Re: [PATCH v2] Remove upper limit on tunable MALLOC_MMAP_THRESHOLD

[Patrick McGehearty <patrick.mcgehearty@oracle.com>]

I thought about documentation issues.
The change in behavior for single threaded programs will
match the current documentation more closely. No change needed there.

I decided not to suggest documentation changes for the multi-thread
case as other threads have wanted to avoid tying documentation to
implementation details. If/when we change how MAX_HEAP_SIZE is handled
as described in the next paragraph, the current documentation would
be fine.

Longer term, it probably would be useful to change the max heap
size to be a tunable variable instead of a fixed value.
That change might also have the max heap size changed appropriately
when the MALLOC_MMAP_THRESHOLD is set beyond the default value.
Of course documentation would need to be added about the new
tunable for MAX_HEAP_SIZE.

- patrick


On 11/29/2021 2:42 PM, DJ Delorie wrote:
> Ok, based on all that, I think it's safe to increase the threshold -
> although, given your description, setting it bigger than the mmap'd heap
> size will still be kinda pointless in multi-threaded programs.  Safe,
> but pointless.
>
> I wonder if any of these details or caveats need to be documented,
> either in the code or the manual...
>