Copyright (C) 1992, 1993, 1994, 1995 Free Software Foundation, Inc.
Permission is granted to make and distribute verbatim copies of this manual provided the copyright notice and this permission notice are preserved on all copies.
Permission is granted to copy and distribute modified versions of this manual under the conditions for verbatim copying, provided that the entire resulting derived work is distributed under the terms of a permission notice identical to this one.
Permission is granted to copy and distribute translations of this manual into another language, under the above conditions for modified versions, except that this permission notice may be stated in a translation approved by the Free Software Foundation.
This file contains guidelines and advice for someone who is the maintainer of a GNU program on behalf of the GNU project. Anyone can change GNU software, and you need not read this file just to do that. But if you want to maintain a version for widespread distribution, then this file becomes important for you.
Please send corrections or suggestions regarding this document to
gnu@prep.ai.mit.edu. If you make a suggestion, please include a
suggested new wording for it; our time is limited. We prefer a context
diff to the `maintain.texi' file, but if you don't have that file,
please mail your suggestion anyway.
This document uses the gender-neutral third-person pronouns "person", "per", "pers" and "perself." They are used just like "she", "her", "hers" and "herself", except that they have no implications about gender. For example, "Person placed per new program under the GNU GPL, to let the public benefit from per work, and to enable person to think well of perself." These pronouns were promoted, and perhaps invented, by Marge Piercy in Woman on the Edge of Time.
This release of the GNU Maintenance Instructions was last updated 24 March 1995.
When incorporating changes from other people, make sure to follow the correct procedures. Doing this ensures that the FSF has the legal right to distribute and defend GNU software.
For the sake of registering the copyright on later versions of the software, you need to keep track of each person who makes significant changes. A change of ten lines or so, or a few such changes, in a large program is not significant.
Before incorporating significant changes, make sure that the person has signed copyright papers and that the Free Software Foundation has received and signed them.
You can tell the person what papers to sign by email. For large changes, ask for an assignment. Send the person a copy of `/gd/gnuorg/assign.changes', but first, go to the second page and insert the person's name and the name of the program involved in place of `NAME OF PERSON' and `NAME OF PROGRAM'. Do this before sending, because otherwise the person might sign without noticing them. Then the papers would be useless.
For medium to small changes, ask the person to fill out and return the disclaimer in the file `/gd/gnuorg/disclaim.changes'.
If someone is likely to keep making changes, person might want to sign an assignment for all per future changes to the program. There's a template you can use for this in `/gd/gnuorg/assign.future'. First, person needs to ask per employer for a disclaimer-in-advance; use the template in `/gd/gnuorg/disclaim.future'.
To check whether papers have been received, look in
`/gd/gnuorg/copyright.list'. If you can't look there directly,
fsf-records@prep.ai.mit.edu can check for you, and can also
check for papers that are waiting to be entered and inform you
when expected papers arrive.
You can also send the person `/gd/gnuorg/conditions.text', which explains per options (assign vs. disclaim) and their consequences.
You should maintain copyright notices in all files of the program. A copyright notice looks like this:
Copyright 19xx, 19yy, 19zz copyright-holder
The copyright-holder is usually the Free Software Foundation, Inc., but may be someone else.
The list of year numbers should include each year in which you finished preparing a version which was actually released, and which was an ancestor of the current version.
It is important to understand that rule carefully, much as you would understand a complicated C statement in order to hand-simulate it.
This list is not a list of years in which versions were released. It is a list of years in which versions, later released, were completed. So if you finish a version on Dec 31, 1994 and release it on Jan 1, 1995, this version requires the inclusion of 1994, but doesn't require the inclusion of 1995.
The versions that matter, for purposes of this list, are versions that were ancestors of the current version. So if you made a temporary branch in maintenance, and worked on branches A and B in parallel, then each branch would have its own list of years, which is based on the versions released in that branch. A version in branch A need not be reflected in the list of years for branch B, and vice versa.
However, if you copy code from branch A into branch B, the years for branch A (or at least, for the parts that you copied into branch B) do need to appear in the list in branch B, because now they are ancestors of branch B.
Keep records of which portions were written by whom.
These records don't need to be as detailed as a change log. They don't need to distinguish work done at different times, only different people.
They should say which files or functions were written by each person, and which files or functions were revised by each person. They don't need to say what the purpose of the change was. The Register of Copyrights doesn't care what the program does.
For example, this would describe an early version of GAS:
Dean Elsner first version of all files except gdb-lines.c and m68k.c. Jay Fenlason entire files gdb-lines.c and m68k.c, most of app.c, extensive changes in messages.c, input-file.c, write.c, revisions elsewhere.
