df7cb opened a new pull request, #1503:
URL: https://github.com/apache/age/pull/1503
On Debian's s390x architecture, some regression tests were failing because
the result set was reordered. Fix by attaching ORDER BY in problematic cases.
On top of this change, another fix is required: `agtype_hash_cmp() changes
values on big-endian architectures. For Debian, I fixed it by adding a
`regress/expected/agtype_1.out` alternate output file where the diff to the
original output file is this:
```
diff -u regress/expected/agtype.out regress/expected/agtype_1.out
--- regress/expected/agtype.out 2024-01-11 13:03:54.051579131 +0100
+++ regress/expected/agtype_1.out 2024-01-11 14:33:37.961477838 +0100
@@ -3593,13 +3593,13 @@
SELECT agtype_hash_cmp('1.0'::agtype);
agtype_hash_cmp
-----------------
- 614780178
+ 1437092844
(1 row)
SELECT agtype_hash_cmp('"1"'::agtype);
agtype_hash_cmp
-----------------
- -888576106
+ -1434266898
(1 row)
SELECT agtype_hash_cmp('[1]'::agtype);
@@ -3659,7 +3659,7 @@
SELECT agtype_hash_cmp('[1, "abcde", 2.0]'::agtype);
agtype_hash_cmp
-----------------
- -1128310748
+ 826120111
(1 row)
SELECT agtype_hash_cmp(agtype_in('null'));
@@ -3701,7 +3701,7 @@
SELECT agtype_hash_cmp('{"id":1, "label":"test",
"properties":{"id":100}}'::agtype);
agtype_hash_cmp
-----------------
- 1116453668
+ -947461933
(1 row)
SELECT agtype_hash_cmp('{"id":1, "label":"test",
"properties":{"id":100}}::vertex'::agtype);
@@ -3713,7 +3713,7 @@
SELECT agtype_hash_cmp('{"id":2, "start_id":1, "end_id": 3,
"label":"elabel", "properties":{}}'::agtype);
agtype_hash_cmp
-----------------
- 1064722414
+ 1662709842
(1 row)
SELECT agtype_hash_cmp('{"id":2, "start_id":1, "end_id": 3,
"label":"elabel", "properties":{}}::edge'::agtype);
```
With that extra file, AGE passes all regression tests on apt.postgresql.org:
https://pgdgbuild.dus.dg-i.net/job/postgresql-16-age-binaries/
But maintaining a full `agtype.out` file along with a full `agtype_1.out`
will be painful, so some other solution will be needed.
* Move the 5 `agtype_hash_cmp` tests to a separate test file, so only these
tests would need two files
* Rewrite the test such that the different value isn't visible (perhaps
check for `<> 0` or similar)
* Just don't test the output value
* Rewrite `agtype_hash_cmp` to output the same value on all architectures
I can help implementing this, but I can't really decide which one of these
options fits best. (The least intrusive version would probably option 1.)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
