The proptest_derive crate & #[derive(Arbitrary)] #79

Centril · 2018-07-31T21:42:43Z

This PR adds the crate proptest_derive to this repo.

The advantage of not having this in a separate repo is that I expect both crates will be easier to maintain and test if they are in a single place. Another crate that does this is serde which sports its serde_derive crate here.

The PR is pretty massive in size (sorry about that) but it is 100% consisting of additions and no code is touched outside of the proptest-derive directory. I recommend checking out my branch and starting from the tests directory to see what is supported and what is explicitly rejected.

I'm sure there are places where the code can be refactored and improved, but I'd like to avoid making too many changes to this PR. We can of course make design changes and code improvements later.

Some things TODO after merging this PR (to a separate branch, not master):

Setup a workspace. Ideally, all the code for the proptest crate should go in its own directory in the root and proptest-derive should also be in the root.
Make sure proptest-derive is tested with Travis CI and AppVeyor CI. For now, I have not made any changes to .travis.yml but there are plenty of tests (+50% of the code). These should all be run with a nightly compiler (because compiletest_rs wants it for pretty tests and because attr_literals is unstable). Until we've ensured that tests are run, it would be inadvisable to make breaking changes to proptest itself.
I have to give you perms on crates.io so you can publish updates to proptest-derive at the same time as proptest.
Release a 0.1 of proptest-derive.

Here are some TODOs into the future:

We should figure out what to do with self-recursive and mutually-recursive types. These are not supported at all and will blow the stack if attempted.
Write extensive documentation (ideally in the form of an mdbook (see Idea: Making a book and splashy website about proptest #36)). I haven't had the time to do that yet but I'll start on it ASAP.
Improve error messages with correct spans and such things.
Eventually use the custom error message API in https://internals.rust-lang.org/t/custom-error-diagnostics-with-procedural-macros-on-almost-stable-rust/8113 to segregate errors by warnings and so on...

…ature/initial-proptest-derive

AltSysrq · 2018-08-09T03:25:46Z

Sorry for my delayed response here. I'll see if I can work through this this weekend.

Some things TODO after merging this PR

All sound like a good plan to me.

Centril · 2018-08-09T03:45:49Z

No problem ❤️ Take your time! :)

AltSysrq · 2018-08-20T03:39:14Z

proptest-derive/src/ast.rs

+/// strategies and then switch to nesting like so:
+///
+/// ```ignore
+/// (1, 2, 3, 4, 5, 6, 7, 8, 9,


This nesting isn't going to produce a uniform distribution. This is why prop_oneof! falls back to boxing everything so it can use a normal Union if it's over the limit.

Oh, I see you're doing something with weights below.

Yeah, the idea is that:

prop_oneof![ w1 => e1, w2 => e2, w3 => e3 ]

can be rewritten in terms of:

prop_oneof![ w1 => e1, (w2 + w3) => prop_oneof![ w2 => e2, w3 => e3 ] ]

and it commutes, so you can also say:

prop_oneof![ (w1 + w2) => prop_oneof![ w1 => e1, w2 => e2 ], w3 => e3 ]

This is then applied for larger groupings (the max number that TupleUnion supports - 1).

In practice this would differ from the distribution you get when using Union + boxing because you are sampling the PRNG more times; but it seems to be that the cost of boxing every part of the union outweighs this slight discrepancy so I'd like to avoid the boxing.

AltSysrq · 2018-08-20T03:51:47Z

proptest-derive/tests/params.rs

+#[derive(Debug, Arbitrary)]
+#[proptest(params(u64))]
+struct TPIS {
+    #[proptest(strategy = "\"a+\"")]


How hard would it be to have a #[proptest(regex = "string")] for String and Vec<[u8]> fields? It would make syntax like this nicer

Not very hard I think; you could do this by adding another attribute regex to the parser and then desugaring that to the same internals as strategy uses in the attribute parser.

I can do this in a follow up PR if you like :)

(I agree #[proptest(strategy = "\"a+\"")] looks suboptimal)

AltSysrq · 2018-08-20T03:57:42Z

These [tests] should all be run with a nightly compiler

Does the plugin itself work with stable?

I've made a pass through the code and it initially looks pretty good, though there's a lot I don't understand since this stuff is extremely new to me. I'll try to work through it in more detail this week.

Centril · 2018-08-20T08:34:22Z

Does the plugin itself work with stable?

It should; I think I have tested it on stable a few times, but I don't remember exactly when the last time was, so it is possible I let some mistake slip through.

