On Sat, Feb 7, 2026 at 9:31 AM Tom Lane <[email protected]> wrote:
> Andrew Dunstan <[email protected]> writes:
> > Fair enough. I agree it's not terribly helpful. In any case, maybe we
> > need to look at my earlier suggestion of adding a suppression file.
>
> Perhaps, but since we're getting pretty hard up against the release
> freeze deadline, I think the thing to do for today is just to update
> .abi-compliance-history.  We know that way works.

So it seems I need to do the attached for 18, and the same again for
17 and 16.  Look about right?

I peeked at the libabigail man page, and I don't see any discussion of
incomplete types, but it looks like OPAQUE_TYPE_LABEL might be a way
to tell it about this.  If that works, wouldn't it be better than
doing this forever?  Where would that be configured?
From a209e9c5cfeef91aa597d83854b1d448cf2c7617 Mon Sep 17 00:00:00 2001
From: Thomas Munro <[email protected]>
Date: Sat, 7 Feb 2026 10:56:04 +1300
Subject: [PATCH] Tell ABI checker that 33e3de6d is OK.

It's not really an ABI break if you change the layout/size of an object
with incomplete type.

Discussion: https://www.postgresql.org/message-id/1871492.1770409863%40sss.pgh.pa.us
---
 .abi-compliance-history | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/.abi-compliance-history b/.abi-compliance-history
index f4e27cfb160..420ad2c3aa1 100644
--- a/.abi-compliance-history
+++ b/.abi-compliance-history
@@ -18,6 +18,16 @@
 # Be sure to replace "<ADD JUSTIFICATION HERE>" with details of your change and
 # why it is deemed acceptable.
 
+33e3de6d77e87d6c3c6f8f878dd8de42d37c3b8f
+#
+# Add file_extend_method=posix_fallocate,write_zeros.
+# 2026-02-06 17:38:39 +1300
+#
+# Modifying GUC tables isn't really an ABI break: the relevant object has
+# incomplete type so its layout is inaccessible from C.
+#
+# Discussion: https://www.postgresql.org/message-id/flat/e1f0cd3b-0164-45f5-9705-e922e59df90f%40dunslane.net#8a350b54012c0042f9869d288e978cfe
+
 492a69e1407029f8c673484f44aa719a63323d77
 #
 # Reject ADD CONSTRAINT NOT NULL if name mismatches existing constraint
-- 
2.52.0

Reply via email to