table of contents
|PAT(1)||General Commands Manual||PAT(1)|
pat - patch generator tools
pat [ -ahmnV ] [ filelist ]
patcil [ -abfhnpqsV ] [ filelist ]
patdiff [ -ahnV ] [ filelist ]
patbase [ -ahV ] [ filelist ]
patlog [ -hnruV ]
patmake [ -hV ]
patsend [ -hiquV ] [ patchlist ] [ recipients ]
patnotify [ -hquV ] [ recipients ]
patpost [ -hrV ] patchlist newsgroups
patftp [ -hV ] [ patchlist ]
patname [ -ahnmV ] -v version [ filelist ]
patsnap [ -ahV ] [ -o snapshot ] [ filelist ]
patcol [ -achnmsCV ] [ -d directory ] [ -f mani ] [ -S snap ] [ filelist ]
patclean [ -ahnmV ] [ filelist ]
Pat and its associated programs generate patches for any package that has been stored under RCS. These programs hide many of the details of RCS that get in your way when constructing and maintaining a package. All you need to do to create a new patch is to edit your files, run pat, and furnish some descriptions to RCS and in the generated patch file. Details such as how to initialize a new RCS file, what the comment string should be, how create a new branch, how to deal with subdirectories, how to do diffs and how to organize the patch file are handled automatically.
Before using any of the pat programs you must initialize your package by running packinit in the top-level directory of your package. This produces a .package file that all of the dist programs make use of.
In any of the programs that want a filelist, if you specify -a instead, all files in MANIFEST.new will be processed. In any of the programs that want a patchlist, a null patchlist means the current patch. You may use hyphens, commas and spaces to delimit patch numbers. If the right side of a hyphen is the null string, the current patchlevel is assumed as the maximum value. All the programs invoked with -h will print a small usage message with the meaning of each available options. The -V option gives the current version number.
Pat itself is a wrapper program that calls patcil, patdiff, and patmake. Usually you can just invoke pat and ignore all the others. Pat will update the MANIFEST file, if necessary (it will be an exact copy of the MANIFEST.new file, provided that a MANIFEST already existed), eventually calling patcil on it.
If you specify -n instead of a filelist, pat will find all files that are newer than patchlevel.h, put you into an editor to trim down the list, then use that file list. If every file of the list is removed, pat will be aborted.
Patcil is used to do a ci -l on any listed files. (It is assumed that you always want to keep your files checked out.) In addition to the -a switch, there is a -b switch which does a quick checkin of a set of files. Instead of calling rcs on each file, it calls rcs on the whole list of files. This is useful for checking in a trunk revision. When you are checking in a new trunk revision you might also want to use the -s flag which will strip out old RCS Log entries from the previous revision so that you can start over fresh. You probably should also use a -f which is passed through to the ci to force unchanged files to be checked in. To check in a new trunk revision, I say
patcil -s -f -a
Patcil will ask for the log entry instead of letting ci do it, and has a little prompter built in that lets you manipulate the message in various ways. Type h for a listing of what you can do. One of the nicest things is that you can pop up into an editor, optionally with a diff listing of the changes since the last patch, in case you've forgotten what you changed. If you type a CR as the first thing, it includes the previous log message. Exit the prompter with a CR.
There are two different ways to use patcil. You can either call patcil yourself, or let pat call it for you. It doesn't matter how many times you call patcil before running pat, since patdiff knows what the last patch base is to compare with. Patcil can be called in any of your directories; the other programs must be called in your top-level directory (or in bugs, when meaningful).
When you are creating a new file at a given patchlevel, you must patcil it with the -p option. Otherwise, it will simply be checked-in as a new trunk revision. The name of the file will be added to the MANIFEST.new if it does not already appear in it. If the name is found along with a description, that description will be passed through to rcs to properly initialize the RCS file.
Patbase can be used to reset the patch base to the current version when you've scrapped the previous patch sequence and are making a new distribution kit. What it really does is an rcs -Nlastpat:REV, where REV is the current revision. If patdiff blows up and you want to set the patch base back to some previous version, you have to call rcs -Nlastpat:REV yourself.
Patdiff actually does the diffs that go into the patch, comparing whatever version -Nlastpat points to with the most recently checked in version. It then updates -Nlastpat to point to the current version. It leaves the diff sitting in the bugs subdirectory for patmake to pick up. It can either use rcsdiff, or a diff command of your choice specified when you run packinit, in case your diff is better than rcsdiff.
Patlog is invoked by patmake usually, to update the ChangeLog file (or whatever name that file has been given when you ran packinit). It will gather log messages and launch an editor for you to make the necessary updates. If you have configured your package to also include RCS logs in the ChangeLog, another editor session will be launched for those too. Finally, a final log is built as a candidate entry for ChangeLog, which you may also modify as you wish.
When you don't have configured a ChangeLog file, patlog will only gather the information it needs to pass on to patmake and will exit. If you wish to call it yourself, you must do that after a least one sucessfull patdiff run. I recommend using the -n option the first time, and then use the -u option along with -n on subsequent runs to recreate files only when needed. The -r option (which supersedes -u) prevents patlog from recreating an existing file, even if it is out of date.
Patlog will call patcil and patdiff on your ChangeLog file (after having stuffed the candidate log entry you edited at the top of the file), unless prevented to do so by the -n option. This means the issued patch will update ChangeLog with current patch information, as you would expect it.
