/ 
parrot-0.5.3
/ project::cage_cleaners_guide

Cage Cleaner Guide

From docs/project/roles_responsibilities.pod:

Fixes failing tests, makes sure coding standards are implemented, reviews
documentation and examples. A class of tickets in the tracking system (RT)
has been created for use by this group. This is an entry level position, and
viewed as a good way to get familiar with parrot internals.

Testing Parrot after making a code cleaning change

To be really really sure you're not breaking anything after doing code cleaning or attending to the newspaper at the bottom of our Parrot's cage here are is the process I (ptc) go through before committing a new change:

make realclean > make_realclean.out 2>&1
perl Configure.pl > perl_configure.out 2>&1
make buildtools_tests > buildtools_tests.out 2>&1
make test > make_test.out 2>&1

Then I diff the *.out files with copies of the *.out files I made on a previous test run. If the diffs show nothing nasty is happening, you can be more sure that you've not broken anything and can commit the change. Then rename the *.out files to something like *.out.old so that you maintain reasonably up to date references for the diffs.

This process should be put into a script and stored somewhere...

Parrot Cage Cleaners high-level goals

Smoke testing on many platforms with many compilers

The more platforms we have, the more likely we are to find portability problems. Parrot has to be the most portable thing we've created.

More platforms also means more compilers. Maybe your DEC compiler is more picky than gcc, and spews more warnings. Good! More opportunities for cleaning!

icc

icc is the Intel C/C++ Compiler and is available for free for non-commercial use. To use icc to build parrot, use the following arguments to Configure.pl:

perl Configure.pl --cc=icc --ccflags=-I<icc-header-path>

(courtesy of Steve Peters, steve at fisharerojo dot org).

For example, those using icc version 9.1 series, the configure line is:

perl Configure.pl --cc=icc --ccflags=-I/opt/intel/compiler91/include/

Compiler pickiness

Use as many compiler warnings as we possibly can. The more warnings we enable, the less likely something will pass the watchful eye of the compiler.

Note that warnings may not just be -W flags. Some warnings in gcc only show up when optimization is enabled.

splint

Splint (http://www.splint.org) is a very very picky lint tool, and setup and configuration is a pain. Andy has tried to get Perl 5 running under it nicely, but has met with limited success. Maybe the Parrot will be nicer.

Solaris lint

Sun has made its dev tools freely available at http://developers.sun.com/prodtech/cc/. Its lint is the best one out there, except from Gimpel's FlexeLint (http://www.gimpel.com/html/flex.htm) which costs many dollars.

Enforcing coding standards, naming conventions, etc

  • Automatic standards checking

    The docs in filename here explains what our code should look like. Write something that automatically validates it in a .t file.

  • const checking

    Declaring variables as const wherever possible lets the compiler do lots of checking that wouldn't normally be possible. Walk the source code adding the const qualifier wherever possible. The biggest bang is always in passing pointers into functions.

Why consting is good

In Perl, we have the use constant pragma to define unchanging values. The Readonly module extends this to allow arrays and hashes to be non-modifiable as well.

In C, we have const numbers and pointers, and using them wherever possible lets us put safety checks in our code, and the compiler will watch over our shoulders.

const numbers

The easiest way to use the const qualifier is by flagging numbers that are set at the top of a block. For example:

int max_elements;

max_elements = nusers * ELEMENTS_PER_USER;

...

array[max_elements++] = n;
/* but you really meant array[max_elements] = n++; */

Adding a const qualifier means you can't accidentally modify max_elements.

const int max_elements = nusers * ELEMENTS_PER_USER;

const pointers

If a pointer is qualified as const, then its contents cannot be modified. This lets the compiler protect you from doing naughty things to yourself.

Here are two examples for functions you're familiar with:

int strlen( const char *str );
void memset( char *ptr, char value, int length );

In the case of strlen, the caller is guaranteed that any string passed in won't be modified. How terrible it would be if it was possible for strlen to modify what gets passed in!

The const on strlen's parameter also lets the compiler know that strlen can't be initializing what's passed in. For example:

char buffer[ MAX_LEN ];

int n = strlen( buffer );

The compiler knows that buffer hasn't been initialized, and that strlen can't be initializing it, so the call to strlen is on an uninitialized value.

Without the const, the compiler assumes that the contents of any pointer are getting initialized or modified.

const arrays

Consting arrays makes all the values in the array non-modifiable.

const int days_per_month[] =
    { 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 };

You don't want to be able to do days_per_month[1] = 4;, right? (We'll ignore that about 25% of the time you want days_per_month[1] to be 29.)

Mixing consts

Combining consts on a pointer and its contents can get confusing. It's important to know on which side of the asterisk that the const lies.

To the left of the asterisk, the characters are constant. To the right of the asterisk, the pointer is constant.

