Some wacky CI test results. Travis passes/Appveyor times out? on the first commit, then Travis fails/Appveyor passes on the second commit (which makes an inconsequential change to a Windows only test).

Don't worry, AppVeyor timeouts are "normal" right now and the failing test on Travis is spurious.

This looks good to me, but cc @cmb69 @weltling for the Windows side (esp. the test).

Thanks @nikic

Actually, we're hitting this:

@cmb69 I didn't fix it there because the stat function calls come from main\streams\streams.c to either _php_stream_stat or _php_stream_stat_path. The former already initializes the buffer so it seemed sensible to make the latter do the same.

However, I can zero initialize buf->st_mode in zend\zend_virtual_cwd.c as well, if you wish (as per PHP-7.4). Please advise.

Regarding support for IO_REPARSE_TAG_MOUNT_POINT, this would be great, but it requires some thought about what to set st_mode to in such cases:

• there is no S_IFJNC (or whatever) so anything used would be non-standard (S_IXJNC) and need to be documented. Also, because it is non-standard there would possibly also be a need for a new windows_is_junction function (or some such).

• And if we did that, would we need to differentiate between junctions and mount points (S_IXMNT and windows_is_mount)?

I'm sure @weltling will have some ideas. My intention here was to supply a quick fix so that the data is consistent, but I will do anything I can to help.

Since it's a fix targeting 7.2, it seems to fine merge (probably should be 7.2 only). The handling of IO_REPARSE_TAG_MOUNT_POINT should target 7.4, if there should be an implementation. There are several irregularities with that, as @johnstevenson mentions, and there should be a lot of testing including network mounts, too.

Thanks.

Thanks @weltling But why "probably should be 7.2 only"?

The buffer is uninitialized in 7.3 as well.

@johnstevenson in 7.3+, any stat call should be running through the ioutil implementation. You mention it in the description actually. If it's not the case, probaly need to check as something should have been missed. Though it might bring regressions, in that case your patch should be ok for 7.3, too.

Thanks.

@weltling From what I can see the ioutil implementation (php_win32_ioutil_stat_ex_w) is not in 7.3. It is in master and 7.4 though.

@johnstevenson ah, indeed. Then 7.2 and 7.3 are good to go, and empty merge upwards :) The missing implementation is a separate thing anyway.

Thansk.

Merged as fe4d724 into 7.2+. I've merged this as-is up to master, as zeroing the struct seems like a good defensive practice independently of the specific issue encountered here.