============================================================================
= This file
* This file attempts to describe the conventions to use when hacking automake.
* After git checkout from Savannah, you can do an initial build with, e.g.:
./bootstrap && ./configure --prefix=/tmp/amdev && make
============================================================================
= Administrivia
* The correct response to most actual bugs is to write a new test case
which demonstrates the bug. Then fix the bug, rerun the test suite,
and check everything in. Run the test suite in parallel or it takes ages.
* If you incorporate a change from somebody on the net:
- First, if it is not a tiny change, you must make sure they have
signed the appropriate paperwork.
- Second, add their name and email address to THANKS.
* If a change fixes or adds a test, mention the test in the commit
message. If a change fixes a bug registered in the Automake debbugs
tracker, mention the bug number in the commit message, using
the short url form, like https://bugs.gnu.org/1234. See section below
about commit messages.
* If a person or people report a new bug, mention their name(s) in the
commit message that fixes or exposes the bug, and add a line for them
in THANKS.
* When documenting a non-trivial idiom or example in the manual, be
sure to add a test case for it, and to reference such test case from
a proper Texinfo comment.
* Some files in the automake package are not owned by automake (many
by gnulib, https://gnu.org/s/gnulib); these files are listed in the
$(FETCHFILES) variable in maintainer/maint.mk. They should never be
edited here. All but two of them can be updated from respective
upstreams with "make fetch" (this should be done especially before
releases). The only exceptions are help2man (install the current
version and copy in by hand) and 'lib/COPYING' (from FSF), which
should be updated by hand whenever the GPL gets updated (which
shouldn't happen that often anyway :-).
Conversely, automake holds the master copy of a number of files that
are copied into other projects, such as install-sh, mdate-sh,
Channels.pm and ChannelDefs.pm, and plenty more; grep for bug-automake
in lib/* for most of the scripts, and see <gnulib>/config/srclist.txt
and <autoconf>/build-aux/fetch.pl for lists of such files on two
notable receiving ends. Do your best not to break them.
* All changes from the last release that are not trivial bug fixes should
be mentioned in NEWS.
* Changes which are potentially controversial, require a non-trivial
plan, or must be implemented gradually with a roadmap spanning several
releases (either minor or major) should be discussed on the list,
and have a proper entry in the PLANS directory. This entry should be
always committed in the "maint" branch, even if the change it deals
with is only for the master branch, or a topic branch. Usually, in
addition to this, it is useful to open a "wishlist" report on the
Automake debbugs tracker, to keep the idea more visible, and have the
discussions surrounding it easily archived in a central place.
===========================================================================
= Setting the development environment
* In development, ./GNUmakefile is used, not (the generated) ./Makefile.
Run make V=1 to see the commands that are run.
* The required and optional dependencies used by Automake and its test suite
can be automatically fetched using the GNU Guix package manager with the
following command:
guix environment automake --ad-hoc \
gettext help2man texinfo libtool flex bison dejagnu zip icedtea \
python gcc-toolchain gfortran pkg-config vala
For other environments, you'll need to install the equivalent.
* To run the bin/automake or bin/aclocal scripts in a checkout, it is
necessary to set PERL5LIB, as in:
am=/path/to/automake/checkout
env PERL5LIB=$am/lib $am/bin/automake --help
env PERL5LIB=$am/lib $am/bin/aclocal --help
* If you need to test different Python versions, pyenv is probably the
most convenient way at this writing. https://github.com/pyenv
============================================================================
= Naming
* Automake's convention is that internal AC_SUBSTs and make variables
should be named with a leading 'am__', and internally generated targets
should be named with a leading 'am--'. This convention, although in
place from at least February 2001, isn't yet universally used.
But all new code should use it.
We used to use '_am_' as the prefix for an internal AC_SUBSTs.
However, it turns out that NEWS-OS 4.2R complains if a Makefile
variable begins with the underscore character. Yay for them.
We changed the target naming convention just to be safe.
* If you'd like to read some general background on automake and the
other GNU autotools, you might try this not-too-long web page (not
affiliated with GNU):
https://www.linux.com/news/best-practices-autotools/
============================================================================
= Editing '.am' files
* For variables, always use $(...) and not ${...}.
