Coccinelle: Source-to-source transformation tool
github.com130 points by anon111332142 3 days ago
130 points by anon111332142 3 days ago
The best thing Julia Lawall ever did!
Not the hardest, not the thing with the most sophisticated theories behind it, not the thing that helped her academic career the most... but definitely the best and the most useful.
There must be a lot of other academics who could do things that are less theoretical but more useful than what they normally do.
There must be a lot of undervalued academics who in effect are punished for doing things that useful without requiring quite as much deep theory as their fields can muster.
I'm glad she did something that she wasn't really rewarded for and I'm sad that the academic reward functions are so off.
I can only agree. It is great work; I met Julia in several occasions were we other academics tried to push our formal methods stuff for checking properties of the Linux kernel. Also ours worked but in a way more complicated way, very resource intense, and less effective than Julia’s work.
Not the same level of sophistication, but ast-grep allows this for far more languages, since it is based on the tree-sitter parser library. I have used it with some success on C++. Of course it only works on the AST level, and C++ famously need types for correct parsing, so it sometimes fall short (also on macros).
I am working on AST level revision control and yes, macros make life difficult. On the other hand, merging/diffing on the AST level is fun.
I found that font extremely hard to read for some reason (on my phone). So I gave up. Maybe due to you using a monospace font for non-code?
But I believe smalltalk represented code as functions in a database somehow, so maybe that is worth looking at.
That is JetBrains Mono. HN is using Verdana, I believe, that one is recommended as the most usable common font.
Smalltalk is ancient. I would say, Unison is an interesting recent experiment, and there are others. But, I am interested in universal revision control, any language.
I think Coccinelle is a really cool tool, but I find its documentation totally incomprehensible for some reason. I've read through it multiple times, but I always end up having to find some preexisting script that does what I want, or else to blunder around trying different variations at random until something works, which is frustrating.
It's a bit of a disservice to call it "The Linux kernel's"; it's its own project that just happens to be used on the Linux kernel quite a bit. It doesn't originate there or belong to the kernel or anything like that.
It’s not only great to fix issues and backport at scale, it’s also very useful to do patches that work universally across kernels no matter how badly modified the sources are.
One example is that there are patches for Android kernels to allow rooting the phone with maximum discretion in order to bypass integrity checks. Unfortunately manufacturers modify Linux sources so sloppily that diffs never apply cleanly and you have to endlessly fix hunk rejects.
I created Coccinelle patches that can universally apply to any Linux kernel between 3.18 and 5.15. It has been tested over 100+ Linux kernel sources from Android phone manufacturers and works flawlessly! https://github.com/devnoname120/kernelsu-coccinelle
The learning curve is steep (documentation is thin so you have to look at the code), but the rewards are well worth it. I compiled a list of conferences, slides, and examples to get started if anyone is interested: https://github.com/devnoname120/kernelsu-coccinelle#how-to-l...
According to https://coccinelle.gitlabpages.inria.fr/website/ce.html :
> Nevertheless, detecting the holding of locks requires a careful and occasionally interprocedural analysis of the source code, and the other conditions, such as "in a completion handler", are not formally defined and require study of multiple files.
> Due to the complexity of the conditions governing the choice of new argument for usb_submit_urb, 71 of the 158 calls to this function were initially transformed incorrectly to use GFP_KERNEL instead of GFP_ATOMIC.
Okay, but how does Coccinelle help? Is it able to do this careful and not formally defined analysis? Or does it automate the undifferentiated heavy lifting and so make it easier for humans to do it?
I forgot about Coccinelle.
I think semantic patching is an idea whose time has come though. I'm making a more modern set of tools for source-to-source transformation that will work with any desired languages as the input and output.
Those tools exist, but you have to pay by the token. I'm not sure if they scale financially to large code bases such as the Linux kernel. They are far more accessible than Coccinelle or Perl, though.
Honestly, I rather use Coccinelle, where I understand exactly what it does, when it does it and why it does it…
I would also rather use a tool that I trust than delegate the task to unreliable third party.
But to the person bringing up AI, you don't have to choose one or the other! Models use tools. Good tools for people are usually also good tools for models. The problem models have in learning to use tools like Coccinelle effectively is that there are too many of the tools and not enough documentation for each tool. If there were a unified, standard platform however then many humans would start to gain abilities through fluent tool use and of enough of those people would write docs and blog posts. Where people lead, models follow without doubt. Once a large enough corpus of writing existed documenting a single platform the models would also be fluent, just like they are fluent in JS and React because of how large the web platform is
What an interesting and very apt choice of name!
