From: Tom Lane <tgl@sss.pgh.pa.us>
To: Arthur Zakirov <a.zakirov@postgrespro.ru>
cc: exclusion@gmail.com, pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #15150: Reading uninitialised value in NISortAffixes
 (tsearch/spell.c)
In-reply-to: <20180412115638.GA19444@zakirov.localdomain>
References: <152353327780.31225.13445405496721177988@wrigleys.postgresql.org>
 <20180412115638.GA19444@zakirov.localdomain>
Comments: In-reply-to Arthur Zakirov <a.zakirov@postgrespro.ru>
	message dated "Thu, 12 Apr 2018 14:56:39 +0300"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <23883.1523571243.1@sss.pgh.pa.us>
Date: Thu, 12 Apr 2018 18:14:03 -0400
Message-ID: <23884.1523571243@sss.pgh.pa.us>
Precedence: bulk

Arthur Zakirov <a.zakirov@postgrespro.ru> writes:
>> It looks that the following condition in NISortAffixes(IspellDict *Conf)
>> uses uninitialised ptr->issuffix:
>> if (ptr == Conf->CompoundAffix ||
>> ptr->issuffix != (ptr - 1)->issuffix ||
 
> Yes, you are right. The second condition isn't right. Instead of
> "ptr->issuffix != (ptr - 1)->issuffix" "Affix->type" should be checked
> because we check for uniqueness of affixes.

Yeah, existing code is clearly wrong, patch looks OK, will push.

But I see from the code coverage report that this bit is unexercised
in the regression tests.  Any chance of getting a test that covers
this?  I'm worried that this means we also lack any coverage of
cases where the CompoundAffix array has more than one entry.

			regards, tom lane