* Prefer ':' over 'true', mostly for consistency with existing code.
* Use '##' comments liberally. Comment anything even remotely unusual,
and even usual things, to help future readers of the code understand.
* Never use basename or dirname. Instead, use sed.
* Do not use 'cd' within backquotes, use '$(am__cd)' instead.
Otherwise the directory name may be printed, depending on CDPATH.
More generally, do not ever use plain 'cd' together with a relative
directory that does not start with a dot, or you might end up in one
computed with CDPATH.
* For install and uninstall rules, if a loop is required, it should be
silent. Then the body of the loop itself should print each "important"
command it runs. The printed commands should be preceded by a single
space.
* Ensure install rules do not create any installation directory where
nothing is to be actually installed. See automake bug#11030.
============================================================================
= Editing conventions
* Indent using GNU style. For historical reasons, the perl code
contains portions indented using Larry Wall's style (perl-mode's
default), and other portions using the GNU style (cperl-mode's
default). Write new code using GNU style.
* Don't use & for function calls, unless really required.
The use of & prevents prototypes from being checked; it's also
uncommon in modern Perl code.
(Perl prototypes are unlike function prototypes in other
languages, so understand what they do.)
* For automake.texi:
- After and during editing, run make info for error checking, and make pdf
to catch TeX-only errors.
- After adding a new node, you can update the @detailmenu with
M-x texinfo-master-menu (not tested with current Emacs releases, but
hopefully it still works).
- You'll probably have already added it to the higher-level menu under
which the new node was added, since that's necessary to run "make info".
============================================================================
= Automake versioning and compatibility scheme
* There are three kinds of automake releases:
- new major releases (e.g., 2.0, 5.0)
- new minor releases (e.g., 1.14, 2.1)
- micro a.k.a. "bug-fixing" releases (e.g., 1.13.2, 2.0.1, 3.5.17).
A new major release should have the major version number bumped, and
the minor and micro version numbers reset to zero. A new minor release
should have the major version number unchanged, the minor version number
bumped, and the micro version number reset to zero. Finally, a new
micro version should have the major and minor version numbers unchanged,
and the micro version number bumped by one.
For example, the first minor version after 1.13.2 will be 1.14; the
first bug-fixing version after 1.14 that will be 1.14.1; the first
new major version after all such releases will be 2.0; the first
bug-fixing version after 2.0 will be 2.0.1; and a further bug-fixing
version after 2.0.1 will be 2.0.2.
* Micro releases should be just bug-fixing releases; no new features
should be added, and ideally, only trivial bugs, recent regressions,
or documentation issues should be addressed by them. On the other
hand, it's OK to include testsuite work and even testsuite refactoring
in a micro version, since a regression there is not going to annoy or
inconvenience Automake users, but only the Automake developers.
* Minor releases can introduce new "safe" features, do non-trivial but
mostly safe code clean-ups, and even add new runtime warnings (rigorously
non-fatal). But they shouldn't include any backward incompatible change,
nor contain any potentially destabilizing refactoring or sweeping change,
nor introduce new features whose implementation might be liable to cause
bugs or regressions in existing code. However, it might be acceptable to
introduce very limited and localized backward-incompatibility, *only*
if that is necessary to fix non-trivial bugs, address serious performance
issues, or greatly enhance usability. But please, do this sparsely and
rarely!
* Major releases can introduce backward incompatibility (albeit such
incompatibility should be announced well in advance, and a smooth
transition plan prepared for them), and try more risky and daring
refactorings and code cleanups. Still, backward incompatibility is
extremely undesirable and should be avoided at all costs.
* For more information, refer to the extensive discussion associated
with automake bug#13578.
============================================================================
= Working with git
* To regenerate dependent files created by aclocal and automake,
use the 'bootstrap' script. It uses the code from the source
tree, so the resulting files (aclocal.m4 and Makefile.in) should
be the same as you would get if you install this version of
automake and use it to generate those files.
