Re: -readnow crash Rust regression [Re: [RFA v2 3/4] Convert Rust to use discriminated unions]

[Keith Seitz <keiths@redhat.com>]

On 04/12/2018 11:09 AM, Tom Tromey wrote:
>>>>>> "Tom" == Tom Tromey <tom@tromey.com> writes:
> 
> Tom> I have a hack to fix that but it reveals another problem, which is that
> Tom> some of the type rewriting can be done multiple times, causing other bugs.
> Tom> I hadn't considered this possibility.  I'm looking into a good way to
> Tom> fix it.
> 
> This analysis was mistaken, and it turned out to be simpler -- good
> news.

This is excellent! I am looking forward to reading about this.

> I will send out my patches, but writing a test case for 23010 seems
> quite painful.  Based on the stack trace I got, and investigating the
> DWARF of the reproducer, you need:
> 
> * A partial CU with no language, which has a function that has some parameters
> * A full C++ CU that references that function using DW_AT_abstract_origin
> 
> In the webkit reproducer this DW_AT_abstract_origin occurs in an inlined
> function IIRC.

It /is/ painful. Very. I have extracted (most of?) the DIE tree from the webkit reproducer, but it does not reproduce the problem, so I am missing something in my test. Maybe it's the inlined function... [If you'd like it, just lemme know. I may yet play with it some more (re: inlined function).]

Keith