Re: [PATCH] Prevent LTO section collision for a symbol name starting with '*'.

[Martin Liška <mliska at suse dot cz>]

On 9/9/19 4:33 PM, Jan Hubicka wrote:
>> PING^1
>>
>> On 8/26/19 12:04 PM, Martin Liška wrote:
>>> Ok. I have a semi-working patch that has issues for inline clones.
>>> When we call cgraph_node::get_untransformed_body for an inline clone,
>>> then one needs to use clone_of->order to find proper LTO stream.
> 
> This seems OK to me - when using inline clone we really look for a body
> of its master, so that seems OK.

Ok!

>>>
>>> What's more problematic is that such clone can be expanded:
>>>
>>> f/12 (f) @0x7ffff769f708
>>>   Type: function definition analyzed
>>>   Visibility: external public
>>>   References: mumble.lto_priv.0/8 (write)
>>>   Referring: 
>>>   Read from file: /tmp/cciAkXHp.ltrans1.o
>>>   Function f/12 is inline copy in main/0
>>>   Availability: local
>>>   Function flags: count:1073741824 (estimated locally) local nonfreeing_fn executed_once
>>>   Called by: main/0 (inlined) (1073741824 (estimated locally),1.00 per call) 
>>>   Calls: 
>>>
>>> and lost. So we end up with an orphan and we ICE with:
> 
> We do some work on removing unnecesary master clone when function is
> fully inlined and I guess in that case you lose the order info.
> One option would be to copy order into all inline clones (it does not
> have very good meaning there) or do that when reshaping the tree.

What do you mean by "reshaping the tree"?

> This is done in cgraph_node::remove at the place clone_of is
> manipulated.

The inline_clone manipulation happens in cgraph_node::find_replacement where
we manipulate the clone_of.

Martin

> This is probably bit cleaner.
>>>
>>> /home/marxin/Programming/gcc/gcc/testsuite/gcc.dg/lto/20081112_0.c: In function ‘main’:
>>> /home/marxin/Programming/gcc/gcc/testsuite/gcc.dg/lto/20081112_0.c:10:3: fatal error: /tmp/cciAkXHp.ltrans1.o: section f is missing
>>>
>>> So usage of symtab_node::order seems awkward to me :/
>>>
>>> Martin
>>>
>>