Re: [RFC] Time to talk about StringNG merge again?

On 07/04/2013 11:51 PM, Kinkie wrote:
>>> void
>>> >> SBuf::reserveCapacity(size_type minCapacity)
>>> >> {
>>> >> Must(0 <= minCapacity && minCapacity <= maxSize);
>>> >> reserveSpace(minCapacity-length());
>>> >> }
>> >
>> > This does not look right. For example, if minCapacity is smaller than
>> > length(), we should do nothing, but will throw instead. I think this
>> > method should look something like this:
>> >
>> > Must(0 <= minCapacity && minCapacity <= maxSize);
>> > cow(minCapacity);

>>> >> if (store_->canAppend(off_+len_, minSpace)) {
>>> >> debugs(24, 7, "not growing");
>>> >> return;
>>> >> }

>> > AFAICT, reserveSpace() should be implemented using
>> > something like this:
>> >
>> > Must(0 <= minSpace && minSpace <= maxSize - length());
>> > reserveCapacity(length() + minSpace);

> The comment is badly worded but the code is correct:

I am not sure which code you are referring to here, but I do not think
it is correct for SBuf::reserveCapacity(minCapacity) to throw when
minCapacity is smaller than length().

> A decision needs to be made if these two methods have different
> purposes or are to be simply different expressions of the same
> underlying concept (as it is now).

Sorry, this question is too abstract for me to answer it -- it feels
like both statements are true: The methods have different purpose but
are manipulating the same underlying concept of available buffer area.
One method deals with the total size of the buffer area, while the other
is concerned with the size of the yet unused buffer area.

> IMO, thinking about it, it does make sense that reserveSpace also
> guarantees that after the call the MemBlob will be exclusively owned
> by this SBuf (aka, blindly cow()), while reserveCapacity can mostly be
> meant as an optimization, to hint the SBuf memory management system
> that an append operation of a known size is about to happen, so that
> only one cow is performed.
> What do you think?

I think that both methods should be adjusted, possibly as sketched
above. They should not throw when the required space or capacity is
available. The documentation should tell users that both reserve*()
methods ensure single-ownership of the buffer by the caller.

HTH,

Alex.
Received on Wed Jul 17 2013 - 23:06:44 MDT