CVSps (c) 2001,2002,2003 David Mansfield
CVSps is a program for generating 'patchset' information from a CVS
repository. A patchset in this case is defined as a set of changes made
to a collection of files, and all committed at the same time (using a
single 'cvs commit' command). This information is valuable to seeing the
big picture of the evolution of a cvs project. While cvs tracks revision
information, it is often difficult to see what changes were committed
'atomically' to the repository.
CVSps is a relatively small program, with only a few modules.
The Makefile is very simple and should work in most GNU type environments.
Unfortunately, I've only been able to test on Red Hat Linux, so YMMV. As
CVSps matures, I'm sure a more sophisticated build environment will evolve
with it. For now, just try 'make' and 'make install'. If you have any
problems let me know.
Note: not all options are necessarily discussed here. Please check the
output of 'cvsps -h' and/or the manual page for the most up-to-date info.
CVSps operates by parsing the 'cvs log' output. So to run it, you must
be in the working directory of a cvs project. CVSps handles
subdirectories fine, so run it in the top directory of your project.
a) the ~/.cvsps/cvsps.cache file (or so it's called)
Because you may have a *lot* of revision history in your project, and/or
your connection to the cvs server may be slow or congested, CVSps uses a
cache file for most operations. The first time you run CVSps, just issue:
and will begin reading and parsing the cvs log. When it is finished it
will output all of the patchset information to stdout, and it will also
generate the '~/.cvsps/cvsps.cache' file. Note: for historical reasons
this file is still called the cvsps.cache file, but in reality it is named
based on the CVS/Root and CVS/Repository contents, and thus is shared for
the same repository checked out in multiple places.
If the cache file exists, it won't ever be automatically updated. To
update the cache with cvs activity that has occurred since the ~/.cvsps/cvsps.cache
was last updated, use:
If you question the integrity of the ~/.cvsps/cvsps.cache, or for some other reason
want to force a full cache rebuild, use (you could also 'rm' the cache file):
b) Reading the output.
CVSps's output is information about patchsets. A patchset looks like:
Date: 2002/07/11 19:50:46
[PATCH] Fix several pdc202xx problems
Misnaming of 20270 as 20268R
Failure of LBA48 on 20262
Incorrect speed detection because the old driver used host not drive side
PDC202xx handling for quirks in udma reporting off some drives
LBA48 for PIO mode
This patchset is taken from the linux kernel BK->CVS tree. It shows the date,
the author, log message and each file that was modified. For each file the
pre-commit and post-commit revisions are given. You can also see (if
applicable, not in this case) if the files are on a branch, as well as the
tag (see TAGS below).
Patchsets are ordered by commit timestamp, so as long as the clock on your
cvs server is monotonic, the numbering of patchsets should be invariant
across cache-rebuilds. (see COMPATIBILITY below).
c) Limiting the patchset output.
The default output of CVSps is to show all patchsets. This can be
filtered in one of many ways. These flags can be combined to really
limit the output to what you're interested in.
By id. With the -s <ps range> you can specify individual PatchSets by
number or by range. Ranges can be of the form '<number>', '<number>-',
'-<number>' and of course '<number>-<number>'. Multiple ranges can be
specified seperated by commas. E.g.
cvsps -s 999-1020,1025,4956-
By author. With the -a <author> flag you limit the output to patchsets
committed by a given author. The author is usually the UNIX login id.
By file. With the -f <file regex> flag you limit the output to patchsets
that have modified the given file. Because a regular expression can have
many pieces 'or'ed together, you can specify many different files here,
for example (note also the use of the ^ character):
cvsps -f '^net/ipv4|^net/core'
By date. With one date specification, CVSps shows only patchsets newer
than the date given, and with two dates, it shows patchsets between the
*NOTE ON DATE FORMAT*. Because I'm lazy, only one date format is
currently acceptable. 'YYYY/MM/DD HH:MM:SS' where time is given as
localtime, and HH is in 24 hour format. NOTE ALSO that cvs tends to
display times as GMT, but parse dates as localtime, so when using the '-D'
with cvs you need to convert from GMT to localtime in your head. CVSps is
not as fancy. It treats all dates as if in localtime, and therefore you
give CVSps dates the same way it gives them to you.
By branch. With the -b <branch> flag you limit the output to patchsets
that have modified the history of the given branch. Note, this
doesn't necessarily mean the commit itself was made on the branch, since
the files in question may have existed prior to the branch point, in which
case changes made to a given file before the branch point affect the file
as it exists in the head of the branch. If you want to restrict to the
main branch, use a branch of 'HEAD'.
By log comment. With the -l <regex> flag you can limit the ouptut to
patchsets with the commit message matching the regex.
By tag. With the -r <tag1> -r <tag2> you can limit the patchsets to
commits after a given tag1 and, optionally, before tag2.
d) viewing the changes made by a patchset.
To show the 'diff' output for a given patchset, use -g.
It will show you the diff of changes made by the selected commits.
Some effort was made to ensure that the patches are valid, even in the
case of removing or creating files, a case in which 'cvs diff' fails.
The patches generated are, generally speaking, applyable in the working
directory with the '-p1' option to the patch command.
e) what is timestamp fuzz factor (-z option)?
There's another annoying feature of cvs. When you commit a large change,
the timestamp on the change is created per file committed. For example:
if you commit changes to 60 files on a slow server, taking, say, 60
seconds, the 'commit time' as given in the log message for the first file
will differ from that of the last file by 60 seconds.
The fuzz factor attempts to workaround this by saying: commits by the same
author, using the same log message, within <fuzz> seconds are considered
part of the same patchset. The default fuzz is 300 seconds (5 minutes).
Please read the manual page.
One of the main goals of cvsps was to make the patchset numbering stable across
all time, as long as no funny-business is done to the repository files themselves.
Unfortunately, as bugs have been fixed, the numbering has changed. This is most
regrettable, but unavoidable.
Additionally, in version 2.0, two changes have been made which will 'renumber'
the patch sets.
1) The false 'globbing' of two commits from nearly the exact same time, by the
same person, with the same log description but to different branches. Now,
these will be reported as 2 patchsets instead of one.
2) The creation of a large volume of patchsets for 'file xyz was originally added on
branch' log messages. This occurs whenever a file is originally born on a branch,
and is exacerbated by the fact that even when all of these files are created with
a single commit, the 'file xyz...' messages, which contains the actual file name,
are different, causing a proliferation of these unwanted patchsets. These patchsets
are now silently eliminated from the output.
Reporting bugs / submitting patches.
Although the current version is perfect and bug free, you can still send
bug reports, feature requests and patches to me at:
I will try to maintain CVSps and make releases regularly. The most recent
version of CVSps will always be available at http://www.cobite.com/cvsps
Special thanks to my employer Cobite and Robert Lippman, who've given me
time to develop this tool.
Known Problems (this will become the FAQ if anyone ever A any Q).
1) What is the '*** file xyz doesn't match strip_path abc' error?
This error occurs when one of the subdirectories of the directory you
ran CVSps in is checked out from a different repository. CVSps tries
to remove the repository path information from the filenames that it
gets to give you working-directory local pathnames. It does this
at startup by parsing the CVS/Root and CVS/Repository files. If
the contents of these two files is different for some subdirectory,
all of the files in that subdirectory will be ignored.
You can always run CVSps in that subdirectory, and since it IS a
separate repository, that does make a little bit of sense.