To make sure everything beside attr_literal (which is getting stabilized soon, rust-lang/rust#53044) tests work on stable, I think it would be a good idea to follow this PR up with one that gates the nightly-only tests behind an unstable cargo feature. We can then use the same testing mechanism we use for stable/unstable in proptest proper.

Centril · 2018-09-11T15:42:17Z

Did you get a chance to review any more of this?

AltSysrq · 2018-09-16T14:02:43Z

I haven't. Sorry about that and how unresponsive I've been in general lately. I should have some time soon.

Centril · 2018-10-21T14:30:26Z

Hey :)

I'd love to make improvements to the proptest_derive crate (for example upgrade the syn version and other changes) but I'd also like to avoid making too many changes to allow you to review the crate more easily.

Any chance you could review this? If you think it's unlikely that you'll get to it in the foreseeable future, I can publish this out of tree, but that will of course make dealing with breaking chances more difficult.

AltSysrq · 2018-10-25T23:54:26Z

I'm going to set myself a hard commitment this time so I stop letting this slide: I will review and likely merge this on Saturday.

AltSysrq · 2018-10-27T14:38:24Z

proptest-derive/src/error.rs

+            "[proptest_derive, ", stringify!($code), "]",
+            " during #[derive(Arbitrary)]:\n",
+            $msg,
+            " Please see: https://PATH/TO/foo#", stringify!($code),


We'll need to update this once the location of the docs are selected. (Just making a note so this is easier to see later.)

Nicely spotted; I might have forgotten about this otherwise. =P

AltSysrq · 2018-10-27T14:47:31Z

Merging this into the proptest-derive branch. I'll get the workspace and CI set up after that, and if you don't already have docs written somewhere, I'll start on that too.

AltSysrq · 2018-10-27T14:49:13Z

Is there a reason some of the .rs files have the execute bit set?

...
 create mode 100644 proptest-derive/tests/compile-fail/E0026-strategy-malformed.rs
 create mode 100755 proptest-derive/tests/compile-fail/E0027-filter-malformed-expr.rs
 create mode 100755 proptest-derive/tests/compile-fail/E0027-filter-malformed.rs
 create mode 100755 proptest-derive/tests/compile-fail/E0028-skipped-variant-has-filter.rs
 create mode 100644 proptest-derive/tests/compile-fail/E0028-skipped-variant-has-param.rs
 create mode 100644 proptest-derive/tests/compile-fail/E0028-skipped-variant-has-strat.rs
...

Centril · 2018-10-27T14:51:05Z

@AltSysrq probably just my editor being silly (using Windows can be painful...). Try running the tests without them and change to 644 or something if the tests all pass... :)

if you don't already have docs written somewhere, I'll start on that too.

No sorry, never got around to writing more docs but I'd like to help out; we can take some inspiration from serde.rs; an mdbook might be fitting here...

AltSysrq · 2018-10-27T16:26:26Z

@Centril any idea about this error off hand? https://travis-ci.org/AltSysrq/proptest/jobs/447128611#L2502

The rustc command line above doesn't appear to be passing in any crates at all and I don't see anything obvious that would control that.

AltSysrq · 2018-10-27T16:38:07Z

Never mind, figured it out

Centril · 2018-10-27T16:38:18Z

@AltSysrq Hmm... this happened to me sometimes when I was using the wrong path when building locally; so my guess would be one of: a) that the target/ directory ends up in a wrong place, b) compile_test looks in the wrong directory, c) the stuff needed are not in the right directory (configured in https://github.com/AltSysrq/proptest/blob/proptest-derive/proptest-derive/tests/compiletest.rs). It's like a) or b)...

Does this build locally with the branch tested or do you also get errors with that?

I'm about to sleep now, so I'll check it out in the morning if you didn't solve it before :)
...lol :D

Centril · 2018-10-27T16:38:39Z

@AltSysrq so... what was the problem? =P

AltSysrq · 2018-10-27T16:39:53Z

353b52f

target is shared by the whole workspace so the directory got moved when I set that up.

Centril · 2018-10-27T16:40:54Z

Aaah; that makes a lot of sense. :)

Centril added 30 commits June 15, 2018 10:00

add WIP proptest-derive logic.

0207841

simplify use of error! macro.

7085b45

add Ctxt based on serde.

43eae2b

derive: get rid of panics, use Result + compile_error! instead.

76530fa

proptest-derive: refactor, improve, test stuff.

