Eric Blake wrote:
> Jim, what do you think of this patch, so that lchmodat only fails with
> ENOSYS when called on a symlink (when called on other file types, it would
> now succeed)?
>
> Also, should "openat.h" provide shortcuts named:
>
> accessat() => faccessat(,0)
> euidaccessat() => faccessat(,AT_EACCESS)
> statat() => fstatat(,0)
> lstatat() => fstatat(,AT_SYMLINK_NOFOLLOW)
>
> to match the existing shortcuts of chownat, lchownat, etc.
Good idea.
I hesitated to appropriate the names initially, but the combination
of increased readability and orthogonality make it worthwhile.
> diff --git a/ChangeLog b/ChangeLog
> index 06287d4..03e91c9 100644
> --- a/ChangeLog
> +++ b/ChangeLog
> @@ -1,5 +1,9 @@
> 2009-09-06 Eric Blake <e...@byu.net>
>
> + openat: with no lchmod, only fail lchmodat on symlinks
> + * lib/fchmodat.c (lchmod) [!HAVE_LCHMOD]: Perform lstat first, to
> + minimize failure cases.
> +
> renameat: new module
> * modules/renameat: New file.
> * lib/at-func2.c: Likewise.
> diff --git a/lib/fchmodat.c b/lib/fchmodat.c
> index 53cafe0..11ace83 100644
> --- a/lib/fchmodat.c
> +++ b/lib/fchmodat.c
> @@ -30,7 +30,16 @@
> system-supplied declaration. */
> # undef lchmod
> # define lchmod lchmod_rpl
> -static int lchmod (char const *f, mode_t m) { errno = ENOSYS; return -1; }
> +static int
> +lchmod (char const *f, mode_t m)
> +{
> + /* This is slightly racy, but better than nothing. */
> + struct stat st;
> + if (lstat (f, &st) == -1 || !S_ISLNK (st.st_mode))
> + return chmod (f, m);
> + errno = ENOSYS;
> + return -1;
> +}
Is this useful to you on Cygwin? Fewer diagnostics?
If so, go ahead.
The drawback is that it makes it harder
to perform a run-time test for lchmod support.
The test requires a symlink.
More importantly, the test writer would need to know
about this half-functional implementation.
Otherwise, an application that doesn't know about this
new feature/requirement risks determining that
lchmod support is available when it is not.
So maybe a word in NEWS or documentation.
