Try to get firebird working on linux x32 CI

[nielsdos]

I'll clean up the history upon merging.

[nielsdos]

@SakiTakamachi I'm trying to get 32-bit Linux CI to run Firebird. I added a service container and the necessary configuration, but the tests fail with Uncaught PDOException: SQLSTATE[HY000] [335544721] Unable to complete network request to host "localhost", so it seems that the test suite cannot connect to the server in the service container. Do you have any idea why?

[cmb69]

Maybe @KentarouTakeda can help?

[SakiTakamachi]

@nielsdos
maybe

Suggested change

	PDO_FIREBIRD_TEST_DSN: firebird:dbname=localhost:test.fdb
	PDO_FIREBIRD_TEST_DSN: firebird:dbname=firebird:test.fdb

edit:
Looking at the mysql DSN above, it appears that the 32-bit CI specifies the service name rather than "localhost".

[KentarouTakeda]

I think @SakiTakamachi is right. It's because of this line.

php-src/.github/workflows/push.yml

Lines 143 to 144 in d8cee06

	container:
	image: ubuntu:20.04

• x64: Tests are run directly on the host specified in runs-on
• x32: A ubuntu:20.04 container is created inside the runs-on host, and tests are run inside it.

I tried to run x32 tests directly in the runs-on container, but it seems very difficult.

So it seems best to fix it the way Saki says. In that case, please note that actions/test-linux and x64 tests will also need some adjustments.

[KentarouTakeda]

Most of the tests passed in my fork.

However, some tests failed because they were written to reference ::1. I don't know how to fix these yet. I created a pull request to fix this: #16115

[nielsdos]

Good finds! I missed this nuance completely, I'll import your commit. Thanks a lot @KentarouTakeda and @SakiTakamachi

[SakiTakamachi]

Looks good to me

Ah, sorry, I mistook it for another PR.

[nielsdos]

All but two tests succeeded:

• ext/pdo_firebird/tests/bug_15604.phpt fails because of another datatype being returned; may be related to a different version of firebird being used? Also I find the test name strange It is not possible to pass a NULL value as an input parameter if the field is marked as NOT NULL. If it's "NOT NULL", why should a "NULL" value be allowed?
• ext/pdo_firebird/tests/bug_76448.phpt fails due to a mutex handling bug, I'll try to get a 32-bit ZTS build going to check what's going on.

[SakiTakamachi]

@nielsdos

Regarding the first test:

First, the data type is different because it's probably a bigint type.

php-src/ext/pdo_firebird/tests/bug_15604.phpt

Line 18 in fec2055

id bigint not null,

Since can't represent a 64 bit number in 32 bits, it will probably be a character.

Also, null is passed to the not null column, but in this test, a sequence is prepared for when null is passed to the id, so no error occurs and the id is completed.

The same is true for mysql's autoincrement, but there is a function that automatically completes by passing null to a not null column, so it is not necessarily impossible to pass null to a not null column.

[nielsdos]

@SakiTakamachi I see, thanks for clarifying and I'll also change the test code to use "int" instead of "bigint".

[cmb69]

• ext/pdo_firebird/tests/bug_76448.phpt fails due to a mutex handling bug, I'll try to get a 32-bit ZTS build going to check what's going on.

I can basically reproduce this on Windows x86 ZTS. After a minute, the test times out, reporting:

Warning: stream_socket_accept(): Accept failed: Ein Verbindungsversuch ist fehlgeschlagen, da die Gegenstelle nach einer bestimmten Zeitspanne nicht richtig reagiert hat, oder die hergestellte Verbindung war fehlerhaft, da der verbundene Host nicht reagiert hat in C:\php-sdk\phpdev\vs17\x86\php-src\ext\pdo_firebird\tests\payload_server.php on line 18

Termsig=-1073741819

The warning is due to WSAETIMEDOUT. The termsig is access violation.

I'll look into this.

Oh, actually, the test apparently always fails for me (x64, NTS, too).

[cmb69]

Oh, actually, the test apparently always fails for me (x64, NTS, too).

Well, should not build against v4 and run against v3.

Anyhow, it looks like the test does not test what it is supposed to, as of PHP 8.4 or when pdo_firebird is built against the v4 API. While it reaches

php-src/ext/pdo_firebird/firebird_driver.c

Line 1253 in d7a37cc

if (!isc_version(&H->db, php_firebird_info_cb, (void*)tmp)) {

the callback is never called (albeit the test might succeed):

php-src/ext/pdo_firebird/firebird_driver.c

Lines 1198 to 1208 in d7a37cc

	/* callback to used to report database server info */
	static void php_firebird_info_cb(void *arg, char const *s) /* {{{ */
	{
	if (arg) {
	if (*(char*)arg) { /* second call */
	strlcat(arg, " ", INFO_BUF_LEN);
	}
	strlcat(arg, s, INFO_BUF_LEN);
	}
	}
	/* }}} */

So the test is badly written, since it should not succeed if it doesn't test what it is supposed to test, and it should be more malleable (like the new MySQL Fake Server). but given that the documentation about the Firebird wire protocol is apparently sparse, one would need to read the implementation, sniff the network and what not. Especially for this test it might not be worth the trouble (the bug was about using strcat() instead of strlcat()). Maybe we should just drop/skip this test case.

[nielsdos]

PR broke, it's stuck on "updating".

[nielsdos]

Replacement: #17045