Version Control

Version control (we use, like virtually everyone nowadays, git) is very important: it allows to roll back to healthy versions of projects, and it goes hand in hand with collaboration (GitLab/GitHub) as well as backups (remotes).

Advantages

Some advantages of using version control (specifically, git, but most is applicable to others as well) are lined out below.

SSOT (Single Source Of Truth)

No more file trees looking like:

directory
│   a.txt
│   help.me.please
│   Important-Document_2018_version1.pdf
│   Important-Document_2018_version2.pdf
│   Important-Document_2018_version3_final.pdf
│   Important-Document_2018_version3_final_really.pdf
│   Important-Document_2018_version3_final_really_I-promise.pdf
│   Important-Document_2018_versionA.pdf
│   Important-Document_2019-03-56.pdf
│   Important-Document_2019-03-56_corrections_John-Doe.pdf
│   Important-Document_2019-03-56_corrections_John-Doe_v2.pdf
│   invoice.docx
│   test - Copy (2).tex
│   test - Copy.tex
│   test.tex
│
└───old_stuff
        Screenshot 1999-09-03-15-23-15(1).bmp
        Screenshot 1999-09-03-15-23-15(2).bmp
        Screenshot 1999-09-03-15-23-15(3).bmp
        Screenshot 1999-09-03-15-23-15(3)_edited.bmp
        Screenshot 1999-09-03-15-23-15.bmp

Instead, there is one working copy looking like:

directory
│   .git (a hidden directory)
│   a.txt
│   Important-Document.pdf
└───Properly-Named-Invoice.docx

assuming that a.txt is actually needed. All the old junk and redundant copies have been pruned. However, nothing was lost. The entire history is contained within git. The history is readily summoned anytime, if so required. Git calls this its log. Git works best (some would say only) on text-based files, but it can deal with images, PDFs etc., too.

The history and everything else git needs is contained in its .git directory, which is hidden on both Linux and Windows. Do NOT mess with this directory directly: interact with git only through its command interface. Everything else in directory, so in this case a.txt, Important-Document.pdf and Properly-Named-Invoice.docx, is accessible as usual. There is no difference to how you would normally work with these files. They are on your local disk. Together, they are called the working tree.

Therefore (provided that git is used correctly):

duplicate files are gone,
the art of cumbersome file naming will finally be forgotten,
old stuff can be safely deleted; this cleans up the working tree and makes it clear which files are no longer needed. Only the currently needed files are visible, the rest is (retrievable!) history.

File Versioning

Everything is versioned. Outputs (e.g. PDFs in the case of LaTeX) can be matched to the source code that generated them exactly. Important states can be tagged and found again easily.

Collaboration

Each contributor has a version of the source on their local machine. Adjustments are made there, and sent to a central, online repository if they are considered ready to be published. Git can also be used in a distributed fashion (its original strong suit), but we assume a remote repository on a platform like GitLab or GitHub. Developers can then also fetch the latest changes from the remote and incorporate them into their local copy.

Do not confuse GitHub, GitLab and others with the tool itself, git. Microsoft's GitHub is not synonymous to git. A crude, mostly wrong analogy would be: OneDrive is the platform you do collaboration, version control and sharing on. This is like GitHub (GitLab, ...). Office programs like Microsoft Word are used to create original content. This is like source code, as created in some editor of your choice. Word's built-in revision history, in conjunction with the process of naming files, for example 2020-05-13_Invoice_John-Doe-Comments_Final.docx (ISO-8601 oh yeah), would be git. It "only" does the version control, but is not a platform for source code.

Backups

The remote repository also serves as a back-up solution. So do all the distributed local copies. At all points, there will be a workable copy somewhere. In general, git makes losing data extremely hard. When (not if!) you get into a fight with git about merging, pulling, rebasing, conflicts and the like, think of it as git protecting you and your work. Often, the reason for git "misbehaving" and making a scene is because it flat-out refuses an operation that would destroy unsaved changes. In the long run, this behavior is the desired one, as opposed to losing unsaved (git lingo: uncommitted) data.

Branching

Branches, essentially deviating paths in development, are at the heart of git. In git, they are lightweight constructs and not as heavy as in for example SVN. You can experiment liberally using branching, combine and split ideas, files etc.

