HACKING 15.4 KB
Newer Older
1
2
3
4
Bootstraping from the GIT repository
====================================

(If you are building from a tarball, skip this section.)
5
6
7

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
8
all of them under GIT because it can generate lots of changes or
9
10
conflicts.

11
Here are the tools you need to bootstrap the GIT tree, or more
12
generally if you plan to regenerate some of the generated files.
13
14
(None of these tools are required by end users installing a tarball
since the generated files they produce are distributed.)
15

16
  GNU Autoconf >= 2.61
17
  GNU Automake >= 1.11
18
  GNU Libtool >= 2.4
19
  GNU Flex (the version seems to matters, we used 2.5.35)
20
  GNU Bison >= 2.7
21
  GNU Emacs (preferably >= 24 but it may work with older versions)
22
  SWIG >= 2.0
23
  Doxygen >= 1.4.0
24
25
26
  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.
27

28
Bootstrap the GIT tree by running
29

30
  % autoreconf -vfi
31

32
and then go on with the usual
33

34
35
  % ./configure
  % make
36

37

38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
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documentation is out-of-date.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.


60
61
62
63
64
65
66
67
68
69
70
71
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


72
73
74
Debugging Libtool executables
-----------------------------

75
The executables generated in the various testsuite directories of Spot
76
are not real binaries.  Because we use libtool to compile the spot
77
78
library in a portable manner, these executables are just scripts that
run the actual binary after setting some environment variables so that
79
80
the OS can find the library in the build tree.

81
82
A consequence is that tools like gdb or valgrind, that expect to work
on a binary, will be confused by the script.  Example:
83
84
85
86
87
88
89
90

  % 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

91
92
The proper way to run any command on these fake binaries is via
libtool:
93
94
95
96
97
98
99

  % ../../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

100
101
102
103
104
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
105
106
107
108
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.

109
There is an undocumented feature of libtool that allows you to
110
shorthand "libtool --mode=execute" as "libtool execute" or even
111
112
113
114
115
116
117
118
"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.)
119
120


121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
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


138
139
140
141
142
Running coverage tests
----------------------

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

Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
143
  % ./configure CXX='g++ --coverage'
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
  % 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/.


162
163
164
165
166
167
168
Link-time optimizations
-----------------------

This is currently (April 2011) tricky to setup, because the
toolchain is not mature enough.

You need:
169
170
171
172
173
174
  1. a version of GCC (>= 4.5) with gold and pluing linker enabled.
  2. a version of Libtool that knows how to deal with
     -flto flags (Libtool 2.4 will not work -- currently only
     the development version does.)
  3. to use static libraries instead of shared libraries
     so that you get inter-libraries optimizations.
175
176
177

Here are example options to pass to configure:

178
179
180
181
  ./configure CC=gcc-4.6 CXX=g++-4.6 \
              --disable-devel --disable-debug \
              CFLAGS='-flto' CXXFLAGS='-flto' LDFLAGS='-fuse-linker-plugin' \
              --disable-shared --enable-static
182

183
Using --disable-debug prevents the -g flag to be passed to the
184
185
186
187
188
189
190
compiler, which seems to help avoiding some internal compiler errors.

Some binaries (like ltl2tgba) currently fail to compile (internal
compiler error), while most others (like randtgba, dve2check, randltl,
...) do fine.


191
192
193
194
195
196
197
198
199
200
201
202
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


203
204
Coding conventions
==================
205

206
Here some of the conventions we follow in Spot, so that the code looks
207
208
209
210
211
212
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).
213

214
215
216
217
218
219
220
221
222
223
224
225
226
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.

227

228
229
230
231
C++11
-----

  Spot uses some C++11 features, and therefore requires a C++11
232
233
234
235
236
237
  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.
238

239
240
241
242
243
244
245
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.
246
247
    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.
248

249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
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).
Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
266
267
    This will ensure that the header is not installed.
    Needless to say, no public header should include such a private
268
269
    header.

Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
270
271
272
  * 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.
273
274
275

  * Functions and classes that are public should be marked with
    the SPOT_API macro.  This macro is defined in misc/common.hh,
Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
276
    but you need not include it in a file that already includes
277
278
279
280
281
    another public header.

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

  * Read http://www.akkadia.org/drepper/dsohowto.pdf for more
Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
282
    information about how shared libraries work and why.
283

284
285
286
Comments
--------

287
  * The language to use is American English.
288
289
290
291
292

  * 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.

293
294
295
  * Prefer C++-style comments (// foo) to C-style comments (/* foo */).
    Use /// for Doxygen comments.

296
297
298
Formating
---------

299
  * Braces are always on their own line.
300
301
302
303
304

  * Text within braces is two-space indented.

    {
      f(12);
305

306
307
308
309
310
311
312
313
314
315
316
317
    }

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

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

318
  * Braces from function/structure/enum/class/namespace definitions
319
320
321
322
323
324
325
326
327
328
329
    are not indented.

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

  * The above corresponds to the `gnu' indentation style under Emacs.
330
331
332

  * Put return types and linkage specifiers on their own line in
    function/method _definitions_:
333
334
335
336
337
338
339
340
341

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

    This makes it easier to grep functions in the code.

342
343
344
    Function/method declaration are usually written on one line:

    int get_bar(int i);
345

346
  * Put a space before the opening parenthesis in control statements
347
348
349
350
351
352
353
354
355
356
357
358

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

  * No space before parentheses in function calls.
    (`some()->foo()->bar()' is far more readable than
359
    `some ()->foo ()->bar ()')
360
361
362

  * No space after opening or before closing parentheses, however
    put a space after commas (as in english).
363

364
    func(arg1, arg2, arg3);
365

366
367
368
369
370
371
372
373
374
  * 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));
375

Alexandre Duret-Lutz's avatar
typo    
Alexandre Duret-Lutz committed
376
377
  * No space after prefix unary operators, or before postfix unary
    operators:
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393

    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)
      {
        ...
      }

394
395
  * `else if' can be put as-is on a single line.

396
397
  * No line should be larger than 80 columns.
    If a line takes more than 80 columns, split it or rethink it.
398

399
400
401
402
403
404
405
406
407
    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.

408
  * Labels or case statements are back-indented by two spaces,
Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
409
    without space before the `:'.
410

Alexandre Duret-Lutz's avatar
Alexandre Duret-Lutz committed
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
    if (something)
      {
      top:
        bar = foo();
        switch (something_else)
	  {
	  case first_case:
	    f();
	    break;
	  case second_case:
	    g();
	    break;
	  default:
	    goto top;
	  }
      }

428
429
430
431
432
433
434
435
436
437
438
439
440
  * 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;
441
    The former declarations also allow you to comment each variable.
442
443
444
445

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

446
Naming
447
------
448
449
450
451
452
453
454

  * 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;
455

456
457
458
459
      typedef int int_array[];

    That is the style used in STL.

460
  * Private members end with an underscore.
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481

    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
    {
      ...
    };

482
  * Enum members also use capitalized name, with joined words.
483
484
485

  * C Macros are all uppercase.

486
487
  * Use *.hxx for the implementation of templates that are private to
    Spot (i.e., not installed) and need to be included multiple times.
488

489
Other style recommandations
490
---------------------------
491

492
493
  * 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.
494

495
496
497
  * Limit the scope of local variables by defining them as late as
    possible.  Do not reuse a local variables for two different things.

498
  * Do not systematically initialize local variables with 0 or other
499
    meaningless values.  This hides errors to valgrind.
500
501
502

  * Avoid <iostream>, <ostream>, etc. in headers whenever possible.
    Prefer <iosfwd> when predeclarations are sufficient, and then
503
    for instance use just <ostream> in the corresponding .cc file.
504
    (A plain <iostream> is needed when using std::cout, std::cerr, etc.)
505
506
507
508
509
510

  * 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.)
511
512
513

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