NAME

App::Multigit - Run commands on a bunch of git repositories without having to deal with git subrepositories.

PACKAGE VARS

%BEHAVIOUR

This holds configuration set by options passed to the mg script itself.

Observe that mg [options] command [command-options] will pass options to mg, and command-options to mg-command. It is those options that will affect %BEHAVIOUR.

Scripts may also therefore change %BEHAVIOUR themselves, but it is probably badly behaved to do so.

report_on_no_output

Defaults to true; this should be used by scripts to determine whether to bother mentioning repositories that gave no output at all for the given task. If you use App::Multigit::Repo::report, this will be honoured by default.

Controlled by the MG_REPORT_ON_NO_OUTPUT environment variable.

ignore_stdout

ignore_stderr

These default to false, and will black-hole these streams wherever we have control to do so.

Controlled by the MG_IGNORE_{STDOUT,STDERR} environment variables.

concurrent_processes

Number of processes to run in parallel. Defaults to 20.

Controlled by the MG_CONCURRENT_PROCESSES environment variable.

skip_readonly

Do nothing to repositories that have readonly = 1 set in .mgconfig.

Controlled by the MG_SKIP_READONLY environment variable.

@SELECTED_REPOS

If this is not empty, it should contain paths to repositories. Relative paths will be determined relative to <mg_root|/mg_root>.

Instead of using the .mgconfig, the directories in here will be used as the list of repositories on which to work.

Each repository's origin remote will be interrogated. If this exists in the .mgconfig then it will be used as normal; otherwise, it will be treated as though it had the default configuration.

FUNCTIONS

These are not currently exported.

mgconfig

Returns .mgconfig. This is a stub to be later configurable, but also to stop me typoing it all the time.

mg_parent

Tries to find the closest directory with an mgconfig in it. Dies if there is no mgconfig here. Optionally accepts the directory to start with.

all_repositories

Returns a hashref of all repositories under mg_parent.

The keys are the repository directories relative to mg_parent, and the values are the hashrefs from the config, if any.

selected_repositories

This returns the repository configuration as determined by <@SELECTED_REPOS|/@SELECTED_REPOS>. Directories that exist in the main config (all_repositories) will have their configuration honoured, but unknown directories will have default configuration.

each($command[, $ia_config])

For each configured repository, $command will be run. Each command is run in a separate process which chdirs into the repository first. Optionally, the $ia_config hashref may be provided; this will be passed to "run" in App::Multigit::Repo.

It returns a convergent App::Multigit::Future that represents all tasks. When this Future completes, all tasks are complete.

Subref form

The most useful form is the subref form. The subref must return a Future; when this Future completes, that repository's operations are done.

The convergent Future ($future below) completes when all component Futures (the return value of then, below) have completed. Thus the script blocks at the $future->get until all repositories have reported completion.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run(\&do_a_thing)
            ->then($repo->curry::run(\&do_another_thing))
        ;
    });

    my @results = $future->get;

See examples/mg-branch for a simple implementation of this.

The Future can complete with whatever you like, but be aware that run returns a hash-shaped list; see the docs for run. This means it is often useful for the very last thing in your subref to be a transformation - something that extracts data from the %data hash and turns it into a usefully-shaped list.

The example examples/mg-closes does this, whereas examples/mg-branch uses report.

report in App::Multigit::Repo implements a sensible directory-plus-output transformation for common usage.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run(\&do_a_thing)
            ->then($repo->curry::run(\&do_another_thing))
            ->then($repo->curry::report)
        ;
    });

The subref given to run is passed the %data hash from the previous command. %data is pre-prepared with blank values, so you don't have to check for definedness to avoid warnings, keeping your subrefs nice and clean.

    sub do_a_thing {
        my ($repo_obj, %data) = @_;
        ...
    }

Thus you can chain them in any order.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run(\&do_another_thing)
            ->then($repo->curry::run(\&do_a_thing))
            ->then($repo->curry::report)
        ;
    });

