Обсуждение: Use correct macro for accessing offset numbers.
Hi hackers! While working on pageinspect support for GIN and SpGiST (welcome to review them [0] & [1]), I spotted $subi. PFA trivial patch that uses UInt16GetDatum for OffsetNumber rather than Int16GetDatum [0]https://www.postgresql.org/message-id/CALdSSPiN13n7feQcY0WCmq8jzxjwqhNrt1E%3Dg%3Dg6aZANyE_OoQ%40mail.gmail.com [1] https://www.postgresql.org/message-id/CALdSSPhbAQbFtjK0nT8_G5GsXmsSEVx8J735Ga%2BZxLp9osHcRA%40mail.gmail.com -- Best regards, Kirill Reshke
Вложения
> On 11 Jan 2026, at 16:21, Kirill Reshke <reshkekirill@gmail.com> wrote: > > Hi hackers! > > PFA trivial patch that uses UInt16GetDatum for OffsetNumber rather > than Int16GetDatum Hi! LGTM, should we check another places of offset number conversations to Datum as part of this thread? -- Best regards, Roman Khapov
On Sun, 11 Jan 2026 at 16:41, Roman Khapov <rkhapov@yandex-team.ru> wrote: > > should we check another places of offset number conversations to Datum > as part of this thread? Maybe, I have stopped some more cases, in v2-0001 -- Best regards, Kirill Reshke
Вложения
On Sun, Jan 11, 2026 at 04:58:39PM +0500, Kirill Reshke wrote: > Maybe, I have stopped some more cases, in v2-0001 Right. It's true that we could be more consistent for all these based on their base type, some of them, particularly in the GIN code now, caring about using the correct macro. It may be a good occasion to double-check the whole tree for similar holes based on unsigned types. -- Michael
Вложения
> > On Sun, Jan 11, 2026 at 04:58:39PM +0500, Kirill Reshke wrote: > > Maybe, I have stopped some more cases, in v2-0001 > > Right. It's true that we could be more consistent for all these based on their > base type, some of them, particularly in the GIN code now, caring about using > the correct macro. It may be a good occasion to double-check the whole tree > for similar holes based on unsigned types. > -- > Michael Hi Kirill, Roman, and Michael, While double-checking the tree for similar holes as Michael suggested, I noticed a couple more inconsistencies in contrib/pageinspect/ginfuncs.cwhere we are using signed macros for unsigned types. Specifically, in gin_page_opaque_info, we use Int32GetDatum for maxoff: values[1] = Int32GetDatum(opaq->maxoff); Since maxoff is of type OffsetNumber (uint16), this seems to be the exact same pattern Kirill is addressing in other partsof the GIN code. Although it is widened to int32 here, for the sake of consistency, it should probably be using a 16-bitor unsigned macro. Similarly, in gin_metapage_info, tailFreeSize (which is defined as uint32 in GinMetaPageData) is also converted using Int32GetDatum: values[2] = Int32GetDatum(metadata->tailFreeSize); It might be better to include these cleanups in the next version of the patch to ensure all pageinspect GIN functions followthe same standard. Best regards, Yuan Li(carol)
Hi,
I’ve also seen such cases in the kernel code, and I’m wondering if this should be added to the patch here?
```
postgres@zxm-VMware-Virtual-Platform:~/code/postgres$ git diff
diff --git a/src/backend/utils/adt/lockfuncs.c b/src/backend/utils/adt/lockfuncs.c
index bcbc226125c..9dadd6da672 100644
--- a/src/backend/utils/adt/lockfuncs.c
+++ b/src/backend/utils/adt/lockfuncs.c
@@ -329,7 +329,7 @@ pg_lock_status(PG_FUNCTION_ARGS)
values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1);
values[8] = ObjectIdGetDatum(instance->locktag.locktag_field2);
values[6] = ObjectIdGetDatum(instance->locktag.locktag_field3);
- values[9] = Int16GetDatum(instance->locktag.locktag_field4);
+ values[9] = UInt16GetDatum(instance->locktag.locktag_field4);
nulls[2] = true;
nulls[3] = true;
nulls[4] = true;
@@ -343,7 +343,7 @@ pg_lock_status(PG_FUNCTION_ARGS)
values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1);
values[7] = ObjectIdGetDatum(instance->locktag.locktag_field2);
values[8] = ObjectIdGetDatum(instance->locktag.locktag_field3);
- values[9] = Int16GetDatum(instance->locktag.locktag_field4);
+ values[9] = UInt16GetDatum(instance->locktag.locktag_field4);
nulls[2] = true;
nulls[3] = true;
nulls[4] = true;
```
--
Regards,
Man Zeng
www.openhalo.org
