rooting of boxes in borrowck fails to take moves into account

[nikomatsakis]

The following code:

// exec-env:RUST_POISON_ON_FREE=1

fn free<T>(-x: @T) {}

fn lend(+x: @{f: @{g: int}}) -> int {
    let y = &x.f.g;
    free(x);
    *y
}

fn main() {
    assert lend(@{f: @{g: 22}}) == 22;
}

is handled incorrectly by borrowck right now. It will opt not to root the x.f box even though it ought to. This is because it sees that x.f is found in immutable data which will outlive the lifetime of the borrow: but it doesn't consider moves. To be correct, it should either (1) know which variables might be moved (liveness can tell us) and conservatively root boxes that are found in those variables or (2) take out a loan on x so that an error would be reported. But (2) seems suboptimal since this is not really an error, just a failure of borrowck to make the right call on whether to root x.f.

[nikomatsakis]

I have a fix for this but it will not make it to 0.6

[Aatch]

@nikomatsakis is this fixed? There have been a lot of borrowck changes recently.

Also, that example is so old I'm not even sure how to modernize it...

[catamorphism]

@nikomatsakis , what's the status of this?

[nikomatsakis]

I think I fixed this :) let me see if I added a test

[nikomatsakis]

Just added a PR with a test, since I didn't see one. I violated rule #1 of test writing though in that I did not observe a test failure; it'd be a pain to do so, but I did inspect code and generated LLVM to make sure that we're creating the right pattern here.

[nikomatsakis]

for reference, here is the code that attempts to fix this scenario: https://github.com/mozilla/rust/blob/master/src/librustc/middle/borrowck/gather_loans/lifetime.rs#L96

[huonw]

Could @cmr use one of his many old rustc's to check it fails? (Does he only need to go back before #10153?)

[emberian]

I don't see a failure with d2cf295

[huonw]

Blame suggests this was fixed by a896440 (2013-03-15), I guess you don't go back that far, @cmr?

[emberian]

I don't have back that far, unfortunately.

On Tue, Nov 12, 2013 at 8:06 PM, Huon Wilson notifications@github.comwrote:

Blame suggests this was fixed by a896440a896440c(2013-03-15), I guess you don't go back that far,
@cmr https://github.com/cmr?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/3053#issuecomment-28351795
.