Re: [PATCH v2 4/4] Fix parameter-less template regression in new DWARF reader

[Simon Marchi <simark@simark.ca>]

On 1/17/23 14:39, Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:
> 
> Simon> Starting with this patch, I get (using _GLIBCXX_DEBUG):
> [...]
> 
> Ugh, sorry about that.
> I've reproduce it and I'm working on a fix.
> 
> Tom

I started to look into this, using that other patch I sent that dumps
the contents of the cooked index.  You probably already know this, as
you have been looking at the problem already, but it can perhaps help
pothers.  The specific instance I look at is:

  $ ./gdb -nx -q --data-directory=data-directory testsuite/outputs/gdb.cp/cpexprs/cpexprs -ex start -ex "print policy1::function" -batch

When searching in the cooked index, the search string (canonicalized
version of the input, I think) is:

  "policy<int, operation_1<void*> >::function"

The problem is that the lower bound function used in
cooked_index::find (cooked_index_entry::compare) is invalid for these
two consecutive entries:

    [365] ((cooked_index_entry *) 0x621000115080)
    name:       policy
    canonical:  policy
    DWARF tag:  DW_TAG_subprogram
    flags:      0x0 []
    DIE offset: 0x3951
    parent:     ((cooked_index_entry *) 0x6210001137a0) [policy<int, operation_2<void*> >]

    [366] ((cooked_index_entry *) 0x621000113740)
    name:       policy<int, operation_1<void*> >
    canonical:  policy<int, operation_1<void*> >
    DWARF tag:  DW_TAG_class_type
    flags:      0x0 []
    DIE offset: 0x22a3
    parent:     ((cooked_index_entry *) 0)

It returns that [365] greater or equal to our search string, but also
that [366] is less than our search string.  This is a contradiction,
because elements are supposed to be sorted according to the comparison
function.

Simon