[PATCH] tests: chdir.test

[Dmitry V. Levin]

On Sat, Mar 12, 2016 at 01:37:20PM +0530, Jay Joshi wrote:
> On Sat, Mar 12, 2016 at 7:53 AM, Dmitry V. Levin wrote:
> > On Fri, Mar 11, 2016 at 11:22:23PM +0530, Jay Joshi wrote:
> > >
> > > This is my first attempt to add test for syscall (here, chdir). Any
> > > suggestion, comment, criticism is welcomed.
> >
> > chdir is already well covered, see
> > $ git grep chdir tests
> >
> > You must have missed my recommendation to check current coverage report
> > at https://codecov.io/github/strace/strace before adding new tests.
> >
> > I think you'd rather take something that's not tested at all.
> 
> Thanks. I couldn't understand coverage structure properly. Is coverage
> calculated manually, semi-automatically or automatically?

It's fully automatic, see travis-ci.sh script.

> > > +#if defined __NR_chdir
> >
> > Why do you think that libc wrapper for chdir syscall is not suitable?
> 
> Syscall should only be used when glibc wrapper function for that
> syscall is not available. Please correct me if I'm wrong.

Either not available or implemented using different system call(s).

> If so, what
> is the advantage of using syscall function for, say "chmod"? Please
> help me understand where to use what.

Some recently added architectures (see e.g. linux/32/syscallent.h and
linux/64/syscallent.h) provide no "chmod" syscall; on these architectures,
"chmod" libc wrapper uses "fchmodat" syscall.

This is certainly not the case for "chdir"; there seems to be no other way
to implement "chdir" besides calling "chdir" syscall.

> > > +  if (errno == ENOSYS) {
> > > +    printf("chdir(\"%s\") = -1 ENOSYS (%m)\n", sample_dir);
> >
> > chdir failing with ENOSYS?  Really?
> 
> Not sure. Where can I find whether system would have surely
> implemented a particular system call? In other words, would never
> throw ENOSYS.

Besides syscalls marked with NF flag
(see git grep -Fw NF linux/*/syscallent*.h), there is no single place that
would give you an answer for this question, which is just a special case
for a more general question how a particular system call works on
different architectures and kernel versions.

manual page usually answers the question whether the function implemented
by syscall is mandatory and when the system call was introduced in the
kernel.

linux/*/syscallent*.h answers the question whether the syscall is hooked
up in all architectures.

General understanding of linux kernel helps to find out alternative
syscalls or prove their absense.


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20160312/01ec2dc3/attachment.bin>