Re: [PATCH 4/4] Skip .cold functions in search_minsyms_for_name

[Bernd Edlinger <bernd.edlinger@hotmail.de>]

On 1/31/21 6:03 PM, Andrew Burgess wrote:
> * Bernd Edlinger <bernd.edlinger@hotmail.de> [2021-01-31 10:13:54 +0100]:
> 
>> When a function with the same name is also available,
>> that is preferred.
>>
>> The .cold function is not the external interface.
>>
>> Fixes 77f2120b200 ("Don't drop static function bp locations w/o
>> debug info")
> 
> I assume that this last sentence means that there's a bug / oversight
> with this previous commit.
> 
> You should explain what this issue with this commit is and how this
> commit fixes it.  For contrast, notice that the commit you reference
> took ~200 lines to explain the change, while you've given us just 7.
> A good commit message makes reviewing this a lot easier.
> 

Thanks for your review.

I'll add this to the commit message, hope it clarifies the patch:

The motivation for this patch, is that I became aware that the "real"
get_alias_set function (in gcc's cc1) had a two-PC breakpoint
location with gdb-9 but a three-PC breakpoint location with gdb-10.
The third breakpoint location was not an entry point but somewhere in the
error handling, which is named "get_alias_set.cold".

Bisection pointed to the above commit.  Prior to this commit the
static "get_alias_set.cold" was ignored, since there is also global
get_alias_set function.  Basically the previous behavior was ignore
any static function, when there is a global function with the same name.
The current behavior is to consider a static function and a global
function with the same name as different entities.

But a function named with ".cold" is not a real function at all.
It is just a wrapper for all the cold code paths in that function.
Therefore this patch checks if a function with the same name is
found and skip the ".cold" overload in that case.


Bernd.