We welcome your contributions to the Performance Co-Pilot project!

Our preferred method of exchanging code, documentation, tests and new
ideas is via git.  To participate, create a cloned git tree using a
public git hosting service (e.g. github) and send mail to the list
with a description of the code and its location.

Patches are fine too - send 'em through to the list at any time - even
just ideas, pseudo-code, etc - we've found it is often a good idea to
seek early feedback, particularly when modelling the metrics you'll be
exporting from any new PMDAs you might be writing.


Philosophy
==========

PCP development has a long tradition of focussing on automated testing.

New tests should be added to the testsuite with new code.  This is not
a hard requirement when sending out patches or git commits, however,
particularly as a new contributor.  Someone else will work with you to
introduce an appropriate level of testing to exercise your fix or new
code, and that will end up being committed along with your new changes.

For the more practiced, regular PCP committers: testing is assumed to
have been done for all commits - both regression testing, using all of
the existing tests, and addition of new tests to exercise the commits.
A guide to using the automated quality assurance test suite exists in
the qa/README file.

For the very practiced, life members of the team: please help out in
terms of getting new contributors moving with their patches - writing
tests with them, giving feedback, and merging their code quickly.  We
aim to have new PMDA or monitoring tool contributions included in the
release immediately following the first arrival of code.  Also ensure
attribution of other contributors code is handled correctly using the
git-commit --author option.

Above all, have fun!


Conventions
===========

Add permanent diagnostics to any code of middling-to-high complexity.
The convention is to test the pmDebug variable (from libpcp) which is
supported by every tool via the -D/--debug option - see pmdbg(1) for
details, and see also the many examples of its use in existing code.

Use the same coding style used in the code surrounding your changes.

All the existing code is GPL'd, so feel free to borrow and adapt code
that is similar to the function you want to implement.  This maximizes
your chance of absorbing the "style" and "conventions" and producing
code that others will be able to easily read, understand and refine.


Practicalities
==============

Refer to the qa/README file for details on using the testsuite locally.
When writing new tests, bear in mind that other people will be running
those tests *alot*.  An ideal test is one that:

...is small - several small tests trump one big test
   (allowing someone else debugging a test failure to quickly get their
    head around what the test does)

...runs in as short an amount of time as possible
   (allowing someone debugging a failure to iterate quickly through any
    ideas they have while working on the fix)

...uses the same style as all the other tests in the "qa" directory
   (other people will be debugging tests you write and you may need to
    debug other peoples tests as well - sticking to the same script
    style and conventions is universally helpful)

...uses filtering intelligently ... because tests pass or fail based
   on textual comparison of observed and expected output, it is critical
   to filter out the fluff (remove it or make it determinisitic) and
   retain the essence of what the test is exercising.  See common.filter
   for some useful filtering functions, and watch out for things like:
   o dates and times (unless using archives)
   o use -z when using archives so the results are not dependent on
     the timezone where the test is run
   o the pid of the running test, especially when part of $tmp that is
     used extensively
   o collating sequences and locale ... take explicit control if in doubt
     

...is portable, testing as many platforms as possible, for example:
   o  use pmdasample rather than a pmda<kernel> metric;
   o  use archives for exercising exact behaviour of monitoring tools;
   o  watch out for differences in output between system utilities;
   o  before running, test for situations where it cannot run - such
      as unsupported platforms/configurations - use _notrun() here.