Patmake combines all the pieces of the patch into one file and invokes an editor so you can add the subject and description. It throws all your log messages in as Subjects and as Description, under the assumption that it's easier to delete what you don't want than to remember everything you did. You'll also want to expand each item in the Description so they don't just repeat the Subject lines. If you have a ChangeLog file, this must have been done already, or your ChangeLog will not accurately represent what is described in the patch, given that it has already been updated when patmake puts together all the pieces (see the note about patlog above).
Big patches will be split in order to keep size of each patch to a reasonable size. This is handled automatically, so you don't have to bother with it. The priority of each patch is merely intuited by patmake, given the assumption that small changes have a great priority.
Patsend, patpost and patftp are used to distribute your patches to the world. Patsend mails a set of patches to a set of recipients. The -u switch adds all the currently registered users who have asked for patches to be mailed to them, as well as the recipients specified while running packinit. The -i switch includes information with the patch about how the user may deregister themselves so they do not receive future patches automatically; this is also the default when the -u switch is used. Patpost posts a set of patches to a set of newsgroups. Patftp merely copies the patch into your public ftp directory.
Patnotify simply notifies users that a new patch has been released so that they can retrieve it by themselves from an archive site or via e-mail if they are interested. The -u switch can be used to include all the currently registered users who have asked for such a notification. The message includes the patch priority and description, as well as instructions on how to automatically request the patch (which will work only if you have mailagent installed).
Both patsend and patnotify let you edit the address list before actually sending anything, unless you add the -q option.
Patname can be used to tag a set of files with a symbolic name (specified with -v). This will set the name for the most recent revision of each file.
Patsnap will get a snapshot of your release by creating a SNAPSHOT file (name can be changed via -o) listing the file names and the latest RCS revision number for that file. Such snapshots can be used to identify the release at some random patchlevel and then later be able to retrieve it by feeding the snapshot file to patcol.
Patcol will check out a locked version of a file, eventually in an alternate directory (specified with -d, thus mirroring the distribution tree). All the files which have no RCS counterpart (e.g. patchlevel.h) will be simply copied by patcol. This is used by makedist to fake the distribution before making the kits. By default, patcol will not do the copyright expansion processing, but clients like makedist force it by using its -C option. Alternatively, you may force copying of the checked-out version into a directory by using the -c switch in conjunction with -d (or that former switch is simply ignored).
Patcol can also take its file list from a SNAPSHOT file via the -S switch, in which case it will check out the files using the RCS version specified by the snapshot file, such as one created by patsnap. You may instead specify -a, -m or -n to respectively use all the files in MANIFEST.new, all the modified files (the one which have been patciled), or all the files newer than patchlevel.h.
Patclean will remove the working files after having checked in all the changes. You may restores your working files by using patcol.
Patindex may be used from the top level directory or within the bugs directory. It will list all the patches and their Subject: lines. This program knows about compressed patches and will decompress them while producing the listing.
This section describes the RCS layer, in case something in the tools breaks, so that you may fix your RCS files and restart the operation.
All the patch tools get the main RCS trunk revision number out of your .package files, say it's 2.5. Then, at the time you ran packinit, you have chosen a branch for patches, usually number 1, which means all your modifications will be stored on the 2.5.1 RCS branch. The tools will create the branch for you when the time comes.
Each last released revision is tagged with an RCS lastpat symbol. When the patch is built by patdiff, the lattest version on the 2.5.1 branch is compared with the one tagged as lastpat. This is why you may safely issue more than one patcil beffore issuing the patch and still have it all worked out. Of course patdiff will move the lastpat tag to the lattest branch revision after processing a given file.
All the log messages and the modified files are kept in your bugs directory, in hidden files (name starting with a dot). Those logs will be collected when the patch is issued and the modified files are used by pat's -m switch.
Patdiff collects its patch hunks under the bugs directory, in files terminating with a .nn extension, where nn represents the current patch level + 1. (Which is going to be the next patchlevel when the patch will be made by patmake, unless it is too big to fit in one file).
Patlog prepares a set of files for patmake: the .clog file collects the information that will go under the Description: section within the patch, and .xlog ones collect the ChangeLog candidate entry. Finally, .rlog files store the RCS information that is to be included in the ChangeLog, if requested. Note that the topmost three lines are garbage and are ignored by all the tools handling those files.
In order to start up a new baseline (i.e. to change the RCS trunk revision number), you need to rerun packinit and change that number. Then issue a new patcil, probably with the -s, -a and -f options...
- Diffs for each file, gathered by patmake to create a patch
- Issued patches (can be compressed with compress only)
- Description to be filled into the patch (or the first part if the patch is split into several parts).
- Log messages for that patch
- Files modified in that patch (checked in with patcil)
- The priority of the next patch, computed by patlog for patmake's perusal.
- The RCS logs computed by patlog.
- The Subject: lines for the next patch, computed by patlog for patmake's perusal.
- The candidate entry for ChangeLog.
- File filled in by mailagent's "@SH package" command, normally issued by Configure, recording some of the users who kindly registered themselves.
Most of this should be built into RCS.
Larry Wall (version 2.0).
Raphael Manfredi <Raphael.Manfredi@pobox.com>.