![[jump to content]](/_.gif){kind=small}
![[No Software Patents]](/patent_button.gif){kind=small}
Tools (new)
: Looking into a tool
[ Home
| Scylla+Charybdis old home
| Tools (old)
| Tools (new)
| FAQ
| Downloads
]Name Last modified Size Description
Parent Directory 05-Jan-2009 18:48 -
GAT/ 05-Jan-2009 18:48 -
ANNOUNCE 26-Apr-2005 00:56 1k
$Header$ Status: ------- This is vaporware today, just a placeholder. What is GAT? ------------ Linus Torvalds wrote GIT, the File Manager From Hell. GIT is ingenous in how it treats files. However it has some slight flaws, which made it impossible for me to use it out of the box: - GIT uses zlib as an implicite compression. I consider zlib a security risk, as in the past, several flaws in zlib were detected (from archive bombs over buffer overruns to denial of service conditions). Also, zlib makes it a complex task to use a better or other compression techniques, also the resulting files are not readable by a native tool like /bin/cat. Therefor GAT uses no compression at all. However in a future, it will allow compressed file store. - GIT stores it's repository under a directory .git. That is not ok for me. I need the files to reside on a central repository on my computer, such that it's more easy to write sync jobs for this. So GAT has an implicite default repository in the environment variable GATROOT (which defaults to "~/.GATROOT/" under Unix and 'C:\GATROOT\' under Windows). If you want this to point somewhere else, just set this variable or use a softlink. GAT is NFS aware. - GIT uses one global .git directory for everything. However I personally like how CVS treats it, such that you can share the same directory under several trees without fear. Therefor I have a GAT directory in each directory (it does not start with a dot, as this is inconvenient under some operating systems which have no Unix like filesystem). - GIT works with changesets, however I like the CVS way of 'tagging'. Therefor GAT tries to resemble this type of CVS tagging, with naming the GAT trees. - GIT does not support CVS tags. However I like how the CVS tags work. So I added this to GAT (at least, I will add this to GAT). - GIT uses SHA-1. I am more fond of MD5, as this is wider available on more platforms (there are even implementations for Microsoft Visual Basic). - GIT does not use the plain file, as it writes a nifty header to it. As you always can guess this header, GAT includes this header in the MD5 hash calculation. This way, after loading the file, it fails the MD5 hash if it's not of the right type. Note that the type is encoded in the file name as well, such that you can easily check it from shell. This way, on CDROM, the GAT file repository and the unpacked source code can share the same data and they can be hardlinked on a readonly repository publishing area. - GIT uses mmap(). That's clever, but I hate it, because I want to be able to process files with arbitrary length. Therefor GAT uses normal file IO. Perhaps more to follow later. What is that name "GAT"? ------------------------ This is no name. It's just a variant of GIT with I replaced by A, because its pronounciation is more like "get" and similarily is no usual english word, so it cannot be mistaken for one. Also I like it, that it's very near 'cat', as implicitely GAT uses a lot of features from cat (like unaltered cleartext files). I don't want look at GIT, so what does GAT? ------------------------------------------- GAT is a basic layer of an VCS. It can replace CVS. It can replace Subversion. There is no fancy UI, but you can trust in GAT, that it keeps your valuables as safe as possible. It works internally much like GIT, so to understand how GAT ensures the integrity of everything you should read what Linus Torvalds writes about GIT. His words are ingenious and contain a deep secret. If you interested in how to use GAT, there is a short HowTo in the following text. It helps, if you know CVS. Prerequisites: -------------- You need curl to build gat. INSTALL: -------- curl http://www.scylla-charybis.com/downloads/gat.tgz | gtar xfz - cd gat make ./gat update make make test make install Note that you will get "release" versions of GAT in this case. Release versions are considered stable code. If you want release early code, you have to switch to a different "branch" using "gat branch". Command overview: ----------------- NOTE THAT THIS SHOWS THE FUTURE. NOT EVERYTHING IS IMPLEMENTED TODAY. Most commands which have no parameters allow you to give a file list to use this command only on the file list. However, as the nature of GAT, nearly any circumstance can be autodetected, so you usually do not need to enter anything. gat sync This does a synchronization with the repository keepers(s). In case you have only read access, this means, only new patches are downloaded. Your patches are always uploaded to private keepers, and published patches are uploaded to public keepers, too. Note that 'gat update' implies 'gat sync' in case you have no local repository set up yet for this branch. gat help Show help gat help <command> Show help of command gat branch List all branches. Note that this list can become very big. gat branch name Switch to another branch. Don't forget to use "gat update". If name contains wildcards the branch is searched for. Branches are: HEAD, ALPHA, BETA, GAMMA, PRE, RELEASE, STABLE Several developers can publish their own branches. You can mix branches by concatenating with +, so gat branch RELEASE+HEAD(tino) says, that