[Pymilter] no memory release when assigning chunks to self.fp in milter-template.py

[joao reis]

Sorry, but this is just the example provided in the pymilter 
documentation, and as I said, the same problem occurs when assigning the 
value of "chunks" in a variable or even in a file, nothing related to 
BytesIO or StringIO. I made sure to keep the example provided in the 
pymilter documentation (as you can see from the link pasted 
https://github.com/joaoreis81/milter-template/blob/master/milter-template.py) 
that there is no change that could affect assignments to self.fp. I'm 
starting to think that there really isn't going to be a solution to this.

On 2/4/20 7:38 PM, Stuart D. Gathman wrote:
> On Tue, 4 Feb 2020, joao reis wrote:
>
>> With "del self.fp" it is consuming 244.6 MiB in just one hour.
>
> That proves pretty conclusively that the problem is not with postfix or
> pymilter.  Either BytesIO/StringIO are badly broken (unlikely, we can
> make a non-milter test), or you have "leaked" a reference to self.fp
> (or self.fp.getvalue()) somewhere where it is being colllected globally.
>
> Although, you are doing it wrong:
>
>>       if self.fp:
>>       self.fp.close()
>>       self.fp = None
>>       del self.fp
>>       return Milter.CONTINUE
>
> You need to delete *before* setting to None, otherwise it is a noop.
> But even just setting it to None will immediately reclaim the memory
> in C-Python when it is the last reference.  So apparently, it
> is not the last reference.
>