Sunday, May 22, 2011

My First SVN Hook

Finally, I got a chance to write my first SVN hook.
Seems to be very easy and useful.

My Goal

My first hook relates to the migrate plugin in play!framework.

For those of you who use rails, you might know this plugin/module/gem already.
It helps you version your DB schema.
Each file has the following pattern :




As you can see - the version is a number that increments each time. When you work in a team - you will get a lot of clashes between versions.

The version of this module I was using (which is quite old by now), had a bug. If you had the combination of "up" or "down" in the name of the version, it would fail. So for example :


would fail, as the word "Group" has the combination UP in it..

There are 2 things I needed to resolve with my pre-commit hook :
  • make sure the version does not clash
  • make sure the name does not have the combination "up" or "down"

Writing the Hook

In unix it's quite easy to write the hook in any language you want as long as the Shebang line is there.

I decided to write my hook in Python.

The svnlook Command

It seems that the SVN comes with a simple command called svnlook that gives you all the API you need to write a hook.
You execute it in a command line (from code, as a different Process) - and you parse the output.
I found the commands :

svnlook changed REPOS_PATH



most useful.

Where to Start From

After you read the API a bit, and you have an idea about how to write the hook, you start by reading the command line arguments.
The first one holds the "REPOSITORY_PATH" and the second one holds the "TRANSACTION_ID".

The TRANSACTION_ID - exists only while committing, which makes it harder on development. I replaced the flag "-t" for transaction with the flag "-r" for revision while developing, because it is easier to run the script.

In Python, you will usually have the following code :
#!/usr/bin/env python

import sys
import os

if __name__ == '__main__':
        if ( len(sys.argv) ) < 3:
                sys.stderr.write("Usage: %s REPOS TXN\n"%(sys.argv[0]))
                repository_location = sys.argv[1]
                transaction_id = sys.argv[2]
                main(repository_location, transaction_id)

where "main" is the function you should implement. I implemented it like this :
def main(rep_location, tx_id):
        added_migrate_files = get_added_migrate_files(rep_location, tx_id )
        for  migrate_file in added_migrate_files:
                check_added_migrate_file(migrate_file, rep_location)
ass you can see - I first get all the new files that are added, and filter out only those who match the "migrate" files pattern. Then, I iterate over them, and check each one. Lets see how I implemented them
def get_added_migrate_files(rep_location, tx_id):
        changes = os.popen('svnlook changed ' + rep_location + ' -t ' + tx_id ).read().split('\n')
        print('found changes : ' + ",".join(changes) )
        res = []
        for  change in changes :
                lchange = change.lower()
                print('checking if [' + lchange + '] starts with a %s'%(lchange.startswith('a')))
                if ( lchange.startswith('a') and migrate_path in lchange ):
                        print('looking at change : ' + lchange)
        return res
In order to get the added files, I execute with "os.popen" the "svnlook changed" command. I read the output from the process with "read()" and split the result to get separate lines. For each line I check if it represents an addition. In subversion, an addition is noted by a capital "A" at the beginning, but since I am working with "lower()" I compare it with "a" instead. The "migrate_path" is the relative path in repository where the migrate scripts' folder resides. I decided to refer to a single path. It might not be your choice, but I preferred to be safe than sorry. So when I run "lchange.split(migrate_path)[1]" - I am actually extracting the name of the migrate script file. this function returns all the names of added migrate files. Now lets have a look at how I check if they are valid
def check_added_migrate_file(added_migrate_file, rep_location):
        migrate_file_parts = added_migrate_file.split(".")
        print('checking migrate file : ' + added_migrate_file )
        migrate_version = migrate_file_parts[0] + "." + migrate_file_parts[1] # a version is a combination of number and up/down
        print('migrate_version : ' + migrate_version)
        migrate_file_name = added_migrate_file
        lower_name = migrate_file_parts[2].lower()
        print('name : ' + lower_name )
        tree = os.popen('svnlook tree ' + rep_location + ' ' + migrate_path,'r').read().split('\n')
        for tree_file in tree:
                filename = tree_file.strip()
                if ( filename.startswith(migrate_version) ):
                        sys.stderr.write("migrate file with version = " + migrate_version + "already exists!")
        if ( "up" in lower_name.lower() ):
                sys.stderr.write("file " + migrate_file_name + " contains the sequece 'up' (case insensitive). need to remove it")
        if ( "down" in lower_name.lower() ):
                sys.stderr.write("file " + migrate_file_name + " contains the sequence 'down' (case insensitive). need to remove it")

This algorithm looks complicated but it's quite simple. The main thing to note here is that I am running the "svnlook tree" command. giving me the existing migrate scripts in the repository. Once I have them I simply compare the version to the added files with 2 loops. (I don't care about performance here). I also verify that a file name does not contain the sequence "up" and "down". As a new python developer, I was amused to discover it is accomplished with the "in" keyword.

Important Things to Note

  • I keep writing to stderr. These messages will appear later in SVN output incase the commit fails
  • debugging a hook is hard. Even though I print to stdout, you have to wrap the "pre-commit" with a script that outputs the stdout to a file in order to be able and debug it.
  • while developing, I simply replaced the "-t" flag with "-r" - allowing me to run the hook from command line, giving it revision numbers instead of transactions. A transaction id only exists during an actual commit.
  • In my installation, pre-commit hooks did not run by default. I had to go to my repository manually, and under a folder named "hooks" I had to rename a file named "pre-commit.tmp1" to "pre-commit" (in unix).
    Then I had to add a line of code in "pre-commit" to add my new hook to the flow.
  • I had some problems with permissions. So make sure you give permissions to execute your hook.

Great Tip for Debugging

I suggest you change "pre-commit" script, to return "1" instead of "0" while you develop a new hook.

This way - the commit will always fail, and you can easily repeat it.

For example - my hook takes affect on "added files". So if the commit succeeds, I need to create new files, add them to SVN and then commit = 3 operation.
However, when I fail the commit each time, all I have to do is commit again = 1 operation.
Saves a lot of time.

Wrapping Pre-Commit for Debugging

This is pretty simple, but not a lot of people know how to do it.
/svn_rep_path/hooks/pre-commit-run $* > /svn_rep_path/hooks/my.log
exit 1
The trick here is to use "$*" in order to pass all command line arguments to the real script. "pre-commit-run" here Once you're done, you should be able to commit wrong files and get errors. I use tortoise SVN so I got this :

No comments :

Post a Comment