path: root/HACKING

-*- mode: text; -*-
$QuaggaId: Format:%an, %ai, %h$ $

Contents:

* GUIDELINES FOR HACKING ON QUAGGA
* COMPILE-TIME CONDITIONAL CODE
* COMMIT MESSAGE
* HACKING THE BUILD SYSTEM
* RELEASE PROCEDURE
* TOOL VERSIONS
* SHARED LIBRARY VERSIONING
* GIT COMMIT SUBSMISSION
* PATCH SUBMISSION
* PATCH APPLICATION
* STABLE PLATFORMS AND DAEMONS
* IMPORT OR UPDATE VENDOR SPECIFIC ROUTING PROTOCOLS


GUIDELINES FOR HACKING ON QUAGGA

[this is a draft in progress]

GNU coding standards apply.  Indentation follows the result of
invoking GNU indent (as of 2.2.8a) with no arguments.  Note that this
uses tabs instead of spaces where possible for leading whitespace, and
assumes that tabs are every 8 columns.  Do not attempt to redefine the
location of tab stops.  Note also that some indentation does not
follow GNU style.  This is a historical accident, and we generally
only clean up whitespace when code is unmaintainable due to whitespace
issues, to minimise merging conflicts.

For GNU emacs, use indentation style "gnu".

For Vim, use the following lines (note that tabs are at 8, and that
softtabstop sets the indentation level):

set tabstop=8
set softtabstop=2
set shiftwidth=2
set noexpandtab

Be particularly careful not to break platforms/protocols that you
cannot test.

New code should have good comments, which explain why the code is correct.
Changes to existing code should in many cases upgrade the comments when
necessary for a reviewer to conclude that the change has no unintended
consequences.

Each file in the Git repository should have a git format-placeholder (like
an RCS Id keyword), somewhere very near the top, commented out appropriately
for the file type. The placeholder used for Quagga (replacing <dollar> with
$) is:

	$QuaggaId: <dollar>Format:%an, %ai, %h<dollar> $

See line 2 of HACKING for an example;

This placeholder string will be expanded out by the 'git archive' commands,
wihch is used to generate the tar archives for snapshots and releases.

Please document fully the proper use of a new function in the header file
in which it is declared.  And please consult existing headers for
documentation on how to use existing functions.  In particular, please consult
these header files:

  lib/log.h	logging levels and usage guidance
  [more to be added]

If changing an exported interface, please try to deprecate the interface in
an orderly manner. If at all possible, try to retain the old deprecated
interface as is, or functionally equivalent. Make a note of when the
interface was deprecated and guard the deprecated interface definitions in
the header file, ie:

/* Deprecated: 20050406 */
#if !defined(QUAGGA_NO_DEPRECATED_INTERFACES)
#warning "Using deprecated <libname> (interface(s)|function(s))"
...
#endif /* QUAGGA_NO_DEPRECATED_INTERFACES */

To ensure that the core Quagga sources do not use the deprecated interfaces
(you should update Quagga sources to use new interfaces, if applicable)
while allowing external sources to continue to build. Deprecated interfaces
should be excised in the next unstable cycle.

Note: If you wish, you can test for GCC and use a function
marked with the 'deprecated' attribute. However, you must provide the
#warning for other compilers.

If changing or removing a command definition, *ensure* that you properly
deprecate it - use the _DEPRECATED form of the appropriate DEFUN macro. This
is *critical*. Even if the command can no longer function, you *must* still
implement it as a do-nothing stub. Failure to follow this causes grief for
systems administrators. Deprecated commands should be excised in the next
unstable cycle. A list of deprecated commands should be collated for each
release.

See also below regarding SHARED LIBRARY VERSIONING.


COMPILE-TIME CONDITIONAL CODE

Please think very carefully before making code conditional at compile time,
as it increases maintenance burdens and user confusion. In particular,
please avoid gratuitious --enable-.... switches to the configure script -
typically code should be good enough to be in Quagga, or it shouldn't be
there at all.

When code must be compile-time conditional, try have the compiler make it
conditional rather than the C pre-processor - so that it will still be
checked by the compiler, even if disabled. I.e.  this:

    if (SOME_SYMBOL)
      frobnicate();

rather than:

  #ifdef SOME_SYMBOL
  frobnicate ();
  #endif /* SOME_SYMBOL */

Note that the former approach requires ensuring that SOME_SYMBOL will be
defined (watch your AC_DEFINEs).


