add subslice support in drop ladder for array #46686

mikhail-m1 · 2017-12-12T15:09:09Z

second step for fix #34708

rust-highfive · 2017-12-12T15:09:14Z

r? @petrochenkov

(rust_highfive has picked a reviewer for you, use r? to override)

mikhail-m1 · 2017-12-12T15:09:24Z

r? @arielb1

arielb1 · 2017-12-12T16:22:29Z

src/librustc_mir/borrow_check/mod.rs

@@ -1279,6 +1278,7 @@ impl<'cx, 'gcx, 'tcx> MirBorrowckCtxt<'cx, 'gcx, 'tcx> {
                    );
                }
            }
+            Write(WriteKind::Move) |


Why this change?

Why did you make this change.

arielb1 · 2017-12-13T16:26:03Z

src/librustc_mir/util/elaborate_drops.rs

-    fn clear_drop_flag(&mut self, location: Location, path: Self::Path, mode: DropFlagMode);
+    fn get_drop_flags(&mut self, path: Self::Path) -> Option<Operand<'tcx>>;
+    fn clear_drop_flag(&mut self,
+                       location: Location, path:


nit: path should go in the next line

location: Location, path: Self::Path

arielb1 · 2017-12-13T17:03:04Z

src/librustc_mir/transform/elaborate_drops.rs

+    fn array_subpaths(&self, path: Self::Path, size: u64)
+                      -> Vec<(Place<'tcx>, Option<Self::Path>, Option<Local>)> {
+        if dataflow::move_path_children_matching(self.ctxt.move_data(), path, |_| {
+                assert!(size <= (u32::MAX as u64),


can you move the assertion out of the loop?

arielb1 · 2017-12-13T17:03:44Z

src/librustc_mir/transform/elaborate_drops.rs

+        }
+
+        let size = size as u32;
+        let flags =  self.ctxt.array_items_drop_flags.get(&path);


nit: stray double space

flags = self.ctxt ^^

arielb1 · 2017-12-13T17:07:57Z

Do you handle the case where either individual indexes or subslices are moved out, aka:

fn subslice_pattern_test(a: &Allocator, arg: bool) {
    let a = [a.alloc(), a.alloc(), a.alloc(), a.alloc(), a.alloc()];
    if arg {
        let[.., _x, _] = a;
    } else {
        let[_, _y..] = a;
    }
}

arielb1 · 2017-12-13T17:10:13Z

src/librustc_mir/util/elaborate_drops.rs

@@ -88,14 +88,19 @@ pub trait DropElaborator<'a, 'tcx: 'a> : fmt::Debug {
    fn param_env(&self) -> ty::ParamEnv<'tcx>;

    fn drop_style(&self, path: Self::Path, mode: DropFlagMode) -> DropStyle;
-    fn get_drop_flag(&mut self, path: Self::Path) -> Option<Operand<'tcx>>;
-    fn clear_drop_flag(&mut self, location: Location, path: Self::Path, mode: DropFlagMode);
+    fn get_drop_flags(&mut self, path: Self::Path) -> Option<Operand<'tcx>>;


this still can't return multiple drop flags, so why did you rename it?

arielb1 · 2017-12-13T17:14:29Z

Another test:

fn subslice_pattern_test(a: &Allocator, arg: bool, arg2: bool) {
    let a = [a.alloc(), a.alloc(), a.alloc(), a.alloc(), a.alloc()];
    if arg2 {
        drop(a);
        return;
    }

    if arg {
        let[.., _x, _] = a;
    } else {
        let[_, _y..] = a;
    }
}

arielb1 · 2017-12-14T23:40:43Z

src/librustc_mir/transform/elaborate_drops.rs

@@ -389,6 +492,7 @@ impl<'b, 'tcx> ElaborateDropsCtxt<'b, 'tcx> {
                }
            };

