Test failures due to integer size differences between architectures

[lechner]

Hi,

Your package was uploaded to the Debian archive earlier this year, where we use it for internal databases. Thank you for providing the extension!

Perhaps somewhat related to the discussion in #47, we see variances in your test suite across architectures. On both i386 and armhf, the TODO test number 31—which is expected to fail—actually succeeds.

Is there a better way to handle it than to disable the test? Thanks!

Kind regards
Felix Lechner

[theory]

That's weird. Here's the declaration of the struct for semver:

pg-semver/src/semver.c

Lines 59 to 64 in 87cc30c

	/* memory/heap structure (not for binary marshalling) */
	typedef struct semver {
	int32 vl_len_; /* varlena header */
	vernum numbers[3];
	char prerel[]; /* pre-release, including the null byte for convenience */
	} semver;

Note that it's explicitly using an int32. Compiling to make int 64 bit should not impact int32. Any idea what compilation options are impacting this?

[lechner]

Hi, why is that test marked TODO if it fails as expected, please? Thanks!

[theory]

The point of a TODO is to say "yes, I know this test fails, but want my coverage to be completely, and we will (may?) try to fix it in the future." At that time, it would not longer be a TODO.

[lechner]

But the test cannot be fixed. It performs as expected. You could only modify the underlying code, i.e the typedef int32 vernum;.

[theory]

I expect one day we will refactor semver to support larger integers, and then the test will pass. I don't know when that will be, but as this is one of the tests from the corpus, I wanted to be sure to include it for completeness.

The test should not pass. If it does, it's weird, and I'm curious to know how Postgres was compiled to make that happen.

[theory]

See the TAP spec for details on the meaning of TODO tests.

[lechner]

I'm curious to know how Postgres was compiled

You can find the logs for all architectures by clicking in the Status column here. Does that help?

[theory]

Nothing obvious there. Can you send me the regression.diffs file?

[theory]

Oh, never mind, it's in the logs you linked to, thanks.

[theory]

I think 07b0c8b fixes the problem, bit of a 🤦🏻 actually. Appreciate the report!

[lechner]

The tests now pass on i386 and armhf. Thanks so much for the quick fix!