Updated platform lists in MACHINES.

[nmh] / docs / README.developers

# README.developers
#

This file is intended to provide a few tips for anyone doing development on nmh.
Developers who learn things "the hard way" about the nmh codebase (as opposed to
local info best encoded in a comment) are encouraged to share their wisdom here.

Following a commit checklist, the topics are organized alphabetically.


----------------
commit checklist
----------------

1. code updated?
2. test added?
3. make distcheck passed?
4. man page and other documentation updated?
5. docs/pending-release-notes updated?
6. should commit message reference bug report?
7. be sure that commit message starts with one-line synopsis, then a blank line
8. update/close bug report (with commit id)?
9. notify nmh-users?


---------------------------------
C library/system call usage notes
---------------------------------

* Use m_mktemp2() or m_mktemp() instead of mkstemp(3) (see section on
  nmh temporary files below).
* Use m_unlink() instead of unlink(3).
* Use done() instead of _exit(3) except after a fork(3).


-------------------------
autoconf & automake files
-------------------------

If you wish to change the `configure' script, the generated Makefile
or other related files, you'll need to first install GNU m4, available
from <ftp://ftp.gnu.org/pub/gnu/m4/>, then GNU autoconf
(<ftp://ftp.gnu.org/pub/gnu/autoconf/>) and GNU automake
(<ftp://ftp.gnu.org/pub/gnu/automake/>).  Nmh is currently using a
minimum of autoconf 2.68 and automake 1.12.

Most of the configure-related files are automatically generated.
The only files you should need to manually edit are configure.ac
and any autoconf macros in the m4 directory.  Don't, for instance,
edit config.h.in.  Though it is an input file from the point of
view of the users (and the configure script) it is an output file
from the point of view of the developers (and the autoconf script).

If you wish to add a new autoconf macro, it should be placed in it's
own file and put in the m4 directory; aclocal will automatically pick
it up and automake will add it to the distribution target automatically.

If you wish to make changes to the Makefile, you will need to edit
Makefile.am.  See the automake documentation if you need further help.
You should always check changes to Makefile.am by using "make distcheck".

Note that the automatically generated autotools files (such as config.h.in,
Makefile.in, and configure), are NOT kept in git.  Thus, when you check out
a git tree, you need to run the autogen.sh script before you can build
anything:

        % ./autogen.sh


-------------------
directory structure
-------------------

Following is a list of nmh's directories along with a brief description of the
purpose of each one.  Meanings are given for the abbreviations, but note that
these meanings are just informed guesses as to what the MH developers were
thinking.

./
    The top-level directory.  Contains files like README and INSTALL.

config/
    Contains utility files for the `configure' process.  Ordinarily nothing in
    here needs to be messed with.

docs/
    Contains more specialized documentation, such as this file and
    the FAQ.

etc/
    Contains files, file templates, and scripts to generate files that will be
    installed in the ${prefix}/etc directory.  Stuff like replcomps.

h/
    Most of nmh's header (.h) files are kept not in the individual source
    directories, but in this central location.

man/
    Contains all the input files that are processed to generate nmh's manual
    pages.

mts/
    "mts" stands for "Message Transfer Service".  Source files specific to the
    different MTSs go in the subdirectories.

mts/smtp/
    When nmh is configured to just talk to an SMTP server over TCP/IP, the
    source in this directory is compiled.

sbr/
    "sbr" stands for "subroutine(s)".  For the most part, each source file in
    this directory contains a single function with the same name as the source
    file.  These functions are of general use and are called from throughout
    nmh.

SPECS/
    Contains files such as RPM specs.

test/
    The num unit test suite.

tools/
    "tools" contains tools, scripts, and supporting files used by the
    developers while writing, debugging, and testing the code.

uip/
    "uip" stands for "User Interface Programs".  Most nmh commands have a file
    in this directory named <command>.c containing the code for that command
    (e.g. repl.c).  In some cases there is also an auxiliary file called
    <command>sbr.c which contains additional subroutines called from <command>.c
    (which would contain not much else besides main()).


---
git
---

As of December 2010, nmh has switched to using git for revision control
instead of CVS.  While the topic of git is beyond the scope of this FAQ,
to get started with git & nmh, you can run the following command to checkout
the nmh repository (with read-only access to it):

    % git clone git://git.savannah.nongnu.org/nmh.git

That will create a workspace called nmh.  To update that workspace
with changes to the master, cd to it and run:

    % git pull

If you are a project member and want write access to the repository,
you'll have to checkout with the following command instead of the one
above:

    % git clone <username>@git.sv.nongnu.org:/srv/git/nmh.git

We suggest using git pull --rebase instead of the default merge for
git pull.  If you don't want to add the --rebase option every time,
you can tell git pull to always rebase in your nmh workspace by
cd'ing to it and running the following command:

    % git config --bool branch.master.rebase true

And you'll probably want the following, also, so that --rebase applies
to any new branches that you create:

    % git config branch.autosetuprebase always


-------------------------------------------------------
nmh-local functions to use in preference to OS versions
-------------------------------------------------------

For some system functions whose availability or behavior varies from OS to OS,
nmh conditionally uses a local definition with the same name as the OS function
(e.g. snprintf()).  For other functions, developers need to avoid the OS
versions and always use the nmh-supplied function.  Here is a list of such
functions:

OS function  nmh-local version to use instead
===========  ================================
getpass()    nmh_getpass()


-------------------
nmh temporary files
-------------------

To create a temporary file, use m_mktemp2() or m_mktemp().  They use
mkstemp(3), but they also register the temporary file for removal on
program termination.  So, do not use mkstemp() directly.