Bug Fixing

Bug fixing can be accelerated through git bisect, a search algorithm that helps pinpoint commits (stages of development) that introduced regressions. If you know that something is broken today but worked two weeks ago, and there are say 80 commits in your history since then, bisection helps you find the culprit much, much faster than manually testing 80 versions.

Getting Started

Download for Windows here. Install it and spam Next without reading the installation and warning prompts, as always. For Linux, apt-get update && apt-get install --yes git or whatever. You likely have it available already.

Then, somewhere on your machine (example for Linux), see also here:

# Create empty directory
$ mkdir test
# Go there
$ cd test
# Set up your git credentials; this will show as the 'Author' of your work
$ git config --global user.name "Foo Bar"
$ git config --global user.email "foo@bar.com"
# Initialize an empty git repository
$ git init
Initialized empty Git repository in ...
# Create some dummy file
$ echo "Hello World!" > test.txt
$ git status
On branch master

No commits yet

Untracked files:
  (use "git add <file>..." to include in what will be committed)
        test.txt

nothing added to commit but untracked files present (use "git add" to track)
# Rather listen to what we're being told. The dot adds everything:
$ git add .
# There is now a change! The file is ready to be committed, aka "saved" into
# the history.
$ git status
On branch master

No commits yet

Changes to be committed:
  (use "git rm --cached <file>..." to unstage)
        new file:   test.txt
$ git commit -m "Initial file!"
[master (root-commit) 3aaded0] Initial file!
 1 file changed, 0 insertions(+), 0 deletions(-)
 create mode 100644 test.txt
# Have a look at the history up to here:
$ git log --patch
commit 3aaded0365524e9c0cf7c3bc3cb72e1e993def74 (HEAD -> master)
Author: Foo Bar <foo@bar.com>
Date:   Mon Sep 28 15:29:25 2020 +0200

    Initial file!

diff --git a/test.txt b/test.txt
new file mode 100644
index 0000000..de39eb0
Binary files /dev/null and b/test.txt differ

And that is the gist of it. Next, you would want to create a project on GitLab or similar and connect that to your local repository. You can then sync changes between the two, enabling collaboration (or a bunch of other advantages if you keep to yourself).

Graphical Interfaces

There are numerous git GUIs available. For example, they are great at visualizing the commit history (which can get convoluted, especially if you're doing it wrong and enter git merge hell), but also offer all the regular CLI functionality in GUI form (try a dog for terminal log visualization). Git for Windows comes with a built-in GUI as well, but it's basic:

Modern, popular IDEs like VSCode have native git support and also offer git extensions which should satisfy all needs. They might look like:

GitLab

GitLab is a platform to host git repositories. Projects on there can serve as git remotes. As mentioned, in this sense, it is like Microsoft's GitHub, the first large website to offer such a service (still by far the largest today). We use GitLab here because https://collaborating.tuhh.de is GitLab instance and therefore freely available to university members. GitLab is a company with their own GitHub competitor at https://gitlab.com, the source code of which is also open source and therefore be self-hosted (as done in the case of https://collaborating.tuhh.de).

GitLab offers various features for each project. This includes a Wiki, an issue tracker and pull request management. Pull requests (PRs for short; GitLab calls these Merge Requests) are requests from outside collaborators who have forked and subsequently worked on a project. Forking projects refers to creating a full copy of the project in the own user space of collaborators. As such, they can then work on it, or do whatever else they want. If for example they add a feature, their own copy is now ahead of the original by that feature. To incorporate the changes back to the original, the original repository's maintainers can be requested to pull in the changes. This way, anyone can collaborate and help, without ever interfering with the main development in the original.

Continuous Delivery (CI)

Continuous Delivery refers to continuously shipping out the finished "product". In the case of LaTeX for example, these are the compiled PDFs. This is done with the help of so-called Docker containers. The advantages are:

collaborators no longer rely on their local tool chain, but on a unified, common, agreed-upon one. It is (usually) guaranteed to work and leads to the same, reproducible, predictable results for everyone.

Docker (also usable locally, not only on the GitLab platform) helps reproduce results:
- across space: results from coworkers in your office or from half-way across the globe.
  
  You no longer rely on some obscure, specific machine that happens to be the only one on which compilation (PDF production) works.
- across time: if fixed versions are specified, Docker images allow programs, processes pipelines etc. from many years ago to run.
If LaTeX documents become very long, full compilation runs can take dozens of minutes. This is outsourced and silently done on the remote servers, if Continuous Delivery is used. As such, for example, every git push to the servers triggers a pipeline which compiles the PDF and offers it for download afterwards. The last part could be called Continuous Deployment, albeit a very basic version.