COMMIT MESSAGES

The commit message MUST provide:

* A suitable one-line summary followed by a blank line as the very
  first line of the message, in the form:

  topic: high-level, one line summary

  Where topic would tend to be name of a subdirectory, and/or daemon, unless
  there's a more suitable topic (e.g. 'build').  This topic is used to
  organise change summaries in release announcements.

The remainder of the commit message - its "body" - should ideally try to
address the following areas, so as to help reviewers and future browsers of
the code-base understand why the change is correct (note also the code
comment requirements):

* The motivation for the change (does it fix a bug, if so which? 
  add a feature?)
* The general approach taken, and trade-offs versus any other approaches.
* Any testing undertaken or other information affecting the confidence
  that can be had in the change.
* Information to allow reviewers to be able to tell which specific changes
  to the code are intended (and hence be able to spot any accidental
  unintended changes).

The one-line summary must be limited to 54 characters, and all other
lines to 72 characters.

Commit message bodies in the Quagga project have typically taken the
following form:

* An optional introduction, describing the change generally.
* A short description of each specific change made, preferably:
  * file by file
    * function by function (use of "ditto", or globs is allowed)

Contributors are strongly encouraged to follow this form.

This itemised commit messages allows reviewers to have confidence that the
author has self-reviewed every line of the patch, as well as providing
reviewers a clear index of which changes are intended, and descriptions for
them (C-to-english descriptions are not desireable - some discretion is
useful).  For short patches, a per-function/file break-down may be
redundant.  For longer patches, such a break-down may be essential.  A
contrived example (where the general discussion is obviously somewhat
redundant, given the one-line summary):

zebra: Enhance frob FSM to detect loss of frob

Add a new DOWN state to the frob state machine to allow the barinator to
detect loss of frob.

* frob.h: (struct frob) Add DOWN state flag.
* frob.c: (frob_change) set/clear DOWN appropriately on state change.
* bar.c: (barinate) Check frob for DOWN state.

Please have a look at the git commit logs to get a feel for what the norms
are.

Note that the commit message format follows git norms, so that "git
log --oneline" will have useful output.

HACKING THE BUILD SYSTEM

If you change or add to the build system (configure.ac, any Makefile.am,
etc.), try to check that the following things still work:

	- make dist
	- resulting dist tarball builds
	- out-of-tree builds

The quagga.net site relies on make dist to work to generate snapshots. It
must work. Common problems are to forget to have some additional file
included in the dist, or to have a make rule refer to a source file without
using the srcdir variable.


RELEASE PROCEDURE

* Tag the apppropriate commit with a release tag (follow existing
  conventions).
  [This enables recreating the release, and is just good CM practice.]

* Create a fresh tar archive of the quagga.net repository, and do a test
  build:

    git-clone git:///code.quagga.net/quagga.git quagga
    git-archive --remote=git://code.quagga.net/quagga.git \
        --prefix=quagga-release/ master | tar -xf -
    cd quagga-release

    autoreconf -i
    ./configure
    make
    make dist

The tarball which 'make dist' creates is the tarball to be released! The
git-archive step ensures you're working with code corresponding to that in
the official repository, and also carries out keyword expansion. If any
errors occur, move tags as needed and start over from the fresh checkouts.
Do not append to tarballs, as this has produced non-standards-conforming
tarballs in the past.

See also: http://wiki.quagga.net/index.php/Main/Processes

[TODO: collation of a list of deprecated commands. Possibly can be scripted
to extract from vtysh/vtysh_cmd.c]


TOOL VERSIONS

Require versions of support tools are listed in INSTALL.quagga.txt.
Required versions should only be done with due deliberation, as it can
cause environments to no longer be able to compile quagga.


SHARED LIBRARY VERSIONING

[this section is at the moment just gdt's opinion]

Quagga builds several shared libaries (lib/libzebra, ospfd/libospf,
ospfclient/libsopfapiclient).  These may be used by external programs,
e.g. a new routing protocol that works with the zebra daemon, or
ospfapi clients.  The libtool info pages (node Versioning) explain
when major and minor version numbers should be changed.  These values
are set in Makefile.am near the definition of the library.  If you
make a change that requires changing the shared library version,
please update Makefile.am.

libospf exports far more than it should, and is needed by ospfapi
clients.  Only bump libospf for changes to functions for which it is
reasonable for a user of ospfapi to call, and please err on the side
of not bumping.

There is no support intended for installing part of zebra.  The core
library libzebra and the included daemons should always be built and
installed together.