+


nit: stray newline

arielb1 · 2017-12-14T23:47:03Z

src/librustc_mir/transform/elaborate_drops.rs

+                min_length: size,
+                from_end: false
+            }),
+            move_path,


So reading this, I'm afraid that this can cause a drop to be run with a "wrong" MovePath - a Subslice when the drop can also be accessed through a ConstantIndex (or vice versa), which might cause the "unconditional" drop code to trigger when it shouldn't, as in

let mut x = [a.alloc(), a.alloc(), a.alloc()]; // The MP for x[1..3] is now initialized, x[2] is now uninitialized let [_, _, _x] = x; // this shouldn't rely on any drop flags, and just drop x[0] & x[1] x = [a.alloc(), a.alloc(), a.alloc()]; let [_, .._x] = x; // The MP for x[1..3] is now uninitialized, x[2] is now initialized // end of scope drop for x - should drop x[0] without any drop flags

And in some cases you might need the drop flags:

let alloc = || (a.alloc(). a.alloc()); let mut x = [alloc(), alloc(), alloc()]; // The MP for x[1..3] is now initialized, x[2].0 is now uninitialized let [_, _, (_x,_)] = x; // this shouldn't rely on any drop flags, and just drop x[0], x[1], and x[2].1 x = [a.alloc(), a.alloc(), a.alloc()]; let [_, .._x] = x; // The MP for x[1..3] is now uninitialized, x[1].1 is now initialized // end of scope drop for x - should drop x[0] without any drop flags

So I think you do need to generate move paths for all the array elements when you see a subslice, to have a place to focus (similarly, you want to normalize the move paths so that an access to array[-2 of 3] is the same as an access to array[1 of 3])

At least, this is how it appears to me. There might be a better solution but it eludes me.

I'll go back to european time and no meetings in Monday (I'm currently in the US for the Mozilla All Hands), so if you need any help I'll go on IRC then.

nikomatsakis · 2017-12-20T10:21:08Z

src/librustc_mir/transform/elaborate_drops.rs

-        })
+    fn array_subpaths(&self, path: Self::Path, size: u64)
+                      -> Vec<(Place<'tcx>, Option<Self::Path>, Option<Local>)> {
+        if dataflow::move_path_children_matching(self.ctxt.move_data(), path,


Nit: speaking as someone who doesn't know this code that well, I'd appreciate some comments here. =)

I guess that this is just a "trivial case" check, however? i.e., if we never touched any of the indices in this array, then we there is nothing to do here?

arielb1 · 2017-12-25T20:45:52Z

I don't think I can finish handling this PR.

r? @nikomatsakis

arielb1 · 2017-12-25T20:46:13Z

@nikomatsakis could you try to get this working with @mikhail-m1 ?

nikomatsakis · 2018-01-03T19:29:18Z

@arielb1 @mikhail-m1 yep!

kennytm · 2018-01-10T13:05:48Z

Hi @mikhail-m1, just to check if you're still working on this. Looks like @arielb1 has some questions for you too!

mikhail-m1 · 2018-01-10T18:49:49Z

@kennytm Hi, @arielb1 showed me serious issue in my current solution, I will be back to finding new approach next week.

arielb1 · 2018-01-13T19:17:47Z

@mikhail-m1

Cool! I'll be sure to review your PR when you get it done.

shepmaster · 2018-01-26T17:39:49Z

Ping from triage @mikhail-m1! It's been over a week since we last heard from you — will you still be able to make the needed revisions?

mikhail-m1 · 2018-01-29T21:07:31Z

@shepmaster sorry, I am still working on it but don't have enough time to finish, hope I will do it in next few days

bors · 2018-01-29T22:47:27Z

☔ The latest upstream changes (presumably #47837) made this pull request unmergeable. Please resolve the merge conflicts.