* To get a faithful and correct rebuild, run:
./bootstrap && ./config.status --recheck && make clean all
* Usually, it is best to run against the latest stable versions of
Autoconf, Libtool, etc., since that is what most users will
do. However, sometimes it may be necessary to use the development
versions due to bugs fixed, etc. Whatever is first in PATH wins.
(Some background: https://bugs.gnu.org/11347)
* The Automake git tree currently carries three basic branches:
'master', 'next' and 'maint'. In practice, for quite a few years now,
as of this writing in 2023, we have been using only the master branch,
due to lack of time and desire to deal with anything but fixing bugs.
The branch information in the items below and above should thus be
taken with a large dose of salt.
* The 'master' branch is where the development of the next release
takes place. It should be kept in a stable, almost-releasable state,
to simplify testing and deploying of new minor version. Note that
this is not a hard rule, and such "stability" is not expected to be
absolute (emergency releases are cut from the 'maint' branch anyway).
* When planning a release a dedicated branch should be created and after
the release is done, the release branch is to be merged both into the
'master' branch and the 'maint' branch.
* Besides merges from release branches, the 'maint' branch can contain
fixes for regressions, trivial bugs, or documentation issues, that
will be part of an emergency regression-fixing or security releases.
As a consequence it should always be kept in a releasable state and no
"active" development should be done whatsoever.
* The 'next' branch is reserved for the development of the next major
release. Experimenting a little is OK here, but don't let the branch
grow too unstable; if you need to do exploratory programming or
over-arching change, you should use a dedicated topic branch, and
only merge that back once it is reasonably stable.
* The 'master' branch should be kept regularly merged into the 'next'
branch. It is advisable to merge only after a set of related
commits have been applied, to avoid introducing too much noise in
the history.
* There may be a number of longer-lived feature branches for new
developments. They should be based off of a common ancestor of all
active branches to which the feature should or might be merged later.
* When merging, prefer 'git merge --log' over plain 'git merge', so that
a later 'git log' gives an indication of which actual patches were
merged even when they don't appear early in the list.
* The 'master', 'maint' and 'next' branches should not be rewound,
i.e., should always fast-forward, except maybe for privacy issues.
For feature branches, the announcement for the branch should
document the rewinding policy.
If a topic branch is expected to be rewound, it is good practice to put
it in the 'experimental/*' namespace; for example, a rewindable branch
dealing with Vala support could be named like "experimental/vala-work".
* If you need to trivially fix a change after a commit, e.g., a typo in
the NEWS entry, edit the file and then:
git commit --amend --no-edit NEWS
git commit --amend --date="$(date -R)" # update date of commit
* If you want to just completely lose your local changes:
git fetch && git reset --hard origin && git checkout && git submodule update
(the submodule stuff is for gnulib).
* The preferred way to create a patch to mail around:
git format-patch --stdout -1 >/tmp/some-patch.diff
which can then be applied with:
git am mboxfile
============================================================================
= Writing a good commit message
* Here is the general format that Automake's commit messages are expected
to follow. See the further points below for clarifications and minor
corrections.
topic: brief description (this is the "summary line").
<reference to relevant bugs, if any>
Here goes a more detailed explanation of why the commit is needed,
and a general overview of what it does, and how. This section
should almost always be provided, possibly only with the exception
of obvious fixes or very trivial changes.
And if the detailed explanation is quite long or detailed, you can
want to break it in more paragraphs.
Then you can add references to relevant mailing list discussions
(if any), with proper links. But don't take this as an excuse for
writing incomplete commit messages! The "distilled" conclusions
reached in such discussions should have been placed in the
paragraphs above.
Finally, here you can thank people that motivated or helped the
change. So, thanks to John Doe for bringing up the issue, and to
J. Random Hacker for providing suggestions and testing the patch.
<detailed list of touched files>
* To choose the topic label, please review the previous commits in the
git log or the ChangeLog file from a release tarball. There are no
hard-and-fast rules; the aim is a usable description. So script
names, Makefile targets, and more are all viable. Some examples:
aclocal automake build cosmetics doc maint python release test ylwrap
* The <detailed list of touched files> should usually be provided (but
for short or trivial changes), and should follow the GNU guidelines
for ChangeLog entries (described explicitly in the GNU Coding
Standards); it might be something of this sort:
* some/file (func1): Improved frobnication.
