FAILED TEST exist in make test on FreeBSD13: Test lstat() and stat() functions: usage variations - effects of is_dir() [ext/standard/tests/file/lstat_stat_variation10.phpt]

[project2021-code]

Description

The following code:

cd /tmp && \
fetch https://www.php.net/distributions/php-8.2.8.tar.xz && \
tar xvf php-8.2.8.tar.xz && \
cd /tmp/php-8.2.8 && \
./configure --disable-all && \
make test

Resulted in this output:

=====================================================================
FAILED TEST SUMMARY
---------------------------------------------------------------------
Test disk_free_space and its alias diskfreespace() functions : basic functionality [ext/standard/tests/file/disk_free_space_basic.phpt]
Test lstat() and stat() functions: usage variations - dir/file name stored in object [ext/standard/tests/file/lstat_stat_variation18.phpt]
Test lstat() and stat() functions: usage variations - dir/file names in array [ext/standard/tests/file/lstat_stat_variation19.phpt]
Test lstat() and stat() functions: usage variations - link names stored in array/object [ext/standard/tests/file/lstat_stat_variation20.phpt]
symlink() & friends [ext/standard/tests/file/symlink.phpt]
=====================================================================

php_test_results_20230711_0411.txt

I also tried to compile in /mnt. Another Failed throw.

=====================================================================
FAILED TEST SUMMARY
---------------------------------------------------------------------
Test disk_free_space and its alias diskfreespace() functions : basic functionality [ext/standard/tests/file/disk_free_space_basic.phpt]
Test lstat() and stat() functions: usage variations - dir/file name stored in object [ext/standard/tests/file/lstat_stat_variation18.phpt]
Test lstat() and stat() functions: usage variations - dir/file names in array [ext/standard/tests/file/lstat_stat_variation19.phpt]
Test lstat() and stat() functions: usage variations - link names stored in array/object [ext/standard/tests/file/lstat_stat_variation20.phpt]
Bug #74090 stream_get_contents maxlength>-1 returns empty string on windows [ext/standard/tests/streams/bug74090.phpt]
=====================================================================

php_test_results_20230711_0419.txt

But I expected this output instead:
Bug tests are in Passed / Skipped / Expected Failed / Warned, no Bug test is Failed

PHP Version

PHP 8.2.8

Operating System

FreeBSD 13.2-RELEASE

[nielsdos]

The problem is that rdev is a negative number (-1) in your case, the test should probably be amended. I'll have a look soon.

[project2021-code]

Thank you, nielsdos. You saved my time.
If you need to me to test with changes, please let me know.

[nielsdos]

You could test the linked PR, but I dont think it's necessary, I'm pretty sure it works.
I still have to figure out why the diskspace test failed, that won't be fixed by the PR because its failure has a different reason.

[project2021-code]

Bad news, the rdev is still -1.

cd /tmp && \
fetch https://www.php.net/distributions/php-8.2.8.tar.xz && \
tar xvf php-8.2.8.tar.xz && \
cd php-8.2.8 && \
fetch https://raw.githubusercontent.com/php/php-src/8a712626b0134b485e8c428c33f9558c1b94da8c/ext/standard/tests/file/lstat_stat_variation18.phpt \
-o ext/standard/tests/file/lstat_stat_variation18.phpt && \
fetch https://raw.githubusercontent.com/php/php-src/8a712626b0134b485e8c428c33f9558c1b94da8c/ext/standard/tests/file/lstat_stat_variation19.phpt \
-o ext/standard/tests/file/lstat_stat_variation19.phpt && \
fetch https://raw.githubusercontent.com/php/php-src/8a712626b0134b485e8c428c33f9558c1b94da8c/ext/standard/tests/file/lstat_stat_variation20.phpt \
-o ext/standard/tests/file/lstat_stat_variation20.phpt

php_test_results_20230711_0611.txt

I find a discussion about this problem, not sure if it helps you to identify the problem.
Perl/perl5#18837

[nielsdos]

It seems your test files didn't actually get replaced as you have the same diff as before.
A value of -1 is possible because of the macro check. However, you might be right that something more is going on here: https://github.com/freebsd/freebsd-src/blob/main/sys/sys/stat.h does still contain st_rdev, so I'm not sure why it didn't detect it on your system.
I'll leave this for someone else for now.

I find a discussion about this problem, not sure if it helps you to identify the problem.
Perl/perl5#18837

This seems unrelated, as we don't ship our own stat.h.

[project2021-code]

Strange, the file should be replaced.

For the discussion, I tried the command below (from the discussion)

perl -e '$file = shift; my @p = stat($file); print join(" ", @p[0,6]), "\n"' /tmp ; stat -f "%d %r" /tmp

and get

FreeBSD 12.3-RELEASE iocage Jail

TrueNAS 12.0U8.1

FreeBSD 13.2-RELEASE

It seems that the value of (dev & rdev) from perl and from stat are different. Only the dev value are same in Jail.
If this is the behavior of FreeBSD, should we ignore the FAILED test lstat_stat_variation18/19/20?

For the disk_free_space test, it may relate to ZFS system.

I would also like to know the information of symlink() & friends and Bug #74090 test.

[nielsdos]

Thanks for the info.

Ah right, so in a way it is somewhat related to the perl issue, but also different. The value 18,446,744,073,709,551,615 you're seeing, which is the right value, has the same representation as -1. What seems to have changed is that rdev is now unsigned instead of signed. But because the 64-bit numbers in PHP are signed numbers instead of unsigned, you see it represented as -1. Not sure what we can do here, representing it as a float instead doesn't work because of the precision loss, and representing it as a string is a BC break...

I tested the patch in my PR and it works when I manually force the rdev output to -1. The PR does still seem necessary because there are systems where the actual value will be -1 (i.e. if the struct lacks the rdev field).

For the disk_free_space test, it may relate to ZFS system.

Hmm, I don't know a lot about ZFS, so maybe someone else who knows more about this could explain.

I would also like to know the information of symlink() & friends and Bug #74090 test.

The test isn't very helpful in finding the issue unfortunately, because it only outputs true or false.
It clearly went far enough to create the server stream, otherwise we would see an error in the output.
It could be a spurious failure. Unfortunately, I don't have access to a FreeBSD13 machine right now so I can't debug it myself.

The symlink test fails because linkinfo() returns the st_dev field, which is negative. The same underlying problem as for rdev probably. This one isn't an easy fix because from the manual: Returns a non-negative integer on success, -1 in case the link was not found, (...).

[Girgias]

For the disk_free_space test, it may relate to ZFS system.

Hmm, I don't know a lot about ZFS, so maybe someone else who knows more about this could explain.

Probably yes, the test is basically useless for BTRFS (and it took me a lot of digging through the speec of understanding why that I don't think I can explain fully) and is a pretty old test, so expectations on a ZFS system might need to be updated too.