Please keep these records in a file named `AUTHORS' in the source directory for the program itself.
If someone sends you changes which are ugly and will make the program harder to understand and maintain in the future, such as a port to another operating system containing ad hoc conditionals, don't hesitate to ask person to clean up per changes before you merge them.
Since the amount of work we do is constant in any case, the more work we get other people to do, the faster GNU will advance.
If person will not or can not make the changes clean enough, then say that you can't afford to merge them. Invite per to distribute per changes another way, or to find other people who can make them clean enough for us to maintain.
The only reason to do these cleanups yourself is if (1) it is easy enough that it is less work than telling the author what to clean up, or (2) users will greatly appreciate the improvement, and you would almost write it yourself if you had time.
The GNU Coding Standards are a good thing to send people who have to clean up C programs (see section `Contents' in GNU Coding Standards). The Emacs Lisp manual contains an appendix that gives coding standards for Emacs Lisp programs; it is good to urge authors to read it (see section `Tips and Standards' in The GNU Emacs Lisp Reference Manual).
Once a program is in use, you will start getting bug reports. Some GNU
programs have their own special lists for sending bug reports. For
miscellaneous programs that don't have their own lists, we use a
catch-all list, bug-gnu-utils@prep.ai.mit.edu. Talk with
gnu@prep.ai.mit.edu to arrange to be added to the proper list
for your program, or to set up a new list.
When you receive bug reports, keep in mind that bug reports are crucial for your work. If you don't know about problems, you cannot fix them. So always thank the people who send bug reports.
The main purpose of bug reports is to enable you to improve the next version of the program. Helping individuals solve their problems is secondary--it is not part of the maintainer's job. So you don't need to give a substantial response unless that helps you maintain the program (for example, to ask for more information, or to ask the user to test a fix). Just "thanks" is sufficient to tell the user that the bug report was useful.
As a practical matter, any time you spend helping individuals beyond what is necessary for you, takes time away from maintaining the program. While this may seem "friendly" and "helpful," actually it is counterproductive. If you help one person when you could instead have helped thousands, the world is worse off.
It is very important to keep backup files of all source files of GNU. You can do this using RCS if you like. The easiest way to use RCS is via the Version Control library in Emacs; section `Concepts of Version Control' in The GNU Emacs Manual.
Alternatively, you can keep backup files.
Emacs makes a backup file by renaming the old source file to a new name. This means that if the file is also pointed to by a distribution directory, the distribution directory continues to point to the same version it always did--the right thing.
We want to keep more than one backup for all GNU sources. So, if you are going to edit GNU sources, make certain to put
(setq version-control t)
into your `.emacs' file, so that Emacs always creates numbered backup files.
Using Emacs backup files works as long as people always make changes
with Emacs. If you change the file in some other way, and use
cp, ftp, or tar to install it, you will
overwrite the old version and fail to make a backup. Don't do
that!
If you want to make a change to a source file with something other than Emacs, you can write the changed file to another name, and use C-x C-w in Emacs to write it under the real name. This makes the backup file properly.
You can use GNU cp or mv to install changed files if you
give them the `--backup' (or, equivalently, `-b') option; then
they make backup copies the same way that Emacs does. You should also
use the `--version-control=t' option, or, alternately, first
export VERSION_CONTROL=t
(or the csh equivalent); this makes GNU cp and mv
create numbered backup files instead of a single backup file with a
`~' appended to the filename. For installing many changed files,
you can use shell wildcards and also give GNU cp or mv the
`--update' (`-u') option, which only copies (or moves) files
that have been modified more recently than the existing destination
files, and the `--verbose' (`-v') option, which prints the
names of the files that are actually copied (or moved).
Before you use mv or cp in this way, make sure it is
the GNU version. Do `which mv' (in csh) or `type mv'
(in bash) to verify you are not getting `/bin/mv' (or
likewise for cp). Or just type `cp' or `mv' and look
at the usage message.
Always answer no when Emacs offers to delete backup files automatically. The way to delete them is by hand, using M-x dired.
When you decide which backup files to delete, it is good to keep one every couple of weeks or so, going back about 2 months.
If there is a long gap between versions, because that file did not change during a long period of time, then keep the version before the gap even if it is 2 months old. Pretend it is just 2 weeks older than the next kept version, so delete it when the next version is >= 6 weeks old.
If the changes in a program have been simple, then you don't need to keep as many backup files. Just one a month for 2 months is enough.
If you have made big changes, keep the versions before and after the big change, until they are old enough.
If you made several changes the same day, usually the last version written that day is best to keep.
It is almost always right to keep the most recent backup version.