Note the difference between a pointer to constant characters:

/* Pointer to constant characters */
const char *str = "Don't change me.";
str++;      /* legal, now points at "o" */
*str = "x"; /* not legal */

and a constant pointer to characters:

/* Constant pointer to characters */
char * const str = buffer;
str++;      /* not legal */
*str = 'x'; /* buffer[0] is now 'x' */

Note the difference between which side of the asterisk that the const is on.

You can also combine the two, with a constant pointer to constant characters:

const char * const str = "Don't change me";

or even an array of constant pointers to constant characters:

const char * const days[] =
    { "Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat" };

If you see a declaration you don't understand, use cdecl. It's standard in many C compiler suites, and is freely available around the net.

$ cdecl
Type `help' or `?' for help
cdecl> explain const char * str;
declare str as pointer to const char
cdecl> explain char * const str;
declare str as const pointer to char

Decreasing the amount of repeated code

PMD (http://pmd.sourceforge.net/) has been used on C code, even though it's a Java tool. It looks for repeated strings of tokens that are candidates for either functions or macros.

PMD usage

General usage:

pmd [directory] [report format] [ruleset file]

To generate html output of unused code within parrot use:

pmd . html rulesets/unusedcode.xml > unused_code.html

Also distributed with PMD is the CPD (Copy/Paste Detector) which finds duplicate code. An easy way to get started with this tool is to use the gui (cpdgui). Set the root source directory to your parrot working directory, and choose the by extension... option of the Language: menu. Then put .c in the Extension: box and click Go.

Automated source macros

Perl5 has a lot of good source management techniques that we can use.

  • Macro for interp argument

    A macro for declaring the interpreter argument, and maybe a macro for passing it

    BTW, our Perl experience teaches us that somebody is going to want to make the interpreter a C++ object for Windows environments, and it wouldn't hurt to make that possible, or at least work in that direction, as long as clarity doesn't suffer.

  • Parrot_xxx macros

    Automated processing that would make a macro to let us write

    somefunc(interp,a,b,c)

    while the linkage is

    Parrot_somefunc(interp,a,b,c)

    for namespace cleanup. This is straight out of embed.fnc and proto.h in Perl5.

Automated generation of C headers

This has started significantly with the headerizer.pl program. Right now, it extracts the function headers correctly, but now I have to have it create the .h files.

Creating automated code checking tools

Documenting function behavior and structure members

Developing coverage tools

Automatically running the coverage tools

Run on many different C compilers

Most of Andy's work right now is with GCC 4.2 on Linux. We need many more.

Run under valgrind

Valgrind (http://valgrind.org/) is a profiler/debugger most notable for the way it magically monitors memory accesses and management.

To run parrot under Valgrind, the following argument set should be helpful:

valgrind --num-callers=500 \
   --leak-check=full --leak-resolution=high --show-reachable=yes \
   parrot --leak-test

(adapted from a post to parrot-porters by chromatic).

IMCC cleanup

From #parrot:

vsoni: there seems to be some dead code/feature....I had a chat
with leo and I am going to send and email to p6i for deprecation
of certain old features

Help other contributors hack their patches into Parrot-style industrial-strength C code.

From chip's comment at http://www.oreillynet.com/onlamp/blog/2006/07/calling_for_parrot_janitors.html

We've just had contributed an improved register allocation
implementation, but since the contributor is new to Parrot,
there are some style and coding standards issues that need to
be worked out. It'd be great if a Cage Cleaner could step up
and help our new contributor bang the code into Parrotish form.

Remove usage of deprecated features

DEPRECATED.pod lists features that are deprecated but not yet removed; An RT ticket will document how this deprecated feature is to be replaced. Help prepare for the actual removal of the feature by replacing its usage.

Clean up skipped tests

Parrot has too many skipped tests. Pick a test file with a skipped test, disable the skip() line, then make it pass. The Parrot code may not compile, or you may have to modify it to bring it up to date. The test may not even be useful anymore; we won't know until you try.

If you can make it pass, great!

If you can make it run, great! Make it a TODO test instead.

If neither, please report your findings so that everyone can decide what to do.

Handy configuration tips

Subversion configuration and automatic properties

There is an example .subversion/config file in editor/subversion_config which is useful in automatically setting the appropriate Subversion properties relevant to Parrot on addition of new files.

Displaying trailing whitespace in vim and emacs

Vim

Add this to your .vimrc:

set list
set listchars=trail:-,tab:\.\

NOTE: there is a space character after the last backslash. It is very important!

Contributed by Jerry Gay <jerry dot gay at gmail dot com>.

Emacs

Add this to your .emacs:

(setq-default show-trailing-whitespace t)

Emacs 22 users can highlight tabs like this:

(global-hi-lock-mode 1)
(highlight-regexp "\t")

Contributed by Eric Hanchrow <offby1 at blarg dot net>.

AUTHOR

Paul Cochrane a.k.a. ptc; original document by Andy Lester

SEE ALSO

docs/project/roles_responsibilities.pod, RESPONSIBLE_PARTIES Also look at docs/ROADMAP.pod and the list of Cage items on RT: http://xrl.us/owsd (Link to rt.perl.org)