
heap image and old libraries with those generated by the previous run
of {\tt makeml}.

$ ./installml
%$emacs unconfuse

This command will replace the system's heap image in {\tt
../../bin/.heap} and its libraries in {\tt ../../lib}.  However, it
will leave alone any unrelated libraries in {\tt ../../lib}.

Sometimes changes to the compiler will render any previously installed
libraries unusable.  In this case one should erase them prior to
issuing the {\tt installml} command:

$ rm ../../lib/*
$ ./installml

Libraries that were installed as part of the SML/NJ installation
process but which are unrelated to bootstrap compilation (e.g., {\tt
\$/inet-lib.cm}, CML, eXene) can be recovered (once they had been
removed) by going back to the installation directory and issuing the
{\tt config/install.sh} command again:

$ cd ../..
$ config/install.sh

Since some changes to the compiler also render old binfiles unusable,
one will occasionally have to remove those first (prior to re-running
{\tt config/install.sh}).  Binfiles for libraries unrelated to
bootstrapping are handled by CM (and not CMB), so the usual CM rules
for locating them apply.  (This means that in such a case the binfile
for {\tt $d$/$f$.sml} will be in
{\tt $d$/CM/{\it arch}-{\it os}/$f$.sml}
where {\it arch} is a string describing the CPU
architecture and {\it os} is a string describing the operating system
kind.  Example: {\tt x86-unix}.)

ViewVC Help
Powered by ViewVC 1.0.0