@nikomatsakis

…omatsakis add transform for uniform array move out reworked second step for fix #34708 previous try #46686 r? @nikomatsakis

rust-highfive assigned petrochenkov Dec 12, 2017

rust-highfive assigned arielb1 and unassigned petrochenkov Dec 12, 2017

kennytm added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Dec 12, 2017

arielb1 reviewed Dec 12, 2017

View reviewed changes

add subslice support in drop ladder for array

2decde6

mikhail-m1 force-pushed the subslice_pattern_array_drop branch from 411c05d to 2decde6 Compare December 13, 2017 09:24

arielb1 reviewed Dec 13, 2017

View reviewed changes

arielb1 added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Dec 13, 2017

bugfix and add tests for elaborate array drop ladder

a4c7d75

arielb1 reviewed Dec 14, 2017

View reviewed changes

nikomatsakis reviewed Dec 20, 2017

View reviewed changes

rust-highfive assigned nikomatsakis and unassigned arielb1 Dec 25, 2017

aidanhs removed the S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. label Jan 17, 2018

aidanhs added the S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. label Jan 17, 2018

mikhail-m1 mentioned this pull request Feb 1, 2018

add transform for uniform array move out #47926

Merged

mikhail-m1 closed this Feb 1, 2018

bors added a commit that referenced this pull request Feb 17, 2018

Auto merge of #47926 - mikhail-m1:subslice_pattern_array_drop2, r=nik…

b85bd51

…omatsakis add transform for uniform array move out reworked second step for fix #34708 previous try #46686 r? @nikomatsakis

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add subslice support in drop ladder for array #46686

add subslice support in drop ladder for array #46686

mikhail-m1 commented Dec 12, 2017

rust-highfive commented Dec 12, 2017

mikhail-m1 commented Dec 12, 2017

arielb1 Dec 12, 2017

arielb1 Dec 12, 2017

arielb1 Dec 13, 2017

arielb1 Dec 13, 2017

arielb1 Dec 13, 2017

arielb1 Dec 13, 2017

arielb1 commented Dec 13, 2017

arielb1 Dec 13, 2017

arielb1 commented Dec 13, 2017

arielb1 Dec 14, 2017

arielb1 Dec 14, 2017 •

edited

Loading

arielb1 Dec 14, 2017

arielb1 Dec 14, 2017 •

edited

Loading

arielb1 Dec 15, 2017

nikomatsakis Dec 20, 2017

arielb1 commented Dec 25, 2017

arielb1 commented Dec 25, 2017

nikomatsakis commented Jan 3, 2018

kennytm commented Jan 10, 2018

mikhail-m1 commented Jan 10, 2018

arielb1 commented Jan 13, 2018 •

edited

Loading

shepmaster commented Jan 26, 2018

mikhail-m1 commented Jan 29, 2018

bors commented Jan 29, 2018

		@@ -389,6 +492,7 @@ impl<'b, 'tcx> ElaborateDropsCtxt<'b, 'tcx> {
		}
		};

add subslice support in drop ladder for array #46686

add subslice support in drop ladder for array #46686

Conversation

mikhail-m1 commented Dec 12, 2017

rust-highfive commented Dec 12, 2017

mikhail-m1 commented Dec 12, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

arielb1 commented Dec 13, 2017

Choose a reason for hiding this comment

arielb1 commented Dec 13, 2017

Choose a reason for hiding this comment

arielb1 Dec 14, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

arielb1 Dec 14, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

arielb1 commented Dec 25, 2017

arielb1 commented Dec 25, 2017

nikomatsakis commented Jan 3, 2018

kennytm commented Jan 10, 2018

mikhail-m1 commented Jan 10, 2018

arielb1 commented Jan 13, 2018 • edited Loading

shepmaster commented Jan 26, 2018

mikhail-m1 commented Jan 29, 2018

bors commented Jan 29, 2018

arielb1 Dec 14, 2017 •

edited

Loading

arielb1 Dec 14, 2017 •

edited

Loading

arielb1 commented Jan 13, 2018 •

edited

Loading