49558fc

Merge branch 'master' of https://github.com/AltSysrq/proptest into fe…

5f79eaf

…ature/initial-proptest-derive

proptest-derive: cleanup and use PAT: TY in proptest!(..).

d0572ec

proptest-derive: reduce rightward drift.

c5fed1f

fix bug in derive and make error_if_set non-fatal.

4551dab

make unknown modifiers, E0018 non-fatal.

d85bfbb

make more errors non-fatal.

567de47

handle case when attr.interpret_meta() fails.

b388b15

make E0031 non-fatal.

0a600d7

proptest-derive: make tests more rigorous.

ed837d1

proptest-derive: make E0008 + E0030 non-fatal.

6360731

proptest-derive: make E0009 non-fatal.

b516e78

proptest-derive: make E0007 non-fatal.

e5eaa73

proptest-derive: make E0021 non-fatal.

86569ee

proptest-derive: make E0023 non-fatal.

dbcce17

proptest-derive: make E0026 non-fatal + cleanup.

01aaaa1

proptest-derive: cleanup.

e702e1a

proptest-derive: make E0029 non-fatal.

b31b0b2

proptest-derive: make E0029 non-fatal.

bffd757

proptest-derive: make E0028 non-fatal.

6da9412

proptest-derive: make E0028 non-fatal.

4ef665e

proptest-derive: make E0028 non-fatal.

e68e46b

proptest-derive: make E0006 non-fatal.

7a8f1d4

proptest-derive: reorder non-fatal and fatal errors.

7f64826

proptest-derive: reorder and make E0003 non-fatal.

6da678c

proptest-derive: make E0001 non-fatal.

4ec37fd

Centril added 7 commits July 31, 2018 20:06

clear a TODO.

dd27aa1

cleanup and clear todos.

0478507

adjust copyright headers.

06d27bf

adjust more copyright headers.

96503d2

adjust more copyright headers.

ae2b859

mv proptest-derive2 -> proptest-derive

31eaf4f

update readme and cargo.toml.

769fa7c

AltSysrq reviewed Aug 20, 2018

View reviewed changes

AltSysrq reviewed Oct 27, 2018

View reviewed changes

AltSysrq changed the base branch from master to proptest-derive October 27, 2018 14:41

AltSysrq merged commit a13212c into proptest-rs:proptest-derive Oct 27, 2018

Centril deleted the feature/initial-proptest-derive branch June 25, 2019 06:36

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

The proptest_derive crate & #[derive(Arbitrary)] #79

The proptest_derive crate & #[derive(Arbitrary)] #79

Centril commented Jul 31, 2018 •

edited

Loading

AltSysrq commented Aug 9, 2018

Centril commented Aug 9, 2018

AltSysrq Aug 20, 2018

AltSysrq Aug 20, 2018

Centril Aug 20, 2018

AltSysrq Aug 20, 2018

Centril Aug 20, 2018

AltSysrq commented Aug 20, 2018

Centril commented Aug 20, 2018

Centril commented Sep 11, 2018

AltSysrq commented Sep 16, 2018

Centril commented Oct 21, 2018

AltSysrq commented Oct 25, 2018

AltSysrq Oct 27, 2018

Centril Oct 27, 2018

AltSysrq commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

Centril commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

The proptest_derive crate & #[derive(Arbitrary)] #79

The proptest_derive crate & #[derive(Arbitrary)] #79

Conversation

Centril commented Jul 31, 2018 • edited Loading

AltSysrq commented Aug 9, 2018

Centril commented Aug 9, 2018

AltSysrq Aug 20, 2018

Choose a reason for hiding this comment

AltSysrq Aug 20, 2018

Choose a reason for hiding this comment

Centril Aug 20, 2018

Choose a reason for hiding this comment

AltSysrq Aug 20, 2018

Choose a reason for hiding this comment

Centril Aug 20, 2018

Choose a reason for hiding this comment

AltSysrq commented Aug 20, 2018

Centril commented Aug 20, 2018

Centril commented Sep 11, 2018

AltSysrq commented Sep 16, 2018

Centril commented Oct 21, 2018

AltSysrq commented Oct 25, 2018

AltSysrq Oct 27, 2018

Choose a reason for hiding this comment

Centril Oct 27, 2018

Choose a reason for hiding this comment

AltSysrq commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

Centril commented Oct 27, 2018

AltSysrq commented Oct 27, 2018

Centril commented Oct 27, 2018

Centril commented Jul 31, 2018 •

edited

Loading