(func2): Adjusted accordingly.
* another/file (foo, bar): Likewise.
* t/foo.sh: New test.
* t/list-of-tests.mk (handwritten_TESTS): Add it.
* doc/automake.texi (Some Node): Document it.
* NEWS: Mention it.
* If your commit fixes an automake bug registered in the tracker (say
numbered 1234), you should put the following line after the summary
line:
Fixes automake bug https://bugs.gnu.org/1234.
* If your commit is just related to the given bug report, but does not
fix it, you might want to add a line like this instead:
Related to automake bug https://bugs.gnu.org/1234.
* When referring to older commits, use 'git describe' output as pointer.
But also try to identify the given commit by date and/or summary line
if possible. Examples:
Since yesterday's commit, v1.11-2019-g4d2bf42, ...
... removed in commit 'v1.11-1674-g02e9072' of 2012-01-01,
"dist: ditch support for lzma"...
* If the commit is a tiny change that is exempt from copyright paperwork, the
commit message should contain a separate line after the detailed list of
touched files like the following:
Copyright-paperwork-exempt: yes
* Generally write the commit message in present tense.
============================================================================
= Test suite
* See file 't/README' for more information.
* Use "make check" and "make maintainer-check" liberally. It is often
useful to set VERBOSE=1 for more reporting on what is being done.
* Export the 'keep_testdirs' environment variable to "yes" to keep
*.dir test directories for successful tests also.
* Use Perl coverage information to ensure your new code is thoroughly
tested by your new tests.
* To run the tests, you should install expect, shar, language compilers,
gettext macros. Anything you don't install won't be tested. The test
suite will report on tests skipped due to software not available.
* Run the test suite in parallel (e.g., "make -j12 check"), both so it
doesn't take forever and because that is what most users will do. You
can also parallelize the makes that run inside each test with, e.g.:
make check AM_TESTSUITE_MAKE="make -j$(( 1*$(nproc) + 1 ))"
If you like, try different levels of parallelization to see what
runs the fastest on your machine. We'll use -j12 in our examples;
adjust as needed.
* To summarize, here is a typical make check invocation:
make -j12 VERBOSE=1 check keep_testdirs=yes
* Then check t/test-suite.log for the overall results. The directory
t/TESTNAME.dir is where the work will be left, if the test fails.
* You can run a single test, with, e.g.,
make check TESTS='t/aclocal-acdir.sh'
where t/aclocal-acdir.sh can be any t/*.sh test, including a new one
you are writing. You may want to add --no-print-dir to silence GNU
make about the many cd commands, and/or env VERBOSE=1 to get more
information about what make is running.
* Sometimes you may want to see when each test starts running, not only
when they complete. This can be done by setting TESTS_ENVIRONMENT:
make -j12 VERBOSE=1 TESTS_ENVIRONMENT='echo RUNNING: "$$f";' check
* Run "make maintainer-check" before commit. Watch out for trailing spaces.
Probably useful to run it more verbosely:
make AM_V_GEN= AM_V_at= VERBOSE=1 maintainer-check
* After "make -j12 check" succeeds. Run "make -j12 distcheck" before
pushing a commit, since that exercises yet more of the code.
* To unsilence Automake's "pretty" output for debugging, see the
"Unsilencing Automake" node in the manual. In short: run make --debug=p.
For the Automake test suite in particular, add VERBOSE=1.
* To set up a new test, first write the test file in t/good-name.sh.
Choose a name that fits with existing ones, as best you can devise.
- You'll likely want to copy material from an existing test, which is
fine and good; depending on how much is copied, it may be useful to
mention the other test(s) you used.
- Nevertheless, start the copyright year list in the new file fresh,
with the current year.
- You can run the new test on its own with
make check TESTS='t/good-name.sh'
- During development of a new test, it can be useful to end it with
"exit 33" (or any random value) to force it to fail even when it would
otherwise succeed, so you can inspect the test.dir/test.log and other
test.dir/* files to be sure things worked as expected.
- At some point before releasing, add the test to the appropriate
variable in t/list-of-tests.mk, most likely the (alphabetical)
handwritten_TESTS.
