Re: 'cp' now uses openat etc. when copying to directories

[Paul Eggert]

On 1/15/22 11:55, Bernhard Voelker wrote:

> yep, `make check-very-expensive` fails due to 3x segmentation faults:
>
>    FAIL: tests/cp/cp-parents
>    FAIL: tests/cp/parent-perm-race
>    FAIL: tests/mv/hard-3

Yes, I think these were due to the fact that my initial tests used GCC 
with optimization levels so high that GCC optimized the bugs away (!). I 
installed commit f39a02a744ae97f6879ba1d20a528192772773e0 to try to fix 
the bugs.

> Furthermore, 'make syntax-check' complains about:
>
>    prohibit_readlink
>    src/copy.c:1266:          if (0 <= readlinkat (dst_dirfd, dst_relname, 
> dummy, sizeof dummy))
>    maint.mk: do not use readlink(at); use via xreadlink or areadlink*
>    make: *** [cfg.mk:446: sc_prohibit_readlink] Error 1

Thanks for mentioning that. I think this syntax-check is too strict, as 
I've found that it's useful to invoke readlinkat with a 1-byte 
destination buffer if one merely wants to test whether a file is a 
symbolic link and does not care where it points to. This approach can be 
better than fstatat since it needn't worry about EOVERFLOW issues. So I 
installed the attached to remove the check.

0001-build-allow-readlinkat-calls.patch
Description: Text Data