On 12/29/2025 7:44 PM, Tom Lane wrote:
> Bryan Green <[email protected]> writes:
>> One notable behavioral change: check hooks using GUC_EXTRA_IS_CONTEXT
>> now use palloc() instead of guc_malloc(). The old approach with
>> guc_malloc() allowed check hooks to return false on OOM, letting the
>> caller handle it at the appropriate error level. With palloc() an OOM
>> throws an immediate ERROR. This seemed like an acceptable tradeoff
> 
> Why?  It seems both inconsistent and unsafe.
> 
>                       regards, tom lane
Fair enough to call me on that.  I mainly thought that if we are having
problems allocating what is usually a few bytes then throwing an error
would have been acceptable.  Based on your comment about unsafe and a
bit deeper thinking I can see where this is probably not a welcome
change in behavior.  I suppose we could catch the error and convert it
to a false return.

-- 
Bryan Green
EDB: https://www.enterprisedb.com


Reply via email to