3 ms·

The pijul model makes even more sense to my brain, being patch based solves a lot of branching issues.

blagie 5y ago
One of the many breakthroughs in git was precisely not being patch based. Being patch-based means that the tools for comparing versions are hard-coded into the data format. One of the major upsides of CVS->SVN, for example, was support for file moves. Simple issues like this weren't possible to fix without reworking the whole system. With a version-based model, how you compare them is determined by the user space, whether that's merges, or viewing history. Even compression isn't all that fundamental to much of anything; it's possible to implement underneath the system, eliminating whatever redundancy you want. With patch-based systems, all of that comes baked into the underlying representation. On a deeper level, this feeds into how git can be distributed, robust, yet very simple.
- lillecarl 5y ago
  It also means cherry picking commits between branches prevents clean merging afterwards meaning you have to force yourself to keep branches short-lived. The patch graph pijul gives you doesn't have this issue. I'm not an expert, but when reading the Pijul blog it made a lot of sense for me.
  - blagie 5y ago
    No, that's mostly the fault of many bad decisions in the git user space. The data model is quite brilliant and quite general. It'd be possible to write a really good user space around the git data model if someone had the time and inclination. My personal gripes: * I like to commit every time I change a few lines of code. To have a clean version history, git requires me to rebase all of those commits. I'd like to have "major" and "minor" commits, which the git data model can support in several different ways (as a list, a tree, or DAG; I prefer the tree model). The user space makes anything like this incredibly painful. It encourages rewriting history, which is actually rather bad. * Large files are managed very, very badly. It's ironic since the git data model is ideal for things with large files. * Related, partial trees aren't really supported. This gives scaling issues. In many cases, I just want to work with a few files, or just the latest commit. Some of this is technically possible but practically painful enough to be not worthwhile. * Nested projects / submodules. Of those, only submodules might require changes to the underlying data model.

One of the many breakthroughs in git was precisely not being patch based. Being patch-based means that the tools for comparing versions are hard-coded into the data format. One of the major upsides of CVS->SVN, for example, was support for file moves. Simple issues like this weren't possible to fix without reworking the whole system. With a version-based model, how you compare them is determined by the user space, whether that's merges, or viewing history. Even compression isn't all that fundamental to much of anything; it's possible to implement underneath the system, eliminating whatever redundancy you want. With patch-based systems, all of that comes baked into the underlying representation. On a deeper level, this feeds into how git can be distributed, robust, yet very simple.

lillecarl 5y ago
It also means cherry picking commits between branches prevents clean merging afterwards meaning you have to force yourself to keep branches short-lived. The patch graph pijul gives you doesn't have this issue. I'm not an expert, but when reading the Pijul blog it made a lot of sense for me.
- blagie 5y ago
  No, that's mostly the fault of many bad decisions in the git user space. The data model is quite brilliant and quite general. It'd be possible to write a really good user space around the git data model if someone had the time and inclination. My personal gripes: * I like to commit every time I change a few lines of code. To have a clean version history, git requires me to rebase all of those commits. I'd like to have "major" and "minor" commits, which the git data model can support in several different ways (as a list, a tree, or DAG; I prefer the tree model). The user space makes anything like this incredibly painful. It encourages rewriting history, which is actually rather bad. * Large files are managed very, very badly. It's ironic since the git data model is ideal for things with large files. * Related, partial trees aren't really supported. This gives scaling issues. In many cases, I just want to work with a few files, or just the latest commit. Some of this is technically possible but practically painful enough to be not worthwhile. * Nested projects / submodules. Of those, only submodules might require changes to the underlying data model.