From: Tom Lane <tgl@sss.pgh.pa.us>
To: Richard Guo <guofenglinux@gmail.com>
cc: pgsql-hackers@lists.postgresql.org
Subject: Re: Use outerPlanState macro instead of referring to leffttree
In-reply-to: 
 <CAMbWs4-vYhh1xsa_veah4PUed2Xq=Ed_YH3=Mqt5A3Y=EgfCEg@mail.gmail.com>
References: 
 <CAMbWs4-vYhh1xsa_veah4PUed2Xq=Ed_YH3=Mqt5A3Y=EgfCEg@mail.gmail.com>
Comments: In-reply-to Richard Guo <guofenglinux@gmail.com>
	message dated "Thu, 14 Apr 2022 14:49:23 +0800"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <95730.1656711122.1@sss.pgh.pa.us>
Date: Fri, 01 Jul 2022 17:32:02 -0400
Message-ID: <95731.1656711122@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/95731.1656711122%40sss.pgh.pa.us>
Precedence: bulk

Richard Guo <guofenglinux@gmail.com> writes:
> In the executor code, we mix use outerPlanState macro and referring to
> leffttree. Commit 40f42d2a tried to keep the code consistent by
> replacing referring to lefftree with outerPlanState macro, but there are
> still some outliers. This patch tries to clean them up.

Seems generally reasonable, but what about righttree?  I find a few
of those too with "grep".

Backing up a little bit, one thing not to like about the outerPlanState
and innerPlanState macros is that they lose all semblance of type
safety:

#define innerPlanState(node)		(((PlanState *)(node))->righttree)
#define outerPlanState(node)		(((PlanState *)(node))->lefttree)

You can pass any pointer you want, and the compiler will not complain.
I wonder if there's any trick (even a gcc-only one) that could improve
on that.  In the absence of such a check, people might feel that
increasing our reliance on these macros isn't such a hot idea.

Now, the typical coding pattern you've used:

 ExecReScanHash(HashState *node)
 {
+	PlanState  *outerPlan = outerPlanState(node);

is probably reasonably secure against wrong-pointer slip-ups.  But
I'm less convinced about that for in-line usages in the midst of
a function, particularly in the common case that the function has
a variable pointing to its Plan node as well as PlanState node.
Would it make sense to try to use the local-variable style everywhere?

			regards, tom lane