assertion failed: (left == right) && (right == left) (left: 1, right: 0)

[carl-eastlund]

Here's the program:

enum Either {One(()), Other{left:~str,right:~str}}
static one : Either = One(());
fn main () {}

Here's the error message:

task <unnamed> failed at 'assertion failed: `(left == right) && (right == left)` (left: `1`, right: `0`)', /Users/cce/git/rust/src/librustc/middle/trans/adt.rs:521
error: internal compiler error: unexpected failure
note: the compiler hit an unexpected failure path. this is a bug
note: try running with RUST_LOG=rustc=1,::rt::backtrace to get further details and report the results to github.com/mozilla/rust/issues
task <unnamed> failed at 'explicit failure', /Users/cce/git/rust/src/librustc/rustc.rs:371

[lilyball]

I tried looking into this a bit. The failed assertion is in middle::trans::adt::trans_const(), when it checks for NullablePointer. After determining that this is the null case, it asserts that vals.len() is 0, and that's the assertion that fails.

I'm not particularly familiar with this code (or really, any of the compiler yet), but the null case in this enum is of course One(()). I imagine the one value that's present is the (). The docstr for NullablePointer says that the null case "is inhabited by at most one value". This seems to be at odds with the assertion that checks vals.len() == 0 (assuming that vals here is what the docstr is referring to).

[jld]

Indeed, that assertion is wrong and should be deleted. A variant with size-0 contents is eligible to be a null pointer, but since size-0 types like () exist, it can have a nonzero number of fields.

The test nullable-pointer-iotareduction.rs would have caught that — the type used there is enum E<T> { Thing(int, T), Nothing((), ((), ()), [i8, ..0]) }, for exactly this reason — but somehow I forgot to test statics.

[emberian]

Updated code, new assertion:

#[feature(struct_variant)];             

enum Either {                           
    One(()),                            
    Other { left:~str, right:~str}      
}                                       

static one : Either = One(());          
fn main () {}                           

foo.rs:3:1: 6:2 warning: code is never used: `Either`, #[warn(dead_code)] on by default                                        
foo.rs:3 enum Either {                                                                                                         
foo.rs:4     One(()),                                                                                                          
foo.rs:5     Other { left:~str, right:~str}                                                                                    
foo.rs:6 }                                                                                                                     
foo.rs:8:1: 8:31 warning: code is never used: `one`, #[warn(dead_code)] on by default                                          
foo.rs:8 static one : Either = One(());                                                                                        
         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                                                                                        
error: internal compiler error: unexpected failure                                                                             
This message reflects a bug in the Rust compiler.                                                                              
We would appreciate a bug report: http://static.rust-lang.org/doc/master/complement-bugreport.html                             
note: the compiler hit an unexpected failure path. this is a bug                                                               
Ok(task 'rustc' failed at 'assertion failed: !is_undef(wrapped)', /build/rust-git/src/rust/src/librustc/middle/trans/adt.rs:796
)                                                                                                                              

[eddyb]

This doesn't need #[feature(struct_variant)];, the source of the problem is One(()).
@carl-eastlund, could you update the issue description?

[carl-eastlund]

I haven't looked at Rust or this issue in several months; if there's
something in the issue that only I can update, you'll have to spell out
what I need to change it to. A lot of these details are paged out of my
mind right now.

Carl Eastlund

On Sun, Mar 2, 2014 at 5:19 AM, Eduard Burtescu notifications@github.heygears.comwrote:

This doesn't need #[feature(struct_variant)];, the source of the problem
is One(()).
@carl-eastlund https://github.com/carl-eastlund, could you update the
issue description?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/8516#issuecomment-36450694
.

[emberian]

@eddyb the issue description doesn't really need updating, especially since there's no specifics still.

[jld]

I'm sure I commented on this somewhere at some point since I wrote that last comment — I was assuming that an undef could be wrapped in a struct to make it not-undef (because undef is used for alignment padding, because we have to add our own for constants in some cases), but that doesn't actually work. So the !is_undef assertion actually does indicate that stuff would break otherwise. The code that fills in undef for the unused fields of non-nullary null pointer variants should just use C_null, I think.

[jld]

…and that's #8506. I'll close this, because the failure this was initially opened for was fixed in 5487f15 / #11441, but I apparently forgot the “Closes” annotation on the commit.