- Related pages
If you are not familiar with Git’s conventional commit approach, jump over here before you continue reading.
I’ve been using Git’s conventional commit approach for at least four years. Recently, it came to my attention that conventional commits don’t fit all codebases; in fact, they can annoy more than they can help in some situations.
The main benefit of using conventional commits that I see is that you can:
CHANGELOGSbased on the semantics of the commit
- Easily allow another developer to understand the impact of a specific commit by reading the commit message
- Implement tools that can act upon a particular commit message
The idea of a changelog is to let anyone understand what has been implemented/fixed/removed from a given version. A changelog is interesting because it provides users with transparency of what is going on within the development team and what they can expect from a specific software version.
The main problem of a changelog is the developer. Developers tend to write software for others to use, but they don’t necessarily care about informing them what has changed and why. That’s when conventional commits came into play. The idea of conventional commits is to use semantic words within the commit message to generate the changelog automatically. So say you have the following commits:
c540106 * Refactor: Move riscv related mmu code to riscv.rs
24694f0 * Docs: Add MMU documentation
946c058 * Fix: Take entries reference instead of an owned value
065dabd * Feat: Display heap size when running main
cd02e93 * Feat: Add MMU Page table support
8eec4b4 * Feat: Add initial MMU page entry bits and entries
You could parse these messages and generate a changelog like so:
# Change Log
* Display heap size when running main
* Add MMU Page table support
* Add initial MMU page entry bits and entries
* Take entries reference instead of an owned value
* Move riscv related mmu code to riscv.rs
* Add MMU documentation
The problem with this approach is that commits are generally too technical and
software-development-focussed. Take the
fix commit: “take entries…” as an
example. What does this tell you as a user of my software? Because from the
user’s perspective, this means nothing.
Changelogs should be written by a human and not by software. Take a look at this alternative changelog:
# Change Log
* Display heap information upon kernel boot
* Add RISC-V paging support
* Fixes then crashing when an user allocates memory in user-mode
* Move any RISC-V MMU related code to the RISC-V module
* Add documentation related to the new MMU component
The previous example is more user-friendly. As a user, I am now aware that there were two main features: the kernel now displays heap information on boot, and RISC-V paging is currently supported. I also know that the documentation related to the new MMU component is now available. Cool, much cleaner, less technical, and separated two main domains: the software developer and the user.
A fix may not necessarily mean anything to a user, but it might mean to a developer. A feature may not necessarily add anything to the end-user. The MMU might mean something to the developer, for which paging is dependent, but not necessarily to the end-user.
Allow developers to understand code impact #
Another problem semantic commits try to solve is by warning the developer of the
impact a commit might have. Specific words, such as
fix is written at
the beginning of a commit message, for example:
c540107 * Fix: Bad memory reference on memor allocation
24694f8 * Feat: Add new MMU component
Commits generally have 50-72 characters (first line) length and 80 for the commit body. The size is limited (with reason), which means we should take every word seriously.
In semantic commits, you can either use
feature, etc., or even by
feat(lang). This takes too much precious character space, limiting
the actual commit argument regarding the change. Countless times, I’ve lost
commit space due to
feat(lang). Sure, it’s ten characters, giving our length
of 50 characters, we have 40 characters left, including space.
The size limit with semantic commits can be a problem because not all commit messages can be 40 or 62 characters.
I do, however, find the scope semantic desirable, even if we have to give up on a few characters.
The scope approach of conventional commits #
As I mentioned previously, I do find the scope approach appealing. It guides the developer on where the change is located by simply reading the commit message. For example:
c540107 * feat(kernel/mem): Check if memory is deallocated before trashing
24694f8 * feat(docs/menu): Add syscall to documentation menu
This is interesting, as I now know which component was modified; however, I have
a few questions regarding the first commit. It does mention it was a
but by reading the message, it does seem like a fix to me or a forgotten logic
that just got added. It seems to me that
before words conflict.
Lets look at the following approach:
c540107 * kernel/mem: Check if memory is reallocated before deleting
24694f8 * docs/menu: Add syscall to documentation menu
I find this much more intuitive, as it still lets me know which component got
added, without the conflict of
The scope approach is interesting in large codebases, where two or more
components exist within the system. I can easily give up a few characters for
Does conventional commits really make sense? #
It depends. A scope-based approach may make sense if you maintain a large codebase. I won’t go on the merit of what exactly is a large project - the project may be complex, but that does not make it necessarily large. How many engineers are working on the code? Answering all these questions may lead to an answer whether conventional commits, either scope-based or specific keywords, or…
You can write a message without scopes or specific words. Take a look at the following commits:
c540107 * Check if memory is reallocated before deleting.
24694f8 * Add syscall to page documentation menu
Straightforward and readable.
Generating changelogs from conventional commits is not user-friendly or developer-friendly (because why would you create a changelog from your commits if the developer can look at the commit logs?). Instead, humanize the process of writing changelogs. If you cannot write humanized reports, you are not following what is going on within the system.
Conventional commits (specially scope-oriented) take too much message space, and they generally conflict depending on the message. Either you stick with conventional commits by carefully writing your commit messages, or you don’t use them at all.
Scope-oriented (i.e., component-related) is interesting, as a developer can quickly read the logs and understand what exactly is modified.
Finally, think through if conventional commits make sense for your project. What are you trying to archive with conventional commits? Commit organization? Trend? Changelog?