you want the release branch, but code from the head branch of tino, too. This probably means, you are tino in a bigger project. For your own projects, just stick to HEAD (your author is implicitely added). Some hints: Use lower case letters if you want to 'name' a branch. Branch names can include any valid character for eMail and domain, with the exception of the '+' (which is perfectly valid for eMail names) and must not start with a -. gat branch +xxx gat branch -xxx Add and remove a branch. Note that you can enter branches, which are unreachable. For example if some author forks of a branch from release and you want to have something from the author from the beta branch, this might lead to nothing valid. gat merge branch Merge the current version with the given branch. Note that this implies, that you have access to the branch can commit to this branch. Afterwards you have to commit, as you have switched the branch implicitely. gat diff Show diff of current directory with selected branch. gat status Like 'diff', only shorter. gat ignore Show the ignores gat ignore filemask This shows which files are ignored. Beware of shell metacharacters, as "gat ignore *.c" works perfectly well, but what you meant was probably "gat ignore '*.c'", which basically is the same (in this case, GAT expands it for you). gat ignore +filemask gat ignore -filemask Add or remove ignores. This probably works well with shell meta characters, as "gat ignore +*.c" usually does expand to "+*.c", as you seldomly have files starting with a +. gat text Shows which files are considered text. Well, gat works well without text. The only difference between text and non-text files is, that CVS-sequences ($Header$ etc.) are replaced by auto generated values and line endings (CR, CR LF, LF) are converted to the machine standard. Note that you need not to worry about files to be text or non-text, as they always are checked in binary. So if you accidentially checked out a binary file in text mode you can fix this by setting it non-text and checkout again. gat text filemask Like "gat ignore filemask" only for text gat text +filemask gat text -filemask Treat files as text, implicitely (+) or as non-text (-). gat co gat update Update the local directory with the data from the repository. Note that you need to use "sync" in case you have added this project to your local repository. GAT will tell you if you probably need to sync. gat ci gat commit Update the GAT repository. Note that you need write access to a repository for this. You always have write access to your local repository, however it might be, that you have not initialized it, so this will tell you and fail. Note that you can ci/co individual files, too. All files checked in together form a changeset. gat author gat trust Shows who currently is author for the changes and who you trust. gat author name Set an author and implicitely trust this author. Note that you might get into trouble if this author is not globally unique. So perhaps use your eMail address or a fake(!) eMail address which belongs to you as the author. Note that you can switch between authors easily. You can only set an author if you are this author. Authors may be protected by an implicite or explicite passphrase. If it's implicite, this means, you have to enter it when you use 'gat author', else you have to use 'gat sign'. Beware! Explicite passphrases are stored on disk and may leak. Never use explicite passphrases on multi user systems, YOU HAVE BEEN WARNED. gat trust +name gat trust -name Trust or untrust author. Note that 'trust' is an alias for 'author'. Only patchsets are downloaded of authors you trust. gat sign Signs the current version. This is only needed if the author is protected by a passphrase and the passphrase is 'explicite'. then you have to enter the passphrase now. Note that "publish" implicitely calls sign. Also note, that you always can 'commit' without signing, as this goes to your local repository. gat publish If you are satisfied what you have, you can publish your code using this command. This means, 'gat sync' will sync this with public masters, too. Note, that his calls 'gat sign' if needed. Note that this automagically calls 'gat sync', however you can interrupt the publishing at any time if you want, as 'gat sync' continues where you left off. Following commands are redundant. They are only needed to make life more safe, just in case you accidentially delete something or such. gat add Commit the add of files/directories. gat remove Commit the remove of files/directories. gat rename Commit the rename of files or directories. This is black magic and often gets it wrong. 'rename' can be archived using 'remove' and 'add' as well, as there is nothing like a real 'rename tracking'. It's just a convenience function. Note that you do not need this commands, as you can switch it implicitely for commit. But this needs discipline on your side. Logging ------- There is an ever growing file "GATLOG", which is written such, that it can be listed using 'tac GATLOG | less'. If you use less directly, it will look a little bit funny. You can remove the file, it will always be created and appended. License: -------- GAT is code under the GNU GPL v2 or higher. Read COPYING. All GAT data structures and scripts which might show up under GAT directories are free of any copyright. So you can use GAT for commercial purpose, it has no impact to you code which you develop with the help of GAT. -Tino $Log$