GIT COMMIT SUBSMISSION

The preferred method for submitting changes is to provide git commits via a
publically-accessible git repository, which the maintainers can easily pull.

The commits should be in a branch based off the Quagga.net master - a
"feature branch".  Ideally there should be no commits to this branch other
than those in master, and those intended to be submitted.  However, merge
commits to this branch from the Quagga master are permitted, though strongly
discouraged - use another (potentially local and throw-away) branch to test
merge with the latest Quagga master.

Recommended practice is to keep different logical sets of changes on
separate branches - "topic" or "feature" branches.  This allows you to still
merge them together to one branch (potentially local and/or "throw-away")
for testing or use, while retaining smaller, independent branches that are
easier to merge.

All content guidelines in PATCH SUBMISSION apply.


PATCH SUBMISSION

* For complex changes, contributors are strongly encouraged to first start a
  design discussion on the quagga-dev list before starting any coding.

* Send a clean diff against the 'master' branch of the quagga.git
  repository, in unified diff format, preferably with the '-p' argument to
  show C function affected by any chunk, and with the -w and -b arguments to
  minimise changes. E.g:

    git diff -up mybranch..remotes/quagga.net/master

  It is preferable to use git format-patch, and even more preferred to
  publish a git repository (see GIT COMMIT SUBMISSION).

  If not using git format-patch, Include the commit message in the email.

* After a commit, code should have comments explaining to the reviewer
  why it is correct, without reference to history.  The commit message
  should explain why the change is correct.

* Include NEWS entries as appropriate.

* Include only one semantic change or group of changes per patch.

* Do not make gratuitous changes to whitespace. See the w and b arguments
  to diff.

* Changes should be arranged so that the least contraversial and most
  trivial are first, and the most complex or more contraversial are last. 
  This will maximise how many the Quagga maintainers can merge, even if some
  other commits need further work.

* Providing a unit-test is strongly encouraged. Doing so will make it
  much easier for maintainers to have confidence that they will be able
  to support your change.

* New code should be arranged so that it easy to verify and test. E.g. 
  stateful logic should be separated out from functional logic as much as
  possible: wherever possible, move complex logic out to smaller helper
  functions which access no state other than their arguments.

* State on which platforms and with what daemons the patch has been
  tested.  Understand that if the set of testing locations is small,
  and the patch might have unforeseen or hard to fix consequences that
  there may be a call for testers on quagga-dev, and that the patch
  may be blocked until test results appear.

  If there are no users for a platform on quagga-dev who are able and
  willing to verify -current occasionally, that platform may be
  dropped from the "should be checked" list.


PATCH APPLICATION

* Only apply patches that meet the submission guidelines.

* If the patch might break something, issue a call for testing on the
  mailinglist.

* Give an appropriate commit message (see above), and use the --author
  argument to git-commit, if required, to ensure proper attribution (you
  should still be listed as committer)

* Immediately after commiting, double-check (with git-log and/or gitk). If
  there's a small mistake you can easily fix it with 'git commit --amend ..'

* When merging a branch, always use an explicit merge commit. Giving --no-ff
  ensures a merge commit is created which documents "this human decided to
  merge this branch at this time".

STABLE PLATFORMS AND DAEMONS

The list of platforms that should be tested follow.  This is a list
derived from what quagga is thought to run on and for which
maintainers can test or there are people on quagga-dev who are able
and willing to verify that -current does or does not work correctly.

  BSD (Free, Net or Open, any platform) # without capabilities
  GNU/Linux (any distribution, i386)
  Solaris (strict alignment, any platform)
  [future: NetBSD/sparc64]

The list of daemons that are thought to be stable and that should be
tested are:

  zebra
  bgpd
  ripd
  ospfd
  ripngd

Daemons which are in a testing phase are

  ospf6d
  isisd
  watchquagga


IMPORT OR UPDATE VENDOR SPECIFIC ROUTING PROTOCOLS

The source code of Quagga is based on two vendors:

   zebra_org (http://www.zebra.org/)
   isisd_sf (http://isisd.sf.net/)

To import code from further sources, e.g. for archival purposes without
necessarily having to review and/or fix some changeset, create a branch from
'master':

	git checkout -b archive/foo master
	<apply changes>
	git commit -a "Joe Bar <joe@example.com>"
	git push quagga archive/foo

presuming 'quagga' corresponds to a file in your .git/remotes with
configuration for the appropriate Quagga.net repository.