Programming-Chaos / evolvedTD

Evolutionary tower defense game from the University of Idaho
GNU General Public License v3.0
5 stars 2 forks source link

Getting Started

Welcome to EvolvedTD, an Evolutionary Tower Defense game!

The following sections walk you through cloning, branching, pulling, commiting, pushing, and reviewing code changes using Git.

Cloning

First, we get download a copy of the repository to our machine (note the use of SSH for authentication):

git clone git@github.com:tsoule88/evolvedTD.git
cd evolveTD

Now we run git status. Notice that it will say we are "on branch master", this branch is the default. Its first commit is the first commit in the repo. Commits are discrete sets of changes to the files tracked in the repo, and what Git does is record these in a tree, effectively keeping track of the entire history of changes to a codebase.

Branching

Since the master branch is special, we want to keep it stable, which means that it should only be changed after a code review. So in order to record and share changes that we wish to eventually merge into master, we commit them to a different branch and share that. Once the branch's changes have been reviewed by our peers (on GitHub, in a Pull Request), the entire branch is merged back into master.

Before changing any code, we need to make a branch for what we are going to work on:

git checkout -b getting-started

This command did two things: it created the branch getting-started and it checked the branch out, meaning that our HEAD reference (a pointer or marker to the checked out commit) now points to the tip of the new branch. When we commit, it is added to the commit tree as a child of the commit that our HEAD reference marks. With each commit we add to this branch, the getting-started branch reference (and HEAD) is moved to the new commit, but master (and other branches) are not, so the history has diverged. This is a good thing, because the differences can be compared, discussed, reviewed, and merged back in easily via pull request.

The two things the above command did are equivalent to this:

git branch getting-started
git checkout getting-started

Git has lots of documentation; use git help subcommand to read it, e.g. git help checkout.

Checking out a reference, i.e. a branch name or a commit hash, will update the local files such that they match their state at the given reference. If git status indicates we have modified files (use git diff to see the differences), then we'll either want to commit them, or if we're not quite ready to do so, we can stash them away for later using git stash.

Use git status -sb to quickly see the current branch and changes

Whenever we start a new branch, unless another line of development makes more sense, we should branch" it off master, which just means checking it out first:

git checkout master
git checkout -b new-topic

Pulling

Let's say a pull request was recently merged into master and we want to obtain those changes. It's as easy as checking it out and pulling:

git checkout master
git pull

We should then integrate the latest changes into our topic branch, which requires merging. For most cases the merge will be trivially resolved by Git; manual intervention is only necessary if the same lines of code have been changed on each side of the merge.

Merges always merge into the current branch, from the one specified on the command line. Since we want the changes in master to be merged into new-topic, we'll start there:

git checkout new-topic
git merge master

Git will create a "merge" commit which brings the two branches of the tree together by having each as a parent, for two parent commits. Git will allow us to edit the commit message before saving it, but we should keep the summary.

Commiting

Go ahead and make some changes to any tracked file and save it. Git status should now list the file as modified under "Changes not staged for commit". This means Git is aware we've changed the file, but no more. If we want to record these changes, we need to stage them to the index (i.e. staging area), which is Git-speak for marking which changes we want in the next commit (it doesn't have to include every change).

See the online documentation for more!

Unstaged changes can be seen using git diff, and staged changes can be seen using git diff --cached (--staged is an alias). The diff command also accepts arbitrary ranges of references as well as paths to particular files, but by default it shows the changes between our working tree and the staging area, hence why staged changes are excluded without the flag.

Perhaps we fixed a typo in this guide and so have modified the file README.md, and we want to add the all the changes in the file to the staging area:

git add README.md

Git status should now list the modifications under "Changes to be committed".

Use git add --patch to interactively stage changes

To finish the commit and thus record the changes to the repository, we need to write a commit message. The first line should be a one-line summary of the changes in the commit; for longer messages, leave a blank line, and then add details. Commit messages should be imperative: "Fix bug", not "Fixed bug" or "Fixes bug", e.g.:

git commit -m "Fix typo in Getting Started"

Executing git commit without a message will open the program defined in our $EDITOR environment variable, where a longer commit message can be written.

