Sem – Semantic version control. Entity-level diffs on top of Git | Hacker News Reader

Sem – Semantic version control. Entity-level diffs on top of Git

github.com

95 pointspabs324 comments

Anyone here using semantic diffing tools in their daily work? How good are they?

I use some for e.g. YAML [0] and JSON [1], and they're nice [2], but these are comparatively simple languages.

I'm particularly curious because just plain diffing ASTs is more on the "syntax-aware diffing" side rather than the "semantic diffing" side, yet most semantic tooling descriptions stop at saying they use ASTs.

ASTs are not necessarily in a minimal / optimized form by construction I believe, so I'm pretty sure you'll have situations where a "semantic" differ will report a difference, whereas a compiler would still compile the given translation unit to the same machine bytecode after all the optimization passes during later levels. Not even necessarily for target platform dependent reasons.

But maybe this doesn't matter much or would be more confusing than helpful?

[0] dyff: https://github.com/homeport/dyff

[1] jd: https://github.com/josephburnett/jd

[2] they allow me to ignore ordering differences within arrays (arrays are ordered in YAML and JSON as per the standard), which I found to be a surprisingly rare and useful capability; the programs that consume the YAMLs and JSONs I use these on are not sensitive to these ordering differences

Fair point on AST vs semantic. sem sits somewhere in between. It doesn't go as far as checking compiled output equivalence, but it does normalize the AST before hashing (we call it structural_hash), so purely cosmetic changes like reformatting or renaming a local variable won't show as a diff. The goal isn't "would the compiler produce the same binary" but "did the developer change the behavior of this entity." For most practical cases that's the useful boundary. The YAML/JSON ordering point is interesting, we handle JSON keys as entities so reordering doesn't conflict during merges.

By the way creator here.

Hey there, thanks for checking in.

Regarding the custom normalization step, that makes sense, and I don't really have much more to add either. Looked into it a bit further since, it seems that specifically with programming languages the topic gets pretty gnarly pretty quick for various language theory reasons. So the solution you settled on is understandable. I might spend some time comparing how various semantic toolings compare, I'd imagine they probably aim for something similar.

> The YAML/JSON ordering point is interesting, we handle JSON keys as entities so reordering doesn't conflict during merges.

Just to clarify, I specifically meant the ordering of elements within arrays, not the ordering of keys within an object. The order of keys in an object is relaxed as per the spec, so normalizing across that is correct behavior. What I'm doing with these other tools is technically a spec violation, but since I know that downstream tooling is explicitly order invariant, it all still works out and helps a ton. It's pretty ironic too, I usually hammer on about not liking there being options, but in this case an option is exactly the right way to go about this; you would not want this as a default.

Ah right, array ordering not key ordering. That's a different beast. You're making a deliberate semantic choice because you know your consumers are order-invariant. We can't really do that at our level since function ordering in code is usually meaningful to the language. Your use case needs domain knowledge about the consumer, which is exactly why an option makes sense there.

If you do end up comparing semantic toolings I'd love to hear what you find. The space is weirdly fragmented between syntax-aware, normalized-AST, and domain-specific (dyff/jd). Everyone calls it "semantic" but they're solving pretty different problems.

I guess I don't understand the difference between semantic and syntax-aware, but I've been trying out difftastic which is a bit of an odd beast but does a great job at narrowing down diffs to the actual meaningful parts.

difftastic is solid. The difference is roughly: syntax-aware (difftastic) knows what changed in the tree, sem knows which entity changed and whether it actually matters. difftastic will show you that a node in the AST moved. sem will tell you "the function processOrder was modified, and 3 other functions across 2 files depend on it." difftastic is a better diff. sem is trying to be a different layer on top of git entirely.

Was "sem" named "graft" last week and "got" a week before that? Everyone is vibing so hard it is difficult to keep track of things. Also, idea theft gets to entirely new levels. Bot swarms promote 100% vibed stolen projects... what a moment in time we all enjoy.

Still, my two cents: Beagle the AST-level version control system, experimental

https://github.com/gritzko/librdx/tree/master/be#readme

It genuinely stores AST trees in (virtually any) key-value database (RocksDB at the moment). In fact, it is a versioned database for the code with very open format and complete freedom to build on top of it.

be is in fact, more of a format/protocol, like in the good old days (HTTP, SMTP, XML, JSON - remember those?)

Different project, same author. sem has been sem since the first commit. Beagle looks interesting, storing ASTs directly in a KV store is a different approach. sem stays on top of git so there's zero migration cost, you keep your existing repo and workflows.

I like this but would think "Semantic" is pushing it a bit as it relies on function names and changes therein having a direct mapping to meaning with standard text processing.

In fact, I fully expected a use of LLM to derive those meaningful descriptions before I checked the repo.

Anyway I definitely see this as a useful thing to try out as a potential addition to the armoury, but as we go further along the route to AI-coding I expect the diffs to be abstracted even further (derived using AI), for use by both agentic and human contributors.

You're right that "semantic" is doing some heavy lifting in the name. We use it to mean "understands code structure" rather than "understands code meaning." sem knows that something like "validateToken" is a function and tracks it as an entity across versions, but it doesn't know what validation means. For the merge use case (weave - https://ataraxy-labs.github.io/weave/), that level of understanding is enough to resolve 100% of false conflicts on our benchmarks. LLM-powered semantic understanding is the next layer, and that's what our review tool (inspect) does, it uses sem's entity extraction to triage what an LLM should look at.

> I like this but would think "Semantic" is pushing it a bit

It would be nice to get to the feature level, meaning across files/classes/functions etc.

sem already does this. sem graph builds a cross-file entity dependency graph and sem impact tells you "if this function changes, these other entities across these many files are affected." It's transitive too, follows the full call chain.

Same mental model problem comes up in AI agent observability. Two conversation flows can produce identical user outcomes and look totally different at the message level, or vice versa. The normalization step that actually captures 'did behavior change' is the hard part in both domains.

That's a really sharp parallel. "Did behavior change" is exactly the question in both cases, and the surface-level representation lies to you in both. We normalize ASTs before hashing so reformatting or renaming a local variable doesn't register as a change. Curious what normalization looks like on the agent observability side, feels like a harder problem when the output is natural language instead of code.

I was just building a POC of something like this a couple of weeks ago. I'm glad someone else already implemented it with support for more languages.

Great, did you opensource it?

cool! Line-level diffs only made sense when humans wrote every line. When AI agents are committing hundreds of changes a session, you need to know which functions changed, not which lines..

Line-level diffs didn't even make sense when humans wrote every line. It's just combining two different things—text-based files and formal languages—in ways that were never going to make sense. We've all wasted days of our lives fixing crappy diffs because nobody bothered to develop something that made sense for the files we were tracking outside of "everything is text who cares text is magic"

Still, some languages/heuristics handle this better than others.

Exactly. The only reason line-level diffs survived this long is that text is the lowest common denominator. Once you have fast enough parsers (tree-sitter parses most files in under 1ms), there's no reason to stay at the line level.

Yeah, as long as the function gets called, it’ll work!

_sad panda face_

Cool, knowledge graph matching on top of commits

Pretty much, yeah. sem graph builds a cross-file entity dependency graph and sem impact does transitive traversal on it. So you can ask "if I change this function, what breaks?" and get a deterministic answer without sending anything to an LLM.

Please could you add cargo binstall support?

Thanks for the request, Yeah I can work on this.

Reply on news.ycombinator.com