On Wed, Feb 6, 2013 at 11:22 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> On Tue, Feb 5, 2013 at 12:18 AM, Phil Sorber <phil@omniti.com> wrote:
>>> On Mon, Feb 4, 2013 at 10:52 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>> I don't believe that callers should be trying to free() the result.
>>>> Whether it's been strdup'd or not is not any of their business.
>
>>> Is that just because of the nature of this specific function?
>
>> I can't presume to speak for Tom, but I think so. Sometimes the API
>> of a function includes the notion that the caller should pfree the
>> result. Sometimes it doesn't. The advantage of NOT including that in
>> the API contract is that you can sometimes do optimizations that would
>> be impossible otherwise - e.g. you can return the same palloc'd string
>> on successive calls to the function; or you can sometimes return a
>> statically allocated string.
>
> Yeah. In this particular case, it seems rather obvious that the
> function should be returning the same string each time --- if it's
> actually doing a fresh malloc, that sounds like a bug.
It does, but it's noted in a comment that it's only expected to be run once.
>
> But in any case, adding or removing a const qualifier from a function's
> result typically goes along with an API-contract change as to whether
> the caller is supposed to free the result or not. My objection here
> was specifically that I don't believe the contract for get_progname
> includes caller-free now, and I don't want it to start being that.
That's fair. Thanks for the explanation.
>
> regards, tom lane
