Improve POSIX compliance in CLI/FileUtils.cpp

[SamuraiCrow]

Some POSIX platforms, such as Haiku and some BSDs, don't supply DT_* identifiers and the corresponding d_type field in stat. This fixes that and has been tested to still work on 64-bit Linux as well as 64-bit Haiku.

[zeux]

Thanks for the contribution!

I'm not sure what ATTOIF is / where it is supposed to come from, but it is not defined at the point of the check on either macOS or Ubuntu Linux with standard toolchains (clang/gcc respectively). As such, this PR would result in us having to issue an extra syscall for each file or folder met in the traversal, which is probably not intentional.

[zeux]

Needs a different compatibility strategy so that macOS/Linux by default do not need to stat the file.

[SamuraiCrow]

@zeux Thanks! I'll try to find where that macro is defined. It's supposed to translate the d_type parameter into the mode_t bitfields. I was so happy to find something that worked on both Haiku and Linux that it didn't occur to me that it might not be defined.

[SamuraiCrow]

Apple has it defined in the last entry in dirent.h. I'll try checking for an include directive for sys/types.h and dirent.h and see if it helps.

[zeux]

Apple has it defined in the last entry in dirent.h. I'll try checking for an include directive for sys/types.h and dirent.h and see if it helps.

Ah, maybe that's the issue here - your patch is using ATTOIF instead of DTTOIF?

[SamuraiCrow]

There. Sorry about the typos. It should be smaller and less branchy by eliminating the double-pipe or statements (d_type==x || mode & S_x) becomes (mode & S_x) on the if statements.

[zeux]

Thanks - this is better, but we're not there yet. I don't think the switch to & is correct. What is it motivated by?

From Linux bits/stat.h:

/* File types.  */
#define __S_IFDIR       0040000 /* Directory.  */
#define __S_IFCHR       0020000 /* Character device.  */
#define __S_IFBLK       0060000 /* Block device.  */
#define __S_IFREG       0100000 /* Regular file.  */
#define __S_IFIFO       0010000 /* FIFO.  */
#define __S_IFLNK       0120000 /* Symbolic link.  */
#define __S_IFSOCK      0140000 /* Socket.  */

A bit test vs IFDIR will also cover block devices. This will cause us to recurse into block devices which will probably not end well.
A bit test vs IFREG also covers IFLNK and IFSOCK. This is particularly a problem for IFLNK that are explicitly ignored here; a link to a folder will now behave differently, and the entire IFLNK branch appears to be dead.

Additionally, because some values (IFLNK in particular) have multiple bits set, testing using & doesn't test the mask correctly.

[SamuraiCrow]

OK. This should be better. The double equals checks are back in place.

[zeux]

Thanks! I think this looks good now.