Re: [PATCH] elf: Sort only uninitialized objects in _dl_map_object_deps()

[Carlos O'Donell <carlos@redhat.com>]

On 7/26/20 6:41 AM, Chung-Lin Tang wrote:
> 
> 
> On 2020/7/26 4:57 AM, Carlos O'Donell via Libc-alpha wrote:
>>> Run the dlopen() for each dynamic library.
>>> Before the patch is installed, it takes 214 seconds.
>>> After patching, it takes 37 seconds.
>> Is it still correct?
>>
>> Do you have a test case you can add?
>>  
>>> Signed-off-by: Xiaoming Ni<nixiaoming@huawei.com>
>> Reviewing this will have to wait until after the release, but this patch is
>> interesting.
> 
> This patch appears to add a linear pass to somewhat reduce the input size
> of a circularly linked case, but frankly speaking, is only useful with the current
> old sorting algorithm, and just to a certain degree.

Chung-Lin, thank you for looking over the suggested fixes.

> The mentioned test case still takes 37 seconds with the proposed patch, while for
> the new DFS-based algorithm, even without any such special case input reduction,
> the sort time will probably be instantaneous.

I expected that might be the case, but it would still be good to put
the example into a test case and verify.

>> Have you looked at Chung-Ling Tang's most recent work in this area?
>>
>> https://patchwork.sourceware.org/project/glibc/patch/1427b370-7400-afd0-16e8-55c1072db20e@mentor.com/
>> https://patchwork.sourceware.org/project/glibc/patch/5de3ab61-3dca-b400-15c6-92ff5ae80877@mentor.com/
> 
> If you're trying out the #17645 sorting patch, remember to add GLIBC_TUNABLES=glibc.rtld.dynamic_sort=2
> to the environment before running the test, or it will still be the old algorithm.
> 
>> Could you use Chung-Ling's test case constructor to write a test case?
>>
> 
> Yeah, a new test case like this is always nice, especially to test if the description language
> is expressive enough to handle this.

Agreed.

-- 
Cheers,
Carlos.