NAME

Cvs::Trigger - Argument parsers for CVS triggers

SYNOPSIS

    # CVSROOT/commitinfo
    DEFAULT /path/trigger

    # /path/trigger
    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();
    my $args = $c->parse("commitinfo");

    if( $args->{repo_dir} =~ m#/secret$#) {
        die "You can't check stuff into the secret project";
    }

    for my $file (@{ $args->{files} }) {
        if( $file =~ /\.doc$/ ) {
            die "Sorry, we don't allow .doc files in CVS";
        }
    }

DESCRIPTION

CVS provides three different hooks to intercept check-ins. They can be used to approve/reject check-ins or to take action, like logging the check-in in a database.

commitinfo

Gets executed before the check-in happens. If it returns a false value (usually caused by calling die()), the check-in gets rejected.

The following entry in the CVS admin file commitinfo calls the hook for all check-ins:

    # CVSROOT/commitinfo
    ALL /path/cvstrig

The corresponding script, /path/cvstrig, parses the arguments which cvs passes to them:

    # /path/cvstrig
    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();
    my $args = $c->parse("commitinfo");

Note that you need to specify the hook name to the parse method, because CVS provides the different hooks with different parameters. In case of the commitinfo hook, the following parameters are available as keys into the has referenced by $args:

repo_dir

Full path to the repository directory where the check-in happens, e.g. /cvsroot/foo/bardir.

files

Reference to an array of filenames involved the check-in. No path information is provided, all files are relative to the repo_dir directory.

opts

Additionally, optional parameters passed to the trigger script are available with this parameter. Note that the number of these parameters needs to be passed to the parse method:

    # CVSROOT/commitinfo
    ALL /path/cvstrig foo bar

    # /path/cvstrig
    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();
    my $args = $c->parse("commitinfo", { n_opt_args => 2 });

        # => "foo-bar"
    print join('-', @{ $args->{opts} }), "\n";
verifymsg

Gets executed right after the user entered the check-in message. Based on the message text, the check-in can be approved or rejected.

This hook is typically used to enforce a certain format or content of the log message provided by the user.

Here's an example that checks if the check-in message references a bug number:

    # CVSROOT/verifymsg
    DEFAULT /path/checkin-verifier

    # /path/checkin-verifier
    #!/usr/bin/perl
    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();
    my $args = $c->parse("verifymsg");
    
    if( $args->{message} =~ m(fixes bug #)) {
        die "No bug number specified";
    }

verifymsg provides the message, accessible by the message key in the hash ref returned by the parse method. Additionally, the opts key provides a list of optional parameters passed to the script (check commitinfo for details).

loginfo

Gets executed after the check-in succeeded. It doesn't matter if the corresponding script fails or not, the check-in has already happend by the time it gets called.

An entry like

   DEFAULT /path/string

will call the loginfo script with the following data on STDIN:

    Update of /cvsroot/m/a
    In directory mybox:/local_root/m/a
    
    Modified Files:
           a1.txt
    Log Message:
    Fixing some bug, forgot which one. Yay!

There's no need to parse this, though, Cvs::Trigger will do that for you. The following hash keys are available:

repo_dir

Full path to the repository directory where the check-in happens, e.g. /cvsroot/foo/bardir.

host

Name of the host where the check-in has been initiated.

local_dir

The directory in the user's workspace where the check-in got initiated.

message

Check-in message.

files

Reference to an array of filenames involved the check-in. No path information is provided, all files are relative to the repo_dir directory.

loginfo scripts can get additional data from cvs. For this to happen, the call syntax in the loginfo administration file needs to change to this format:

   DEFAULT ((echo %{sVv}; cat) | /path/script)

The first line piped into the script's STDIN then consists of the file name, the previous and the new revision number, all space-separated (oh well, this seems to have been invented before spaces in file names came around):

    module/path file1.txt,1.3,1.4 file2,1.1,1.2
    Update of /tmp/RgNSQ4Yomr/cvsroot/module/path
    In directory mybox:/tmp/RgNSQ4Yomr/local_root/module/path

    Modified Files:
        file1.txt file2.txt
    Log Message:
        Here are my check-in notes.

