Liss Heidrich comments

Results 23 comments of


                                            Liss Heidrich

Feature: `Serve::tcp_nodelay`

> Is `TCP_NODELAY` ever on by default, such that `.tcp_nodelay(false)` would make sense to have? I was originally thinking that the method would be parameter-less. The idea was that there...

Assertion triggered on small allocations with custom chunk size

Sorry forgot to mention: allocation is done through ``` auto *x = manager.construct("test")[sizeof(void *)](); ```

Assertion triggered on small allocations with custom chunk size

We were basically trying to see if there is any performance benefit to choosing a greater chunk size. But apart from that/if you think it makes no difference we don't...

Do not rely on implementation details of `boost::interprocess`

Thanks, can be closed

metall manager capacity limit SEGFAULT

You are on MacOS, right? Maybe there is some difference between MacOS and Linux that prevents you from being able to reproduce it. Could you try increasing the maximum `cap`...

metall manager capacity limit SEGFAULT

Thanks, that PR seems to have resolved the issue. You can close this if you want

SEGFAULT through nullptr access in `object_cache::priv_make_room_for_new_blocks`

Disabling the object cache seems to have fixed the issue, the message however is not printed. (I still haven't figured out how to reproduce it in a minimal example yet)

SEGFAULT through nullptr access in `object_cache::priv_make_room_for_new_blocks`

Finally came around to testing that, it does indeed print "Deallocating an offset that was not allocated" a few times close to the beginning. The messages in the screenshot are...

SEGFAULT through nullptr access in `object_cache::priv_make_room_for_new_blocks`

![grafik](https://github.com/user-attachments/assets/1da4dcc4-e41d-4265-bdc9-face44461100)

SEGFAULT through nullptr access in `object_cache::priv_make_room_for_new_blocks`

The program does not abort. And sorry I don't have a minimal example I could send you. Currently we basically need to run it through the whole codebase with 2...