* Some hints for debugging test failures or trying to understand the
(complex) test infrastructure.
- The t/ax/ dir contains most of the test infrastructure files.
- You may want to try a simple test just to exercise the setup;
t/amassign.sh is such a test. For a simple test that runs automake
twice (sometimes useful in finding concurrency failures), try
t/nodef.sh.
- To trace the (complex) test initialization code, change the set -e
to set -ex (or whatever you wish) in t/ax/test-init.sh.
Automake itself enables -x for the *.log and test-suite.log files,
in am_test_setup in t/ax/test-lib.sh. You may want to change to -vx.
- If you want to run the individual commands of a test one by one in a
shell, you have to reproduce the test environment. Start a subshell and:
PATH=$am/bin:$am/t/ax:$PATH # where $am is your automake source checkout
set +u # if needed; automake cannot handle this "nounset" option
. test-init.sh # beginning of test
Then each command you type will be executed in the test environment,
including -x being enabled. Each test prints the PATH value at the beginning.
- If you need to debug a test which fails under installcheck, you need
to invoke make with am_running_installcheck=yes, as is done by the
installcheck-testsuite target in t/local.mk.
- If you want to see the (complex shell) commands that make is
running, despite all of Automake's attempt at silence, run (GNU)
make --debug=p ... other make args ...
============================================================================
= Bug tracker
* Automake uses the debbugs instance at https://bugs.gnu.org. Email
from the tracker is sent to bug-automake@gnu.org, and vice versa.
Ditto automake-patches@gnu.org. (See
https://gnu.org/s/automake for all the mailing lists; if you're
working on Automake, please subscribe.)
* To see all open bugs resp. patches (and recently closed ones):
https://debbugs.gnu.org/cgi/pkgreport.cgi?pkg=automake
https://debbugs.gnu.org/cgi/pkgreport.cgi?pkg=automake-patches
* For a full search form, initialized to show only bugs tagged confirmed:
https://debbugs.gnu.org/cgi/pkgreport.cgi?package=automake;include=tags%3Aconfirmed
* We use the "confirmed" tag to mean bugs that have been reviewed, are
evidently valid, but no one is currently working or plans to work on
them. Thus they are good candidates for new volunteers to get involved.
* To download the so-called maintainer mbox for a bug, say 12345:
https://debbugs.gnu.org/cgi/bugreport.cgi?bug=12345;mbox=yes;mboxmaint=yes
Only the maintainer mbox consistently has the bug# prefix on Subject:
lines, so it is the most usable, as far as we know.
* To close a bug, you can mail (or bcc) to 12345-done@debbugs.gnu.org.
It's best to do this with a note saying that the bug is being closed
because the fix was pushed, or whatever the reason.
* To add tags "help" and "confirmed" to bug 12345, mail to
control@debbugs.gnu.org with a one-line body:
tags 12345 + help confirmed
* To merge bugs 12345 and 56789, mail to
control@debbugs.gnu.org with a one-line body:
merge 12345 56789
* In general, all bug operations are done by mail. Reference information:
https://debbugs.gnu.org/server-control.html # dev info
============================================================================
= Release procedure
* The steps outlined here are meant to be followed for all releases,
whether stable, beta, alpha, or other. Where differences are
expected, they will be explicitly mentioned.
* Fetch new versions of the files that are maintained elsewhere by
running "make fetch". help2man has to be updated separately. If
any files in the automake repository get updated, commit them and
rerun the testsuite.
* Ensure that the copyright notices of the distributed files are up to
date. The maintainer-only target "update-copyright" can help with this.
* Check NEWS; in particular, ensure that all the relevant differences
with the last release are included.
* Update the AC_INIT version number in configure.ac. Leading up to a
minor release (x.y), say 1.17, pretests should be numbered from
1.16.90. Even numbers (1.16.90, 1.16.92, ...) should be pretest
releases; odd numbers (1.16.91, 1.16.93, ...) should be only
development versions. Leading up to a micro release (x.y.z), say
1.17.1, we would have 1.17.0.91 for the initial post-release
development version, 1.17.0.92 for the first pretest, etc.