In order to parse this enhanced format, the call to Cvs::Trigger's parse method needs to be modified:

    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();
    my $args = $c->parse("verifymsg", { rev_fmt => "sVv" });

The result in args will then store the file names and their revisions under the revs key:

    use Data::Dumper;
    print Dumper($args->{revs});

        # $VAR1 = { file1.txt => [1.3, 1.4]
                    file2.txt => [1.1, 1.2]
                  }

Use the same script for multiple hooks

You can call the same trigger script in multiple hooks. Since the parameters passed to the script vary from hook to hook, the easiest solution is to pass the hook name on to the script, so that it can switch the command argument parser accordingly:

    # CVSROOT/commitinfo
    DEFAULT /path/trigger commitinfo

    # CVSROOT/verifymsg
    DEFAULT /path/trigger verifymsg

    #!/usr/bin/perl
    use Cvs::Trigger;
    my $c = Cvs::Trigger->new();

    my $hook = shift;

       # First argument specifies the parser
    my $args = $c->parse( $hook );
    
    if( $hook eq "verifymsg" ) { 
        if( $args->{message} =~ m(fixes bug #)) {
            die "No bug number specified";
        }
    } 
    elsif( $hook eq "commitinfo" ) { 
        if( $args->{repo_dir} =~ m#/secret$#) {
            die "You can't check stuff into the secret project";
        }
    }

Remember fields by caching

THIS FEATURE IS EXPERIMENTAL. USE AT YOUR OWN RISK.

If you want to make a decision based on both the file name and the check-in message, none of the hooks provides all necessary information in one swoop. If, say, .c files need a bug number in their check-in message and .txt don't, here's a tricky way to forward the filenames parsed by commitinfo to the verifymsg hook, which has the check-in message available:

    # CVSROOT/commitinfo
    DEFAULT /path/trigger commitinfo

    # CVSROOT/verifymsg
    DEFAULT /path/trigger verifymsg

    #!/usr/bin/perl
    use Cvs::Trigger;

        # Turn on the cache
    my $c = Cvs::Trigger->new( cache => 1 );

    my $hook = shift;

       # First argument specifies the parser
    my $args = $c->parse( $hook );
    
    if( $hook eq "verifymsg" ) { 
        # We're in verifymsg now, but the cache still holds the file
        # names obtained in the commitinfo phase
        if( grep { /\.c$/ } @{ $args->{cache}->{files} } and
            $args->{message} =~ m(fixes bug #) ) {
            die "No bug number specified in .c file";
        }
    } 

Caching has a couple of gotchas, though. First, items can only stay in the cache for a limited time, to avoid a cache overflow with many simultaneous checkins going on.

However, the time span between commitinfo and verifymsg can hardly be estimated accurately. What if someone types "cvs commit" and then goes to lunch? The editor window will stay open, and if the message gets saved a couple of hours later, the cache still needs to hold a copy of the commitinfo data.

Deleting the cache data once verifymsg is done with it doesn't work either. If you type "cvs commit" in a directory with multiple subdirectories, both the commitinfo and verifymsg will get called for each subdirectory containing modified files. Cvs::Trigger therefore maintains a TTL (time to live) counter to keep track of how many instances of verifymsg are still going to read it. Bottom line: The cache entry will be deleted once the last verifymsg instance is done with it.

Nevertheless, determining the cache timeout is a delicate issue. The default values are set as follows:

        # Turn on the cache
    my $c = Cvs::Trigger->new(
       cache                     => 1,
       cache_default_expires_in  => 3600,
       cache_auto_purge_interval => 1800,
       cache_namespace           => "cvs",
    );

Therefore, the cache will expire entries after an hour and it will run the check/prune procedure every half hour. To set different values, simply call new with different parameters. The cache namespace can also be configured, see the Cache::Cache manual page for details.

The cache makes use of the fact that the commitinfo and verifymsg scripts are run by processes sharing the same parent pid (ppid). The cache indexes its data using this pid value. If the operating system reuses the same pid within the expiration timeframe, a clash will occur.

TODO List

    * Try filenames with commas, spaces, and newlines
    * tests for optional arguments
    * methods vs. hash access
    * no STDIN on loginfo => hangs

SEE ALSO

http://ximbiot.com/cvs/wiki/index.php?title=CVS--Concurrent_Versions_System_v1.12.12.1:_Reference_manual_for_Administrative_files#SEC184

LEGALESE

Copyright 2006 by Mike Schilli, all rights reserved. This program is free software, you can redistribute it and/or modify it under the same terms as Perl itself.

AUTHOR

2006, Mike Schilli <m@perlmeister.com>