Every now and then I've been asked if abapGit can be used to deploy software to production, while it might be technically possible I heavily discourage doing it.
abapGit and CTS
SAP Change and Transport System(CTS) is used in ABAP systems to deploy software between systems, I guess most ABAP developers are familiar with transactions like SE09 and STMS
abapGit is a git client for ABAP written in ABAP, it is open source, and developed by the community on GitHub. It has been around for ~5 years is in active development, and works with versions 702 and up.
abapGit is meant for helping software development, not software deployment, the requirements for these two areas are very distinct:
In software development, one developer might choose to switch from emacs to vim, this should not impact software delivery. A lot has happened in version control software, CVS has moved to SVN, SVN has moved to git, and git will move to something else sometime.
However, in the normal world(outside ABAP), the software delivery method has typically stayed the same with distributing executable/compiled files.
A lot of things break during software development, but it is not good to have errors during software delivery, as it can possibly impact production systems and users.
Having ABAP code in git does not enable parallel development.
ABAP systems have one active version of a program, and abapGit works on package level, ie. it's possible to have one branch checked out at a time. Developer1 can work on feature1, but developer2 cannot work on feature2 if it's the same package, in the same system.
Some features might be added in abapGit to have one branch checked out and commit to a different branch, but it will not be very transparent.
To have parallel development, multiple ABAP systems/containers are required, as suggested by Ethan Jewett back in 2015.
In a typical 3 tier setup, DEV -> TST -> PRD, I don't have an overview of how to do the branching properly. Assuming a setup with weekly fixed releases is quite easy, it will work with trunk based development.
But an agile setup where every feature can move through git and the environments gives more complexity and a lot of CI runs. I assume some manual testing will happen in TST, so some changes might be 1 hour on the way to production, and some might take months.
It feels like it would end up with the same as STMS, a list of branches for each system that has not been merged to that system/branch yet, with similar pitfalls as CTS.
Each traditional CTS transport can be seen as a "mini" branch for the objects that it contains, which is automatically rebased when imported to the target system. It would also be possible to put every object in its own package to simulate this behavior in git, but it would be a lot of work and not help developers.
abapGit works on package level, CTS on the objects that is part of the transport. If doing software delivery with abapGit I see the following three possibilities:
Make assumptions regarding the target system state, look at the git history to find the limited list of objects to import
Import all objects in the package every time
Compare all objects in git vs the system and import everything that does not match
Making assumptions regarding ABAP system state is not something I'd recommend, everything can happen in ABAP 🙂 It would also go against the reliability requirement for software deployment listed in the beginning.
abapGit is slower than CTS, and say the package contains 10k objects, and aiming to have an "agile" setup with changes constantly moved to production, I'm afraid this would end up in a deadlock situation where features cannot be imported because the previous imports are still running.
CTS on the other hand, works on object level, so if a developer changes a single report, only that report is part of the transport, making it the fastest import setup possible.
With git it's easier to do rollback, yes, especially during development on a feature branch, as it would typically not have any consequences.
Rolling back changes to production is more difficult, assuming it's not the latest commit. Say, three unrelated changes, feature1 + feature2 + feature3 have been deployed successfully, and after a while a logical error is found in feature1, with git the rollback would cascade to feature2 and feature3, assuming it's in the same package.
And before doing the rollback it must be manually analyzed if all 3 features can be rolled back, in case of DDIC changes.
Many customer custom developments are decoupled, consisting of small individual programs in a big package. With CTS it is possible to create a new transport rolling back feature1, not having to consider feature2 and feature3, there are also partner solutions that can do this automatically.
During my last 15 years of using CTS for ABAP deployment, I don't recall having experienced a bug in CTS. Having bugs in the software deployment software, possibly causing requirements to upgrade to the latest S/4 HANA would be bad, software deployment needs stability.
Software development, on the other hand, needs agility, the possibility for new tools and processes. Hence abapGit can be upgraded without impact on the landscape, across a wide range of versions(702+), assuming it's used for software development and not deployment.
Keeping the core clean
I see CTS as part of the core, it's a prerequisite for running an ABAP installation. CTS is a good example of keeping the core clean and stable.
abapGit on the other hand is an option, with rapid development and changes, and should not be part of the core.
CTS could provide an API, so that new innovations can be built outside the core, utilizing the API. Integrations and APIs are also very important in the lower levels of the technology stack.
I'm not a git or CTS expert, so comments and corrections are welcome 🙂