]> git.ipfire.org Git - thirdparty/postgresql.git/commit
Future-proof the recursion inside ExecShutdownNode().
authorTom Lane <tgl@sss.pgh.pa.us>
Mon, 19 Sep 2022 16:16:02 +0000 (12:16 -0400)
committerTom Lane <tgl@sss.pgh.pa.us>
Mon, 19 Sep 2022 16:16:02 +0000 (12:16 -0400)
commitcb7d636e0fb2058e93b444f31a28883a113d489d
treed635b7fd99e374770adcdc612aacc205a90748f4
parentf01fd89b1537cbc5f6572ae6c3d45243bcd7dd14
Future-proof the recursion inside ExecShutdownNode().

The API contract for planstate_tree_walker() callbacks is that they
take a PlanState pointer and a context pointer.  Somebody figured
they could save a couple lines of code by ignoring that, and passing
ExecShutdownNode itself as the walker even though it has but one
argument.  Somewhat remarkably, we've gotten away with that so far.
However, it seems clear that the upcoming C2x standard means to
forbid such cases, and compilers that actively break such code
likely won't be far behind.  So spend the extra few lines of code
to do it honestly with a separate walker function.

In HEAD, we might as well go further and remove ExecShutdownNode's
useless return value.  I left that as-is in back branches though,
to forestall complaints about ABI breakage.

Back-patch, with the thought that this might become of practical
importance before our stable branches are all out of service.
It doesn't seem to be fixing any live bug on any currently known
platform, however.

Discussion: https://postgr.es/m/208054.1663534665@sss.pgh.pa.us
src/backend/executor/execProcnode.c