On Mon, 12 Jan 2026 at 04:23, Michael Paquier <michael@paquier.xyz> wrote: > > On Sun, Jan 11, 2026 at 04:58:39PM +0500, Kirill Reshke wrote: > > Maybe, I have stopped some more cases, in v2-0001 > > Right. It's true that we could be more consistent for all these based > on their base type, some of them, particularly in the GIN code now, > caring about using the correct macro. It may be a good occasion to > double-check the whole tree for similar holes based on unsigned types. > -- > Michael Ok > Hi Kirill, Roman, and Michael, > While double-checking the tree for similar holes as Michael suggested, I noticed a couple more inconsistencies in contrib/pageinspect/ginfuncs.cwhere we are using signed macros for unsigned types. > Specifically, in gin_page_opaque_info, we use Int32GetDatum for maxoff: > > values[1] = Int32GetDatum(opaq->maxoff); > > Since maxoff is of type OffsetNumber (uint16), this seems to be the exact same pattern Kirill is addressing in other partsof the GIN code. Although it is widened to int32 here, for the sake of consistency, it should probably be using a 16-bitor unsigned macro. > > Similarly, in gin_metapage_info, tailFreeSize (which is defined as uint32 in GinMetaPageData) is also converted using Int32GetDatum: > > values[2] = Int32GetDatum(metadata->tailFreeSize); > > It might be better to include these cleanups in the next version of the patch to ensure all pageinspect GIN functions followthe same standard. Thank you, I have included your findings in v3. On Mon, 12 Jan 2026 at 11:53, zengman <zengman@halodbtech.com> wrote: > > Hi, > > I’ve also seen such cases in the kernel code, and I’m wondering if this should be added to the patch here? > ``` > postgres@zxm-VMware-Virtual-Platform:~/code/postgres$ git diff > diff --git a/src/backend/utils/adt/lockfuncs.c b/src/backend/utils/adt/lockfuncs.c > index bcbc226125c..9dadd6da672 100644 > --- a/src/backend/utils/adt/lockfuncs.c > +++ b/src/backend/utils/adt/lockfuncs.c > @@ -329,7 +329,7 @@ pg_lock_status(PG_FUNCTION_ARGS) > values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1); > values[8] = ObjectIdGetDatum(instance->locktag.locktag_field2); > values[6] = ObjectIdGetDatum(instance->locktag.locktag_field3); > - values[9] = Int16GetDatum(instance->locktag.locktag_field4); > + values[9] = UInt16GetDatum(instance->locktag.locktag_field4); > nulls[2] = true; > nulls[3] = true; > nulls[4] = true; > @@ -343,7 +343,7 @@ pg_lock_status(PG_FUNCTION_ARGS) > values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1); > values[7] = ObjectIdGetDatum(instance->locktag.locktag_field2); > values[8] = ObjectIdGetDatum(instance->locktag.locktag_field3); > - values[9] = Int16GetDatum(instance->locktag.locktag_field4); > + values[9] = UInt16GetDatum(instance->locktag.locktag_field4); > nulls[2] = true; > nulls[3] = true; > nulls[4] = true; Thank you, I have included your findings in v3. PFA v3 with fixes for signed usage across the tree, with my new findings and suggestions from thread -- Best regards, Kirill Reshke
Вложения
On Mon, Jan 12, 2026 at 01:51:10PM +0500, Kirill Reshke wrote: > PFA v3 with fixes for signed usage across the tree, with my new > findings and suggestions from thread Note that the change in get_opfamily_member() is not right based on the type of "strategy". The rest was OK, so done. -- Michael
Вложения
On 14.01.26 09:08, Michael Paquier wrote:
> On Mon, Jan 12, 2026 at 01:51:10PM +0500, Kirill Reshke wrote:
>> PFA v3 with fixes for signed usage across the tree, with my new
>> findings and suggestions from thread
>
> Note that the change in get_opfamily_member() is not right based on
> the type of "strategy". The rest was OK, so done.
The thread [0] is proposing a patch to change these things in the
opposite direction, effectively reverting commit 6dcfac9696c.
I think the premise of the patch in this thread is incorrect. You have
changed
Int16GetDatum(offset)
to
UInt16GetDatum(offset)
because the variable offset is of type OffsetNumber, which is uint16.
But that is not the meaning of the "UInt16" in UInt16GetDatum(), at
least that's the argument being made in the other thread.
These values end up being converted to an output parameter of type
smallint, and the output function int2out uses DatumGetInt16() to
convert its argument. So the *GetDatum() function should match that, so
we should use Int16GetDatum().
The real problem here is that offset values that are uint32 are being
output via the SQL type smallint, which can't handle the whole set of
values, but this is probably not a problem in practice.
[0]:
https://www.postgresql.org/message-id/flat/CALdSSPhFyb9qLSHee73XtZm1CBWJNo9+JzFNf-zUEWCRW5yEiQ@mail.gmail.com
On Fri, Apr 24, 2026 at 10:23:23AM +0200, Peter Eisentraut wrote: > The thread [0] is proposing a patch to change these things in the opposite > direction, effectively reverting commit 6dcfac9696c. Thanks for the poke. This qualifies as an open item assigned to me, I guess. I'll double-check the whole and reply on the other thread where the patch has been posted at [1]. > The real problem here is that offset values that are uint32 are being output > via the SQL type smallint, which can't handle the whole set of values, but > this is probably not a problem in practice. Will check all that. [1]: https://www.postgresql.org/message-id/CAJ7c6TMcGu8qmRe1gZfJ-gOzVnZq-t%3Dfwn-UuyStx1w6ZyydMw%40mail.gmail.com -- Michael