Observe also that the interface to run allows for the arrayref form as well:

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run([qw/git checkout master/])
            ->then($repo->curry::run(\&do_another_thing))
        ;
    });

A command may fail. In this case, the Future will fail, and if not handled, the script will die - which is the default behaviour of Future. You can use else to catch this and continue.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run([qw{git rebase origin/master}])
            ->else([qw{git rebase --abort])
            ->then($repo->curry::report)
        ;
    });

The failure is thrown in a manner that conforms to the expected Future fail interface, i.e. there is an error message and an error code in there. Following these is the %data hash that is consistent to all invocations of run. That means that when you do else, you should be aware that there will be two extra parameters at the start of the argument list.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run([qw{git rebase origin/master}])
            ->else(sub {
                my ($message, $error, %data) = @_;
                ...
            })
            ->then($repo->curry::report)
        ;
    });

In the case that you don't care whether the command succeeds or fails, you can use finally to catch the failure and pretend it wasn't actually a failure.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run([qw{git rebase origin/master}])
            ->finally($repo->curry::report)
        ;
    });

Despite the name, finally does not have to be the final thing. Think "finally" as in "try/catch/finally". In the following code, finally simply returns the %data hash, because finally transforms a failure into a success and discards the error information.

    use curry;
    my $future = App::Multigit::each(sub {
        my $repo = shift;
        $repo
            ->run([qw{git rebase origin/master}])
            ->finally(sub { @_ })
            ->then(\&carry_on_camping)
            ->then($repo->curry::report)
        ;
    });

Arrayref form

In the arrayref form, the $command is passed directly to run in App::Multigit::Repo. The Futures returned thus are collated and the list of return values is thus collated.

Because run completes a Future with a hash-shaped list, the convergent Future that each returns will be a useless list of all flattened hashes. For this reason it is not actually very much use to do this - but it is not completely useless, because all hashes are the same size:

    my $future = App::Multigit::each([qw/git reset --hard HEAD/]);
    my @result = $future->get;

    my $natatime = List::MoreUtils::natatime(10, @result);

    while (my %data = $natatime->()) {
        say $data{stdout};
    }

However, the %data hashes do not contain repository information; just the output. It is expected that if repository information is required, the closure form is used.

mg_each

This is the exported name of each

    use App::Multigit qw/mg_each/;

mkconfig($workdir)

Scans $workdir for git directories and registers each in .mgconfig. If the config file already exists it will be appended to; existing config will be preserved where possible.

write_config

Write a .mgconfig configuration file.

clean_config

Checks the .mgconfig for directories that don't exist and removes the associated repo section.

base_branch

Returns the branch that the base repository is on -the repository that contains the .mgconfig or equivalent.

The purpose of this is to switch the entire project onto a feature branch; scripts can use this as the cue to work against a branch other than master.

This will die if the base repository is not on a branch, because if you've asked for it, giving you a default will more likely be a hindrance than a help.

set_base_branch($branch)

Checks out the provided branch name on the parent repository. Beware of using a branch name that already exists, because this will switch to that branch if it does.

AUTHOR

Alastair McGowan-Douglas, <altreus at perl.org>

ACKNOWLEDGEMENTS

This module could have been a lot simpler but I wanted it to be a foray into the world of Futures. Shout outs go to those cats in irc.freenode.net#perl who basically architectured this for me.

tm604 (TEAM) - for actually understanding Future architecture, and not being mad at me.
LeoNerd (PEVANS) - also for not being irritated by my inane questions about IO::Async and Future.

BUGS

Please report bugs on the github repository https://github.com/Altreus/App-Multigit.

LICENSE

Copyright 2015 Alastair McGowan-Douglas.

This program is free software; you can redistribute it and/or modify it under the terms of the the Artistic License (2.0). You may obtain a copy of the full license at:

http://www.perlfoundation.org/artistic_license_2_0