Before 1.17 (July 2024), we used suffixed letters for pretests, as is
traditional, but deficient sorting algorithms did not like that. The
code in lib/Automake/Version.pm, along with the regexp in
lib/Automake/Options.pm, has to be updated to understand any change in
version number schemes. The test t/pm/Version[.pl] comprehensively
checks valid and invalid version strings.
* If making a minor or major release (1.x), also update APIVERSION
in configure.ac. But not micro.
* Create an announcement message with "make announcement". Edit the
generated 'announcement' file appropriately, in particularly filling
in by hand any "TODO" left in there. It's useful to do this early,
because carefully writing the announcement can easily bring to light
things that still need to be worked on.
* Run these commands, in this order (as mentioned, adjust -j as desired):
make bootstrap
make -j12 check keep_testdirs=yes
make maintainer-check
make -j12 check-no-trailing-backslash-in-recipes
make -j12 check-cc-no-c-o
make -j12 distcheck # regular distcheck
make -j12 distcheck AM_TESTSUITE_MAKE="make -j12" # parallelize makes
You can run "git clean -fdx" before invoking the bootstrap, to ensure
a completely clean rebuild. However, it must be done carefully,
because that command will remove *all* the files that are not tracked
by git! Run git status first to ensure all removals are ok.
* Create additional info for the announcement with the announce-gen
script that is part of gnulib (it requires the release tarball):
pkg=automake
prever=1.16.92
newver=1.16.94
reltype=alpha # or beta or stable
host=`if test $reltype = stable; then echo ftp; else echo alpha; fi`
gpgkey=0x... # gpg --fingerprint
$gnulib/build-aux/announce-gen --release-type=$reltype \
--package-name=$pkg --previous-version=$prever \
--current-version=$newver --gpg-key-id=$gpgkey \
--url-directory=https://$host.gnu.org/gnu/$pkg
and merge this with the just-written announcement file.
* Run "make git-tag-release".
This will run the maintainer checks, verify that the local git
repository and working tree are clean and up-to-date, and create
a proper signed git tag for the release (based on the contents
of $(VERSION)).
* Run "make git-upload-release".
For pretest releases: "DEVEL_SNAPSHOT=1 make git-upload-release".
This will first verify that you are releasing from a tagged version
and that the local git repository and working tree are clean and
up-to-date, and will then run "make dist" to create the tarballs,
and invoke the 'gnupload' script sign and upload them to the correct
locations. In case you need to sign with a non-default key, you can
use "make GNUPLOADFLAGS='--user KEY' git-upload-release".
* For stable releases you also need to update the manuals on www.gnu.org.
- Check links:
make html
make checklinkx # various perl prerequisites, see contrib/checklinkx
or use:
<https://validator.w3.org/checklink>
- Generate manuals (with the help of the standard gendocs.sh script):
make web-manual
The ready-to-be-uploaded manuals (in several formats) will be left
in the 'doc/web-manual' directory.
- Commit the updated manuals to web CVS. This requires the "cvsu"
utility from the "cvsutils" package.
make web-manual-update
If your local username is different from your username at Savannah,
you'll have to override the 'CVS_USER' make variable accordingly;
for example:
make web-manual-update CVS_USER=slattarini
* Update version number in configure.ac to next alpha number.
Rerun ./bootstrap and commit.
* Don't forget to "git push" your changes so they appear in the public
git tree.
* Send the announcement generated in the earlier steps at least to
<autotools-announce@gnu.org> and <automake@gnu.org>. If the release
is a stable one, the announcement also goes to <info-gnu@gnu.org>;
if it is an alpha or beta release, announcement should be sent also
to <platform-testers@gnu.org>, to maximize the possibility of early
testing on the widest variety of systems. Finally, copy an abridged
version of the announcement into the NEWS feed at:
<https://savannah.gnu.org/projects/automake>.
Be sure to link a version to the complete announcement (from
the version you sent to the automake list, as archived at
<https://lists.gnu.org/archive/html/automake/>).
---------
Copyright (C) 2003-2025 Free Software Foundation, Inc.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2, or (at your option)
any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
Local Variables:
mode: text
End:
