HACKING

Bootstraping from the GIT repository
====================================

(If you are building from a tarball, skip this section.)

Spot's gitlab page is at

  https://gitlab.lrde.epita.fr/spot/spot

The GIT repository can be cloned with

  git clone https://gitlab.lrde.epita.fr/spot/spot.git

Some files in SPOT's source tree are generated.  They are distributed
so that users do not need to tools to rebuild them, but we don't keep
all of them under GIT because it can generate lots of changes or
conflicts.

Here are the tools you need to bootstrap the GIT tree, or more
generally if you plan to regenerate some of the generated files.
(None of these tools are required by end users installing a tarball
since the generated files they produce are distributed.)

  GNU Autoconf >= 2.61
  GNU Automake >= 1.11
  GNU Libtool >= 2.4
  GNU Flex (the version seems to matters, we used 2.5.35)
  GNU Bison >= 2.7
  GNU Emacs (preferably >= 24 but it may work with older versions)
  Groff (a.k.a. GNU troff) >= 1.20
  SWIG >= 3.0 (for its better C++11 support)
  Doxygen >= 1.4.0
  Perl, with its Gettext module (it might be called something like
    liblocale-gettext-perl or p5-locale-gettext in your distribution)
  A complete LaTeX distribution, including latexmk and extra fonts
    like dsfont.sty.
  Python >= 3.2, IPython >= 2.3
  GraphViz

The following additional tools are used if they are present, or
only for certain operations (like releases):

  pandoc    used during Debian packaging for the conversion of
            IPython notebooks to html
  optipng   used during "make dist" if present, to optimize
            distributed png images
  R         used by some example in the documentation (the
            documentation will still compile without R, but that
            example will appear broken)
  ltl2ba    used in the generated documentation and the test suite
  ltl2dstar likewise
  ltl3dra   likewise
  spin      likewise
  glucose >= 3.0  likewise
  lbtt >= 1.2.1a  used in the test suite (ltlcross is now more
            powerful, but additional tests do not hurt)

If you use Debian or a similar distribution, the Dockerfile at
https://github.com/adl/spot-docker/blob/master/debuild/Dockerfile
lists all the Debian packages that should be installed to build
Debian packages out of the GIT tree.  Additionally, the script
https://github.com/adl/spot-docker/blob/master/debuild/install.sh
installs the third-party tools that do not have Debian packages.


Bootstrap the GIT tree by running

  % autoreconf -vfi

and then go on with the usual

  % ./configure
  % make


Tricks
======

Avoiding Doxygen runs
---------------------

When there is no documentation built (e.g., after a fresh checkout
of the GIT tree), when the configure.ac file has changed, or when
the Doxygen configuration has changed, the doc will be rebuilt.

This can take quite some time, even though recent version of Doxygen
have started to parallelize things.  If you have no interest
in generating the documentation, just use the "magic touch":

  touch doc/stamp

Do that right before running make.  The timestamp of doc/stamp
is compared to configure.ac and Doxygen.in to decide if the
documentation is out-of-date.  The above command pretends the
documentation has just been built.


Avoiding org-mode runs
----------------------

The files in doc/org/ are org-mode files (a mode of Emacs that we use
to author documents that embed executable snippets), they are used to
generate the doc/userdoc/ HTML documentation.  If for some reason you
don't have emacs, or you simply want not to rebuild these files, use
another "magic touch":

  touch doc/org-stamp


Debugging Libtool executables
-----------------------------

The executables generated in the various testsuite directories of Spot
are not real binaries.  Because we use libtool to compile the spot
library in a portable manner, these executables are just scripts that
run the actual binary after setting some environment variables so that
the OS can find the library in the build tree.

A consequence is that tools like gdb or valgrind, that expect to work
on a binary, will be confused by the script.  Example:

  % cd src/tgbatest
  % file ltl2tgba
  ltl2tgba: POSIX shell script text executable
  % gdb -q ltl2tgba
  "/home/adl/git/spot/src/tgbatest/ltl2tgba": not in executable format: File format not recognized
  (gdb) quit

The proper way to run any command on these fake binaries is via
libtool:

  % ../../libtool --mode=execute file ltl2tgba
  /home/adl/git/spot/src/tgbatest/.libs/lt-ltl2tgba: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), for GNU/Linux 2.6.18, not stripped
  % ../../libtool --mode=execute gdb -q ltl2tgba
  Reading symbols from /home/adl/git/spot/src/tgbatest/.libs/lt-ltl2tgba...done.
  (gdb) quit

You can see that libtool turns ltl2tgba into .libs/lt-ltl2tgba, but it
also sets environment variables so that the dependent shared libraries
will be found.

If you are building Spot from the GIT repository, the libtool script
generated the root of the build tree should be the same as the libtool
script that is installed on your system.  So you can simply run
libtool instead of ../../libtool.

There is an undocumented feature of libtool that allows you to
shorthand "libtool --mode=execute" as "libtool execute" or even
"libtool e".  But you might also find convenient to define an alias, a
function, or a script to make that invocation even shorter.
For instance:

  alias le='libtool --mode=execute '

(The trailing space makes it possible to follow this command by
another aliased command.)


Profiling with callgrind
------------------------

Install valgrind and kcachegrind.

Then run the command you want to profile through valgrind's callgrind
tool.  For instance:

  % libtool e valgrind --tool=callgrind ltl2tgba -f 'GFa & GFb'

This will output a file called 'callgrind.PID' where PID is the
process ID printed during valgrind's run.  Load this file with
kcachegrind to get a graphical summary.

  % kcachegrind ./callgrind.PID


Running coverage tests
----------------------

First, compile (and link) Spot with coverage enabled.

  % ./configure CXX='g++ --coverage'
  % make

Then run the test suite (or any program you want to study).

  % make check

Executing programs using Spot will generate a lot of *.gc* files
everywhere.  Collect these using lcov:

  % lcov --capture --directory src --output spot.info

Finally generate a coverage report in HTML:

  % genhtml --legend --demangle-cpp --output-directory html spot.info

This should create the directory html/.


Link-time optimizations
-----------------------

This is currently a bit tricky to setup, because the toolchain is not
mature enough.  However this is getting better and better.  The Debian
packages we built nightly are mostly built with link-time optimization
(the shared library uses link-time optimization, but the command-line
binary are built without because of some bug with exception
propagation).


You need:
  1. a version of GCC (>= 4.9) with gold and pluing linker enabled.
  2. a version of Libtool that knows how to deal with
     -flto flags (Libtool 2.4.2 will work)

Here are example options to pass to configure to build a static
version with link-time optimization:

  ./configure CC=gcc-4.9 CXX=g++-4.9 \
              --disable-devel --disable-debug \
              CFLAGS='-flto' CXXFLAGS='-flto' LDFLAGS='-fuse-linker-plugin' \
              --disable-shared --enable-static

If you want to build a shared library, see in debian/rules how it is
done.


Log driver for testsuite
------------------------

The PASS/FAIL status for each test of the testsuite is printed by
tools/test-driver.  This script can be changed to format the output
differently.  When we use Teamcity (for continuous integration) we
change the output format to something that Teamcity will understand
with:

  make check TEST_LOG_DRIVER=$PWD/tools/test-driver-teamcity


Coding conventions
==================

Here some of the conventions we follow in Spot, so that the code looks
homogeneous.  Please follow these strictly.  Since this is free
software, uniformity of the code matters a lot.  Most of these
conventions are derived from the GNU Coding Standards
(http://www.gnu.org/prep/standards.html) with the notable exception
that we do not put a space before the opening parenthesis in function
calls (this is hardly readable when chaining method calls).

Besides cosmetics, some of these conventions are also here
to prevent bugs and make it easier to devise safety checks.

The directory src/sanity/ contains some scripts that are executed
during 'make check' or 'make installcheck' to check some of the
conventions discussed below.

For instance we have a check to ensure that any header can be included
twice, and we have another check to ensure that any header contains a
include guard that follow our naming convention.  This way we do not
forget guards, and we do not forget to rename them when a file is
copied into another one.


C++11
-----

  Spot uses some C++11 features, and therefore requires a C++11
  compiler.  The code relies on features that are not available in
  version of g++ older than 4.8, so this is our minimum requirement
  for now.  Avoid features that require 4.9.

  Reasonably recent versions of clang should work as well.  Our
  build farm has clang++ 3.5.

Encoding
--------

  * Use UTF-8 for non-ASCII characters.

  * If you edit files encoded in Latin-1 (the original default
    encoding for the project), feel free to convert them to UTF-8.
    In emacs the simplest way to convert the file is to add a comment
    with -*- coding: utf-8 -*- at the top or bottom of the file.

Exporting symbols
-----------------

  Since we are building a library, it is important to make a clear
  distinction between what is private and what is public.  In our
  setup, everything is private by default and has to be explicitely
  made public.

  * If a private symbol is needed only by one module, keep it inside
    the *.cc file, in an anonymous namespace.  Also mark it as static
    if it is a function or variable.  This is the best way to let the
    compiler and linker know that the symbol is not used elsewhere.

  * If a symbol could be used by several modules of the library but
    should still be private to the library, use a *.hh/*.cc pair of
    files, but list both files in the _SOURCES variable of that
    directory (see for instance weight.hh in tgbaalgos/Makefile.am).
    This will ensure that the header is not installed.
    Needless to say, no public header should include such a private
    header.

  * The directory src/priv/ can be used to store files that are
    globaly private the library, and that do not really belongs to
    other directories.

  * Functions and classes that are public should be marked with
    the SPOT_API macro.  This macro is defined in misc/common.hh,
    but you need not include it in a file that already includes
    another public header.

  * Do not make a symbol public just because you can.

  * Read http://www.akkadia.org/drepper/dsohowto.pdf for more
    information about how shared libraries work and why.

Comments
--------

  * The language to use is American English.