Enable Runner for the project

To build anything, we need someone to build for us. GitLab calls these build-servers runners. Such a runner does not materialize out of thin air. Luckily, in the case of collaborating.tuhh.de, runner tanis is available to us. Enable it (him? her?) for the project on the GitLab project homepage: Settings -> CI/CD -> Runners -> Enable Shared Runners. Otherwise, the build process will get 'stuck' indefinitely.

Add git info to PDF metadata

After retrieving a built PDF, it might get lost in spacetime. That is, the downloader loses track of what commit it belongs to, or even what release. This is circumvented by injecting the git SHA into the PDF metadata. This allows you to freely hand out PDFs to people, for example for proof-reading or to editors of journals. You will be able to associate and pinpoint their remarks to a specific, reproducible version of the document. Note though that such a revision process is (much, much) better done using Pull Requests, provided the other party uses git and GitLab.

To identify versions in git, every object is uniquely identified by its hash (SHA256):

412ba291b6980ab21f912b5cdf01a13c6268d0ed

It is convenient to abbreviate the full SHA to a short version:

412ba291

Since a collision of even short hashes is essentially impossible in most use cases, we can uniquely identify states of the project by this short SHA. This is why commands like git show 412ba291 work. So if we have this SHA available in the PDF, never again will there be confusion about versions. The PDF will be be assignable to an exact commit. It can look like this (in Adobe Reader, evoke file properties with CTRL + D):

Using this approach, it is hidden and will not show up in print. You can of course also add the info to the document itself, so that it will be printed.

Add PDF Download Button

On the top of GitLab project pages, badges may be added. That's how GitLab calls the small, clickable buttons. They can be used as a convenient way for downloading built artifacts, e.g. PDFs. It might look like (bottom row):

A little image (svg format) can be generated using shields.io. That only needs to be done once. The result might look like:

To add them to a project, navigate to: Settings -> General -> Badges. Give it a Name, enter a file path (within the repo) or URL for the Badge image URL (or do whatever you want here), and finally enter the Link. This part is a bit tricky, since we need a dynamic URL that adapts to our path. For this, GitLab provides variables like %{project_path}. As such, a URL might look like (the hyphen in the middle is intentional):

https://collaborating.tuhh.de/%{project_path}/-/jobs/artifacts/%{default_branch}/raw/document.pdf?job=build_pdf

The project_path is just your project with its namespace, like peter/thesis, the default_branch is usually just master. It visits the job artifacts on master and gets the PDF of the supplied filename. This filename has to be adjusted accordingly.

Note that the download is unavailable while a job is running. To avoid this, work on a git branch and leave master alone. Treat the PDF (or whatever it is) on master as the current stable version that only changes sometimes, not at every commit. For example, you can do your continuous business on a dev branch and then add a second button.

Hints

These hints will make working with source code and git easier.

Use UTF-8 (and only UTF-8, not anything higher) for text encoding. Stop using Windows 1252, Latin etc. Existing files can be easily updated to UTF-8 without much danger for regression (i.e., introducing errors), for example using Visual Studio Code.
Put each sentence, or even part of a sentence, and each instruction onto its own line. This is important to diff files properly, aka git diff. Otherwise, you will have to use git diff --word-diff, but it's just not the natural/default thing to do. Generally, keep lines short.
Use indentation appropriately. Preferably indent using 4 spaces. There are schools of thought that advocate two spaces, or also one tab. Ultimately, it doesn't really matter. 'Four spaces' just seems to generally pull ahead in terms of mindshare.
Be consistent. Even if you pull your own custom stuff, at least be consistent in doing so. This makes things predictable, the code will be easier to read, and also more easily changed programmatically. GNU/Linux and by extension Windows using Windows Subsystem for Linux offer a very wide range of tools that make search, search-and-replace, and various other operations for plain text files easy. The same is true for similar tools in IDEs.

However, if the text is scattered and the style was mangled and fragmented into various sub-styles, this becomes very hard. For example, one person might use $<math>$ for inline-LaTeX math, another the (preferred) $<math>$ style. Suddenly, you would have to search for both versions to find all inline-math. So stay consistent. If you work on pre-existing documents, use the established style. If you change it, change it fully, and not just for newly added work.