To further support this, nmh_init() must be called at the beginning of
main().  And, if a child process is not going to immediately call one
of the exec(3) functions or _exit(3) after a fork(3), it should call
unregister_for_removal(0).  Finally, nmh_init() sets up signal handlers
for several signals:  these signal handlers should not be disabled.


--------------
nmh test suite
--------------

The nmh test suite is run through the Makefile, with "make check"
or "make distcheck".

In the nmh test suite, nmh programs to be tested should be invoked
through the run_test or run_prog shell functions defined in
test/common.sh.

Instead of echoing test progress, use start_test()/finish_test()
from tests/common.sh.  These will report the particular test name,
within the test, only if there is a failure.

To enable the use of valgrind, where available, set the environment
variable NMH_VALGRIND to a non-null value.  However, a separate
environment variable, VALGRIND_ME, triggers the use of valgrind in
test/inc/test-eom-align because it greatly extends the duration of
that test.

If valgrind complains about "serious error when reading debuginfo"
from a library, either update or remove the debuginfo package for
the offending library.


-------------
releasing nmh
-------------

To make a public release of nmh (we'll use version 1.5 as the example
here; the convention for release candidates is to use something like
"1.5-RC1"):

 1. Create a release branch.  The convention is to name release branches
    with the name "<version>-release".

    % git branch 1.5-release

    Note you are still on the master branch at this point.  Mark the
    current revision as the branchpoint for the new release branch:

    % git tag -a -m "This tag marks the point where we started the branch for 1.5" 1.5-branchpoint

    Now mark the master branch with a post-release version number (the
    convention here is to use VERSION+dev as the version number).

    % echo 1.5+dev > VERSION
    % git commit VERSION
    % git push
    % git push --tags

    Then do:

    % git checkout 1.5-release

    You are now on the 1.5 release branch.

 2. % echo 1.5 > VERSION
    % date +"%e %B %Y" > DATE
    (DATE should contain something like "30 December 2000")

 3. % git commit VERSION DATE; git push

 4. % git tag -a 1.5 -m 'Releasing nmh-1.5.'
    % git push --tags

    Note that the new convention for tagging is to simply tag with the
    version number (tag formats in the past have varied).

 5. % make distcheck

    If you want to check the distribution build with some particular
    configure options, set the DISTCHECK_CONFIGURE_FLAGS variable.
    E.g.:

    % make distcheck DISTCHECK_CONFIGURE_FLAGS=--with-cyrus-sasl

 6. Upload the distribution file to savannah.  You can automate this process
    by doing:

    % make upload SAVANNAH_USERNAME=username

    This will automatically call gpg to sign the release.  You can bypass
    this step by setting the SKIP_GPG_SIG variable.

 7. Update the http://www.nongnu.org/nmh/ homepage. (It lives in the CVS
    'webpages repository'; see https://savannah.nongnu.org/cvs/?group=nmh)

 8. Add a news item to the savannah nmh page. You'll have to submit it first
    and then separately approve it (under News->Manage).

 9. Send the release announcement email to the following places:
     nmh-workers@nongnu.org
     nmh-announce@nongnu.org
     exmh-users@redhat.com
     exmh-workers@redhat.com
     mh-e-users@lists.sourceforge.net

    If the release fixes significant security holes, also send an announcement
    to bugtraq@securityfocus.com.  The exmh lists require you to be subscribed
    in order to post.  Note that you don't need to post separately to
    comp.mail.mh, as the mh-users mailing list is apparently bidirectionally
    gatewayed to it.

    Preferably, the announcement should contain the MD5 hash generated above,
    and should be PGP-signed.  It should include the URL for the tarball as
    well as the URL of the website.  It should contain a brief summary of
    visible changes, as well as the URL of the git diff page that would show
    a detailed list of changes.  The changes between 1.5 and 1.4 would be
    shown by [this is just a guess, I don't know anything about cgit, and
    it assumes that we tag with nmh-x_x-release from now on]:

        http://git.savannah.gnu.org/cgit/nmh.git/diff/?h=nmh-1_5-release?h=nmh-1_4-release


---------------
after a release
---------------

Keep an eye on Debian's packaging, especially what patches they have to
apply, and the results of their Lintian checker, which even includes
spelling errors in man pages and binaries.

    https://sources.debian.net/src/nmh/1.6-16/debian/patches/
    https://lintian.debian.org/full/az@debian.org.html#nmh

Perhaps some nmh developer that uses Debian, or Ubuntu?, could provide
package-building commands, including lintian(1), for Makefile.am so
Lintian's complaints are known before release.