Re: [PATCH RFC 10/19] slab: remove cpu (partial) slabs usage from allocation paths

[Roman Gushchin]

Chris Mason <clm@xxxxxxxx> writes:

> On 1/10/26 10:41 AM, Vlastimil Babka wrote:
>> On 1/10/26 14:20, Chris Mason wrote:
>>> On 1/9/26 3:16 AM, Vlastimil Babka wrote:
>>>> On 10/24/25 16:29, Chris Mason wrote:
>>>>> On Thu, 23 Oct 2025 15:52:32 +0200 Vlastimil Babka <vbabka@xxxxxxx> wrote:
>>>
>>> [ ... ]
>>>
>>>> By the way, there was another bug in this patch, causing a severe memory
>>>> leak, which the AI unfortunately didn't flag. Petr reported it during
>>>> performance testing and it took me more than a day to find it. Oh well :)
>>>>
>>>> Wonder if things got better since then perhaps, and your or Roman's tools
>>>> would find it today? :)
>>>
>>> Yes and no.  It didn't find the leak until I changed the prompt to say:
>>> "there is a leak, find it".  I'll see if I can improve things...
>> 
>> Thanks. Hmm even if it has to be done like this, it could be a substantial
>> time saver vs finding the leak myself.
>
> Finding the missing break on the first pass was tricky because claude
> consistently focused on concerns about potential NULL pointers and
> mostly ignored the loop flow control changes.
>
> I think I've fixed things by expanding the loop analysis and also
> forcing it to make a more fine grained list of changes to analyze before
> it jumps into the review.
>
> It caught the missing break 5 out of 6 times in a loop, so maybe?
> That's probably the best I can get right now for a generic review, but
> claude will almost always be more reliable with extra directions like
> "there is a leak, find it" on top of the review prompt.
>
> I've pushed out two new commits to:
> https://github.com/masoncl/review-prompts
>
> 9a44c271 CS-001.md: pay more attention to loop control flow and memory
> allocations
> 7fad3996 review-core.md: make change categories more fine grained

It helped Gemini too. With these changes even the flash-3 model caught it
from the first attempt.

Thanks