For each program, you should keep a special magtape or cartridge as an
archive. Each time you release a new version, dd the tar file onto
the end of the tape. Keep a list of versions on the tape's paper
label, and add to it each time you add to the tape.
For cartridges, you can type
mt -f /dev/nrst8 eom
to go straight to the end of the data on the tape.
For reel-to-reel tapes, there is no automated way to go to the end of the data on the tape. You have to count the number of files (based on the written label), and space forward over them with `mt fsf'.
To be safe, it is important to check your count. If the count is n, then do:
mt -f /dev/nrmt8 fsf n-1
This puts you at the beginning of the last existing tar file.
Then do
tar tf /dev/nrmt8
to list that tar file. If the version number appears in a directory name, which is a good idea, you can use this to verify that you have reached the tar file you wanted to reach.
C-c the tar before it finishes; then do
mt -f /dev/nrmt8 fsf
to skip past it and its end-of-file marker.
To copy the new distribution file onto cartridge tape, do:
dd if=tar-file-name of=/dev/nrst8 bs=102400
(This specifies a blocking factor of 200.)
For reel-to-reel tape, do:
dd if=tar-file-name of=/dev/nrmt8 bs=10240
(This specifies a blocking factor of 20.)
When the tape gets full, put it aside permanently and start writing another.
It is important to follow the GNU conventions when making GNU software distributions.
The tar file for version m.n of program foo should be
named `foo-m.n.tar'. It should unpack into a
subdirectory named `foo-m.n'. Tar files should not
unpack into files in the current directory, because this is inconvenient
if the user happens to unpack into a directory with other files in it.
Here is how the `Makefile' for Bison creates the tar file. This method is good for other programs.
dist: bison.info
echo bison-`sed -e '/version_string/!d' \
-e 's/[^0-9.]*\([0-9.]*\).*/\1/' -e q version.c` > .fname
-rm -rf `cat .fname`
mkdir `cat .fname`
dst=`cat .fname`; for f in $(DISTFILES); do \
ln $(srcdir)/$$f $$dst/$$f || { echo copying $$f; \
cp -p $(srcdir)/$$f $$dst/$$f ; } \
done
tar --gzip -chf `cat .fname`.tar.gz `cat .fname`
-rm -rf `cat .fname` .fname
Source files that are symbolic links to other file systems cannot be
installed in the temporary directory using ln, so use cp
if ln fails.
If the program is large, it is useful to make a set of diffs for each release, against the previous important release.
At the front of the set of diffs, put a short explanation of which version this is for and which previous version it is relative to. Also explain what else people need to do to update the sources properly (for example, delete or rename certain files before installing the diffs).
The purpose of having diffs is that they are small. To keep them small, exclude files that the user can easily update. For example, exclude info files, DVI files, tags tables, output files of Bison or Flex. In Emacs diffs, we exclude compiled Lisp files, leaving it up to the installer to recompile the patched sources.
When you make the diffs, each version should be in a directory suitably named--for example, `gcc-2.3.2' and `gcc-2.3.3'. This way, it will be very clear from the diffs themselves which version is which.
If you use GNU diff to make the patch, use the options
`-rc2P'. That will put any new files into the output as "entirely
different."
If the distribution has subdirectories in it, then the diffs probably include some files in the subdirectories. To help users install such patches reliably, give them precise directions for how to run patch. For example, say this:
To apply these patches, cd to the main directory of the program and then use `patch -p1'. `-p1' avoids guesswork in choosing which subdirectory to find each file in.
It's wise to test your patch by applying it to a copy of the old version, and checking that the result exactly matches the new version.
prep
Only the latest version of any program needs to be on prep. Being an
archive of old versions is not the function of prep.
Diffs are another matter. Since they are much smaller than distribution files, it is good to keep the diffs around for quite a while.
When you release a greatly changed new major version of a program, you might want to do so as a beta test release.
Once a program gets to be widely used and people expect it to work solidly, it is a good idea to do pretest releases before each "real" release. This means that you make a tar file, but send it only to a group of volunteers that you have recruited. (Use a suitable GNU mailing list/newsgroup to recruit them.)
One thing that you should never do is to release a distribution which is considered a pretest or beta test but which contains the version number for the planned real release. Many people will look only at the version number (in the tar file name, in the directory name that it unpacks into, or wherever they can find it) to determine whether a tar file is the latest version. People might look at the test release in this way and mistake it for the real release. Therefore, always change the number when you make a new tar file.
If you are about to release version 4.6 but you want to do a pretest or beta test first, call it 4.5.90. If you need a second pretest, call it 4.5.91, and so on.