Re: BUG #17847: Unaligned memory access in ltree_gist - Mailing list pgsql-bugs

Hi, Alexander and Alexander!

On Tue, 18 Apr 2023 at 14:16, Alexander Korotkov <aekorotkov@gmail.com> wrote:
>
> Hi  Alexander,
>
> Thank you for your feedback.
>
> The revised patch is attached.
>
> On Mon, Mar 20, 2023 at 10:00 AM Alexander Lakhin <exclusion@gmail.com> wrote:
> > 19.03.2023 20:08, Alexander Korotkov wrote:
> > > On Thu, Mar 16, 2023 at 10:35 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > >> What I'm inclined to do about this is add a restriction that the siglen
> > >> value be a multiple of MAXALIGN.  It doesn't look like the reloption
> > >> mechanism has a way to specify that declaratively, but we could probably
> > >> get close enough by just making LTREE_GET_SIGLEN throw an error if it's
> > >> wrong.  That's not ideal because you could probably get through making
> > >> an empty index without hitting the error, but I don't offhand see a
> > >> way to make it better.
> > > Sorry for missing this.
> > >
> > > Please, note that there are infrastructure of reltoption validators.
> > > I think this is the most appropriate place to check for alignment of
> > > siglen.  That works even for empty indexes.  See the attached patch.
> >
> > Thanks for the fix! It works for me.
> >
> > Maybe it's worth to reflect this restriction in the documentation too?
> >       <literal>gist_ltree_ops</literal> GiST opclass approximates a set of
> >       path labels as a bitmap signature.  Its optional integer parameter
> >       <literal>siglen</literal> determines the
> >       signature length in bytes.  The default signature length is 8 bytes.
> >       Valid values of signature length are between 1 and 2024 bytes.
> >
> > How about "The length must be a multiple of <type>int</type> alignment between 4 and 2024."?
> > (There is a wording "<type>int</type> alignment (4 bytes on most machines)" in catalogs.sgml.)
>
> I think it's a bit contradictory to say that int alignment is 4 bytes
> on most machines, but the minimum value is exactly 4.  The revised
> patch says just that length is positive up to 2024.
>
> > Also maybe change the error message a little:
> > s/siglen value must be integer-alignment/siglen value must be integer-aligned/
> > or "int-aligned"? (this spelling can be found in src/)
>
> Thank you, accepted.
>
> > (There is also a detail message, that probably should be corrected too:
> > DETAIL:  Valid values are between "1" and "2024".
> > ->
> > DETAIL:  Valid values are int-aligned positive integers less than 2024.
> > ?)
>
> I can't edit directly the detail message for GUC min/max violation.
> But I've corrected the min value to INTALIGN(1).  Also, I've added
> detail message for alignment validation.
>
> I'm going to push this if no objections.

I've looked into the patch v2 and there is a difference in DETAIL text
for the cases:

(1)
 create index tstidx on ltreetest using gist (t gist_ltree_ops(siglen=2025));
+ERROR:  siglen value must be integer-aligned
+DETAIL:  Valid values are int-aligned positive integers up to 2024.

(2)
+create index tstidx on ltreetest using gist (t gist_ltree_ops(siglen=2028));
+ERROR:  value 2028 out of bounds for option "siglen"
+DETAIL:  Valid values are between "4" and "2024"

Could we stick to the DETAIL like in (2) for both cases?
Overall the patch seems good to be committed.

Regards,
Pavel Borisov