Pushing

Let's suppose we have spent the day working on a feature with its own branch foo. Since we work with a team, we need to share our work, even if it's not completely finished, so we can refactor our code based on input we receive from our peers. First, let's take a look at our log of changes so we can remember what we've done:

git log --decorate --graph --oneline

The --decorate flag causes references to be emitted in the log, so we can see where each branch is. The --graph flag draws an ASCII tree; and --oneline shows just each commit's summary. The top is most recent, so we should see foo but not master right before that commit's summary. Since we are using GitHub Flow, master and the remote branch that it tracks, origin/master, should always be on the same commit if we are not actively merging a finished pull request. Once we are sure we have not erroneously committed to master, we will push our changes to GitHub:

git push --set-upstream origin foo

If we have not pushed this branch to GitHub before, we need to specify it with --set-upstream origin. When we cloned Dr. Soule's repo, Git kept track of where it came from by adding a remote named origin. We are setting origin as the upstream repo, and adding a new branch foo to origin, so that all subsequent executions of git push when we are on our local branch foo will automatically send our changes to GitHub and move the origin/foo reference accordingly.

Our peers can then pull our changes into their local repos by running git pull, and then check those changes out with git checkout foo.

Reviewing

Once we've pushed our changes to GitHub, we want to write a perfect pull request. After navigating to the project's page, we want to find our branch foo and click the "Compare & review". This will present with an overview of changes between our branch and master. After writing the pull request, we then submit it by clicking "Create pull request".

The PR will now be viewable to all our peers, and open for discussion. We can add to the PR by pushing more commits to foo; in fact, any collaborator can do this, so don't forget to pull their changes before continuing your work.

Before a PR can be signed-off, it needs to be able to merge cleanly, which means merging master into foo, such that when we merge foo into master, Git will simply "fast-forward" the master branch reference to the merge commit at the head of foo. Since this commit's parents were master and foo to begin with, the second merge won't make another commit as it can just move the reference forward.

"Any sufficiently advanced technology is indistinguishable from magic." - Arthur C. Clarke

After the PR has been sufficiently reviewed and signed-off by another team member, we will want to merge it back into master. Since we already took the time to ensure this will happen cleanly, GitHub offers us a button at the bottom of the PR to automatically perform the merge from foo into master, thus completing and closing the pull request.

Reverting

If published commits break the game, or for any other reason need to be "undone", the git revert command will create inverse commits of the specified range, which are then committed, effectively undoing the changes in a recorded manner. GitHub also provides a revert button on merged pull requests.

Dependencies

Processing

If you wish to use Processing from the command line (or with processing-mode in Emacs etc.), install the binaries through the menu "Tools" -> "Install 'processing-java'".

Box2D for Processing

Installing

Can be installed through the Processing IDE menu "Sketch" -> "Import Library" -> "Add Library" -> "Box2D for Processing". Can also be installed manually.

Handling Memory Leaks

Box2D has an issue with memory leaks. Specifically, it keeps references to bodies around much longer than it is supposed to. The bodies have a userData field with a reference to our objects (creatures, projectiles, food, etc.). When the body sticks around with the reference, it prevents Java's garbage collector from cleaning up our objects.

After a long search for our memory leak, we have found that the fix for this is to remove the reference via body.setUserData(null) when the body is no longer needed (that is, in killBody() for the respective object). Box2D bodies also have fixture lists, which similarly need their user data set to null. If we ever add joints, they would need to be taken care of as well. Please ensure this is done for any new body types added to EvolvedTD. See #203 and #205 for more information.

Data Collection

If you wish to add a field to the data collection there is a simple process to do so:

1) Add your data field to the correct table

2) Add your row of data to the collection.

Now your data should show up in the designated table and column in our excel and csv files.

Editing the Excel File

Whenever you open the excel file it auto-populates it's information from the csv files generated by the game. It then re-populates ever minute automatically. This should mean that editing the excel file should not really be necessary. It also means that the excel file technically changes everytime it is openned, so be careful to not save it or you will have a new commit to make.

Changes to the excel file should only be saved and made in the master excel file. This is because with everyone adding different data collection, there may be several different formats for the same collumn. Thus the data file for your branch should never be merged to master, rather the master excel file should be edited after your branch has been merged.

Certain edits that you may want to do to the excel file would be coloring and formatting of the cells in the collumn of data that you added.

You can find tutorials on basic excel editing here: Easy Excel Tutorials