Обсуждение: PG_GETARG_TEXT_PP vs PG_GETARG_TEXT_P
In the “Extending SQL” chapter I see both of these forms are mentioned. But can’t find info about when to use which one.
Markur Sens <markursens@gmail.com> writes:
> In the “Extending SQL” chapter I see both of these forms are mentioned.
> But can’t find info about when to use which one.
PG_GETARG_TEXT_P returns a traditional-format, 4-byte-header value.
PG_GETARG_TEXT_PP is allowed to return either that or a 1-byte-header
value, in case that's what the input is.
PG_GETARG_TEXT_PP is preferred in new code since it can avoid one
step of palloc-and-copy-the-value; the only real downside is you
have to use the appropriate macros to get the string's start address
and length.
regards, tom lane
> On 12 Jun 2022, at 12:06 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > > Markur Sens <markursens@gmail.com> writes: >> In the “Extending SQL” chapter I see both of these forms are mentioned. >> But can’t find info about when to use which one. > > PG_GETARG_TEXT_P returns a traditional-format, 4-byte-header value. > > PG_GETARG_TEXT_PP is allowed to return either that or a 1-byte-header > value, in case that's what the input is. > > PG_GETARG_TEXT_PP is preferred in new code since it can avoid one > step of palloc-and-copy-the-value; the only real downside is you > have to use the appropriate macros to get the string's start address > and length. > > regards, tom lane Ah Thanks. I wouldn’t have guessed that. And I don’t see this mentioned in the header files either. Is it worth adding a relevant comment in the documentation section?
Markur Sens <markursens@gmail.com> writes:
> On 12 Jun 2022, at 12:06 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> PG_GETARG_TEXT_PP is preferred in new code since it can avoid one
>> step of palloc-and-copy-the-value; the only real downside is you
>> have to use the appropriate macros to get the string's start address
>> and length.
> Is it worth adding a relevant comment in the documentation section?
It is documented in the source code where these macros are defined
(fmgr.h).
regards, tom lane