Edit: I feel like it needs the addition that Coccinelle was the stage name of Jacqueline Charlotte Dufresnoy (1931), a French performer and probably the first celebrity who underwent a transition from male to female.
There is also https://sparse.docs.kernel.org/ that can be used to help refactoring C code.
See also OpenRewrite:
https://github.com/openrewrite/rewrite
And i assume any large organisation running a monorepo has some vaguely equivalent tooling for making mass changes. Have any of them published about that?
You can write automated refactoring with clang tools if you need AST-level knowledge across your project (or monorepo).
I’m not sure if there’s other public examples leveraging this, but Chromium has this document [0] which has a few examples. And there’s also the clang-tidy docs [1].
[0] https://chromium.googlesource.com/chromium/src/+/80a6fc33dee...
[1] https://releases.llvm.org/21.1.0/tools/clang/tools/extra/doc...
This is a business that I suspect may not survive BABLR.
> Moderne's build plugins allow for LSTs to be serialized to disk. This makes the process of consuming and editing large quantities of them much more efficient. OpenRewrite's build plugins, on the other hand, store everything in memory and need to be reparsed every time there is a change.
So yeah I'm giving away open standards to everyone for free that do the thing they expect people to pay them for...
What's BABLR?
> The next-gen LR parser framework for creating elegant and efficient language tools
> BABLR is a new kind of thing that does not quite fit into any category of things that has existed before it. In purpose it is made to be an instrument of code literacy -- a unified toolchain for software developers that supports a new generation of richly visual interfaces for coding. In form BABLR is a collection of scripts and virtual machines written in plain Javascript that run in almost any modern web browser. BABLR is also a community and an ecosystem, including a small but rapidly growing collection of ready-to-use parsers for popular languages.
At first brush, everything about this sounds like overly ambitious vapourware. Is there a reason to think this is going to deliver? People involved, what's already shipped, etc?
I particularly loved this from their roadmap:
> Completed
> Shift operation
> Enables LR parsing of expressions like 2+2
Being able to parse 2 + 2 is definitely good!
And their thoughts on testing:
> How our project reaches production stability is a process that often surprises people. We don't write a lot of tests for example, and we often don't do much testing before we ship releases. Instead we test exhaustively after we ship releases, which is the only way we know of knowing for sure that the product we shipped does what we think it does. [...] We also don't (usually) practice TDD. If you look at the number of tests we have, it likely won't seem like it's anywhere near enough to keep a project of this size stable! The secret sauce here is that our key invariants aren't written in our test files, they're baked into the core of the implementation. Every time you use the code, you're essentially testing it. To gain confidence in our core, we simply try to use it to do a lot of real work.
Man, why did i not think of that, i could have got out of writing so many tests if i'd just baked the invariants into the core of the implementation!
In this case the tool is meant to parse programming languages, so once I write some parser grammars every valid code file in existence is a test case. Seen that way I have more test cases than I know what to do with.
We've come a ways from 2 + 2. This week my goal is to feed our own whole codebase through the JS parser, and I should be able to. I managed to parse a few hundred lines of real JS last week before running into Automatic Semicolon Insertion trouble that I needed to tinker with the core to fix.
While I get that our low profile smacks of vapor, we actually have working packages published: bablr and @bablr/cli. I'd consider them to be beta quality right now, having gone through many previous releases that I'd only consider alpha-quality, and even more releases before that.
It's not too hard to verify my central claim here which is that we're giving away what they charge money for. Their serialization format is secret, proprietary. Ours, CSTML, is open: https://docs.bablr.org/guides/cstml. Their free product make you re-parse the entire project with every code change you make. Ours is built with copy-on-write immutable data structures so that you can always build new things without losing old ones. Our way you can compose fragments of trees together with new code into new trees like you're playing with lego bricks.
coccinelle's one of those tools that's stupid powerful once it clicks, but man the learning curve is steep. i used it to migrate like ~200 call sites when we tweaked an internal API signature in a big C codebase - doing that by hand wouldve been a multi-day slog. the semantic patch language feels kinda weird at first, but it catches edge cases regex stuff just misses, like matching through macro expansions and all that
I thought this was a misspelled article about Kokinelli, the Greek red wine, fairly accurately described here: https://www.arrse.co.uk/wiki/Kokinelli
I used to drink this stuff back in the late 1960s, when my Dad was an RAF pilot based in Cyprus and I was about 15. You had to take it with a Sprite mixer if you wanted to retain your teeth.
It would be a good name for a project, though.
They are cognate: cochineal is a red dye derived from a red insect a little bit like a ladybug.