X-Git-Url: https://git.ralfj.de/web.git/blobdiff_plain/6d624757438133582f69cf27730f773f316a1ae1..e806a32d3aa3ce53166c46b8eab5d77a9655c251:/ralf/_posts/2017-08-11-types-as-contracts-evaluation.md diff --git a/ralf/_posts/2017-08-11-types-as-contracts-evaluation.md b/ralf/_posts/2017-08-11-types-as-contracts-evaluation.md index 5f0c07a..68318a6 100644 --- a/ralf/_posts/2017-08-11-types-as-contracts-evaluation.md +++ b/ralf/_posts/2017-08-11-types-as-contracts-evaluation.md @@ -4,7 +4,7 @@ categories: internship rust forum: https://internals.rust-lang.org/t/https-www-ralfj-de-blog-2017-08-11-types-as-contracts-evaluation-html/5753 --- -Some weeks ago, I described [Types as Contracts]({{ site.baseurl }}{% post_url 2017-07-17-types-as-contracts %}) as an approach for how to go about defining Rust's aliasing-related [undefined behavior]({{ site.baseurl }}{% post_url 2017-07-14-undefined-behavior %}). +Some weeks ago, I described [Types as Contracts]({% post_url 2017-07-17-types-as-contracts %}) as an approach for how to go about defining Rust's aliasing-related [undefined behavior]({% post_url 2017-07-14-undefined-behavior %}). One key property of this approach is that it is *executable*, which means we can actually have a program (in this case, [miri](https://github.com/solson/miri/)) tell us whether some particular Rust test case exhibits undefined behavior or not. I have since then spent most of my time completing and refining this implementation, and running it on miri's test suite to exercise various bits of the standard library and see whether they are actually following the rules I have been suggesting. @@ -19,9 +19,9 @@ Finally, I discuss some of the things that I would like to look at going forward Unsurprisingly, some of miri's test cases started to fail once I properly implemented the rules I wanted to have checked. Some of these turned out to be [compiler](https://github.com/rust-lang/rust/issues/43457) [bugs](https://github.com/rust-lang/rust/issues/43481), but others are actually examples of code violating the proposed rules. In the following, I describe the patterns that I found. -I assume that you are familiar with Types as Contracts as introduced in my [previous post]({{ site.baseurl }}{% post_url 2017-07-17-types-as-contracts %}). +I assume that you are familiar with Types as Contracts as introduced in my [previous post]({% post_url 2017-07-17-types-as-contracts %}). -## 1.1 Ownership Passing Via Raw Pointers +### 1.1 Ownership Passing Via Raw Pointers I already mentioned this in the original post: `mem::swap` triggers a lock violation in the strictest interpretation of the model. {% highlight rust %} @@ -69,12 +69,12 @@ However, this again overlaps with `self`. This list is not exhaustive by any means; there are problems of this sort pretty much whenever raw pointers are used in non-trivial ways. -## 1.2 Uninitialized memory +### 1.2 Uninitialized memory Any place that calls `mem::uninitialized` instantly has validation blow up because the returned `T` is, obviously, not actually a valid `T` -- it is, after all, completely uninitialized. However, it turns out that `mem::uninitialized` is causing other problems as well (e.g. around uninhabited types), and [better APIs for handling uninitialized data have already been proposed](https://internals.rust-lang.org/t/mem-uninitialized-and-trap-representations/4167/18?u=ralfjung). -## 1.3 `Arc::drop` +### 1.3 `Arc::drop` This is the most interesting case I found. The `Arc` destructor ends up calling `Arc::drop_slow`, which is implemented as follows: {% highlight rust %} @@ -140,7 +140,7 @@ In particular, it turned out that suspensions were actually not modeling borrowi There were also some changes affecting the rules of where validation calls are emitted in some corner cases, but there weren't really any interesting things to learn there. At some point I will write down the exact rules in full detail; until then it is really the general approach that matters much more than such corner cases. -## 2.1 Suspension of Write Locks +### 2.1 Suspension of Write Locks In the original proposal, on every borrow `y = &[mut] x;`, the write locks on `x` would be *suspended*. This means that the locks get released, but will be re-acquired (or *recovered*) automatically when the lifetime of the borrow ends. @@ -200,7 +200,7 @@ Notice that in the above example, both `x_inner` and `r` have the *same* lifetim So, I will have to go back to the drawing board and figure out a different way to identify write locks. I have some ideas, but I am not very happy with them. -## 2.2 Unsafe Code +### 2.2 Unsafe Code As already indicated in the original post, the way I suggest to handle the problems with raw pointers described in [§1.1](#11-ownership-passing-via-raw-pointers) is by relaxing the validation performed around unsafe code. Specifically, the code that emits MIR validation statements now detects if a function is unsafe, contains an unsafe block, or (in the case of closures) is defined inside an unsafe block/function. @@ -219,7 +219,7 @@ In discussions, two possible solutions to this problems became apparent: Both of these options are certainly worth exploring. -# 3 Testing Your Own Code +## 3 Testing Your Own Code The implementation of the model as described above (well, with the open problems mentioned in [§2.1](21-suspension-of-write-locks)) actually landed upstream in rustc and miri, so you can play around with this yourself! To do so, you need to install Rust nightly and miri. @@ -242,7 +242,7 @@ If you find anything odd and things don't work as expected, please [report a bug Notice however that miri currently supports virtually no interaction with the surrounding operating system, like file access or spawning threads. -# 4 Future Work +## 4 Future Work As usual, there are so many things that could be done next. @@ -260,12 +260,12 @@ Interestingly, the [discussion](https://internals.rust-lang.org/t/types-as-contr After all, we want miri to also detect UB related to e.g. illegal pointer arithmetic. These issues are far from settled, and I do have some interesting ideas here that I want so share with you. -Finally, there is of course the discussion around how to handle raw pointers; I already talked in [§3](#22-unsafe-code) about some of the options that could be explored here. +Finally, there is of course the discussion around how to handle raw pointers; I already talked in [§2.2](#22-unsafe-code) about some of the options that could be explored here. Unfortunately, today is the last day of my internship, so I will no longer be able to work on all of this full-time like I did the last three months. Still, I certainly intend to stay involved. This problem is way too interesting to just let it go :) -As always, please comment with your thoughts on the topic. +As always, please [comment](https://internals.rust-lang.org/t/https-www-ralfj-de-blog-2017-08-11-types-as-contracts-evaluation-html/5753) with your thoughts on the topic. I am particularly curious about what kind of test cases you are throwing at miri, and how it is doing! -**Update**: I added a proposal for how to fix the `Arc` problem. +**Update:** I added a proposal for how to fix the `Arc` problem. **/Update**