jcmds

jcmds build tool
================

``jcmds`` is a build tool. Its main inspiration is Cons, a Perl-based
"make-replacement" developed in the late 1990s. Common features are:

- MD5 checksums are used to reliably decide if files need to be rebuilt

- C/C++ #include file dependencies are automatically found by the tool

Differences are:

- ``jcmds`` can build in parallel (eg. -j8)

- ``jcmds`` uses no script language. Instead one just tells ``jcmds``
  what command there is to run. ``jcmds`` then decides for itself which
  commands may need to be run (this is a slight simplification; see below
  for more details).

Examples
--------

::

    $ cat > prog.cmds <<_END_
    gcc -o prog prog.o mod.o
    gcc -c prog.c -o prog.o
    gcc -c mod.c -o mod.o
    _END_

    $ jcmds -f prog.cmds
    gcc -c mod.c -o mod.o
    gcc -c prog.c -o prog.o
    gcc -o prog prog.o mod.o

    $ jcmds -f prog.cmds
    jcons: up-to-date: .

Two things can be seen here:

- the order of the commands given to ``jcmds`` does not matter.
  All commands contribute to the building of the dependency graph.
  The actual order of commands is determined by the structure of
  the dependency graph.

- the second invocation detects that all target are up-to-date
  so no commands get executed.

``jcmds`` has built-in knowledge of compiler/linker commands like those above,
but can of course not understand all possible commands. Say one command is::

    process_file src.txt tgt1.txt tgt2.txt

``jcmds`` can't known by magic that ``src.txt`` is the input and that there are
two output files ``tgt1.txt`` and ``tgt2.txt``. To "help" ``jcmds`` extra information
can be given as "comments" before the command, like this::

    # INPUT: src.txt
    # OUTPUT: tgt1.txt
    # OUTPUT: tgt2.txt
    process_file src.txt tgt1.txt tgt2.txt

Using this extra information, ``jcmds`` can add the command to the
dependency graph.

Built-in knowledge
------------------

``jcmds`` has built-in understanding of a number of commands::

    cp INPUT OUTPUT
    gcc ... -c ... NAME.c ... -o NAME.o ...
    gcc ... -o NAME ... NAME_1.o ... NAME_N.o ... LIBNAME_1.a ... LIBNAME_N.a
    ar qc LIBNAME.a NAME_1.o ... NAME_N.o

Different compilers like CLANG is also recognized, and C++ is also
handled (ie. ``clang``, ``g++`` and ``clang++``).

A command can also be on the form::

    cd some_dir && some_command ...

Then ``jcmds`` understands that file arguments to ``some_command`` are to be
evaluated relative to the directory ``some_dir``.

Extra information via comments
------------------------------

- ``# INPUT: <filename>``

  one or more input files to the command

- ``# OUTPUT: <filename>``

  one or more output files from the command

- ``# DEPEND: <filename>``

  one or more extra "hidden" dependencies

- ``# EXE_DEPEND: <filename>``

  one or more extra "hidden" dependencies of the command executable itself

... TODO: add more info ...


Jcmds and CMake
---------------

One way to use ``jcmds`` is in combination with ``cmake``. For example::

    mkdir build
    cd build
    cmake -GNinja ..
    ninja -t commands > commands.txt
    jcmds -f commands.txt