  * When comments are sentences, they should start with a capital and
    end with a dot.  Dots that end sentences should be followed by two
    spaces (i.e., American typing convention), like in this paragraph.

  * Prefer C++-style comments (// foo) to C-style comments (/* foo */).
    Use /// for Doxygen comments.

Formating
---------

  * Braces are always on their own line.

  * Text within braces is two-space indented.

    {
      f(12);

    }

  * Anything after a control statement is two-space indented.  This
    includes braces.

    if (test)
      {
        f(123);
	while (test2)
	  g(456);
      }

  * Braces from function/structure/enum/class/namespace definitions
    are not indented.

    class foo
    {
    public:
      Foo();
    protected:
      static int get_mumble();
    };

  * The above corresponds to the `gnu' indentation style under Emacs.

  * Put return types and linkage specifiers on their own line in
    function/method _definitions_:

    static int
    Foo::get_mumble()
    {
      return 2;
    }

    This makes it easier to grep functions in the code.

    Function/method declaration are usually written on one line:

    int get_bar(int i);

  * Put a space before the opening parenthesis in control statements

    if (test)
      {
        do
	  {
	    something();
	  }
	while (0);
      }

  * No space before parentheses in function calls.
    (`some()->foo()->bar()' is far more readable than
    `some ()->foo ()->bar ()')

  * No space after opening or before closing parentheses, however
    put a space after commas (as in english).

    func(arg1, arg2, arg3);

  * No useless parentheses in return statements.

    return 2;    (not `return (2);')

  * Spaces around infix binary or ternary operators:

    2 + 2;
    a = b;
    a <<= (3 + 5) * 3 + f(67 + (really ? 45 : 0));

  * No space after prefix unary operators, or before postfix unary
    operators:

    if (!test && y++ != 0)
      {
        ++x;
      }

  * When an expression spans over several lines, split it before
    operators.  If it's inside a parenthesis, the following lines
    should be 1-indented w.r.t. the opening parenthesis.

    if (foo_this_is_long && bar > win(x, y, z)
        && !remaining_condition)
      {
        ...
      }

  * `else if' can be put as-is on a single line.

  * No line should be larger than 80 columns.
    If a line takes more than 80 columns, split it or rethink it.

    This makes it easier to print the code, allow people to work on
    small screens, makes it possible to display two files (or an
    editor and a terminal) side-by-side, ...

    This also puts some pressure on the programmer who writes code
    that has too much nested blocks: if you find yourself having to
    code between columns 60 and 80 because of identation, consider
    writing helper functions to simplify the structure of your code.

  * Labels or case statements are back-indented by two spaces,
    without space before the `:'.

    if (something)
      {
      top:
        bar = foo();
        switch (something_else)
	  {
	  case first_case:
	    f();
	    break;
	  case second_case:
	    g();
	    break;
	  default:
	    goto top;
	  }
      }

  * Pointers and references are part of the type, and should be put
    near the type, not near the variable.

      int* p;        // not `int *p;'
      list& l;       // not `list &l;'
      void* magic(); // not `void *magic();'

  * Do not declare many variables on one line.
    Use
      int* p;
      int* q;
    instead of
      int *p, *q;
    The former declarations also allow you to comment each variable.

  * The include guard for src/somedir/foo.hh is
    SPOT_SOMEDIR_FOO_HH

Naming
------

  * Functions, methods, types, classes, etc. are named with lowercase
    letters, using an underscore to separate words.

      int compute_this_and_that();

      class this_is_a_class;

      typedef int int_array[];

    That is the style used in STL.

  * Private members end with an underscore.

    class my_class
    {
    public:
      ...
      int get_val() const;
    private:
      int name_;
    };

  * Identifiers (even internal) starting with `_' are best avoided
    to limit clashes with system definitions.

  * Template arguments use capitalized name, with joined words.

    template <class T, int NumberOfThings>
    class foo
    {
      ...
    };

  * Enum members also use capitalized name, with joined words.

  * C Macros are all uppercase.

  * Use *.hxx for the implementation of templates that are private to
    Spot (i.e., not installed) and need to be included multiple times.

Other style recommandations
---------------------------

  * The original C++98 code used 0 for null pointers (and never NULL).
    Feel free to replace these by uses of C++11's nullptr instead.

  * Limit the scope of local variables by defining them as late as
    possible.  Do not reuse a local variables for two different things.

  * Do not systematically initialize local variables with 0 or other
    meaningless values.  This hides errors to valgrind.

  * Avoid <iostream>, <ostream>, etc. in headers whenever possible.
    Prefer <iosfwd> when predeclarations are sufficient, and then
    for instance use just <ostream> in the corresponding .cc file.
    (A plain <iostream> is needed when using std::cout, std::cerr, etc.)

  * Always declare helper functions and other local class definitions
    (used in a single .cc files) in anonymous namespaces.  (The risk
    otherwise is to declare two classes with the same name: the linker
    will ignore one of the two silently.  The resulting bugs are often
    difficult to understand.)

  * Always code as if the person who ends up maintaining your code is
    a violent psychopath who knows where you live.