Re: [posixtest-discuss] Cumulative patch for aio_suspend and aio_writetestsuites

[Erik H. <eh...@su...>]

Hello,

>  We really need to remove the signal within some aio_suspend test cases=
. since each asynchronous IO operation may possibly provoke signal when i=
t finish, the error is EINTR instead of EAGAIN.

I have rewritten aio-suspend/9-1.c and aio-suspend/4-1.c testsuites witho=
ut signals.
These are attached to this mail.

> For aio-suspend/9-1.c, it is used to test if no aio indicated in the li=
st completed before timeout, aio_suspend should fail with EAGAIN. So we s=
till need to rewrite it without signal.  It is very strange that aio_susp=
end/4-1.c pass while aio_suspend/9-1.c faild on your side. Could you remo=
ve the signal and re-run it on your machine?

Original versions of these tests didn't pass. Only after my patch which w=
ipes
out at least the sigrt1_handler. (I have wiped out both signals and signa=
l handlers)

> For conformance/interfaces/aio_write/4-1.c, according the codes, the re=
sult should be always UNSUPPORTED or UNTESTED.=20

I can't found in code, how conformance/interfaces/aio_write/4-1.c can end=
s with
UNTESTED. This testsuite always pass on my machine. Only problem is when =
stdout is
redirected to file, which is used by testsuite loader.
Here is the example, how it behaves:

...conformance/interfaces/aio_write 0$./4-1
Test PASSED
...conformance/interfaces/aio_write 0$rm /tmp/test000000001
...conformance/interfaces/aio_write 0$./4-1  >>/tmp/test000000001
...conformance/interfaces/aio_write 0$dd if=3D/dev/zero bs=3D1 count=3D51=
2 >> /tmp/test000000001 2>/dev/null
...conformance/interfaces/aio_write 0$./4-1  >>/tmp/test000000001
File size limit exceeded
...conformance/interfaces/aio_write 153$

The last numger in prompt is return value of executed command.

Have a nice day
Yokotashi
=20
> Really appreciate you help on posixtestsuite. I have checked in your pa=
tch partially. thank you very much!
>=20
> Elva Fu=20
>=20
> -----Original Message-----
> From: pos...@li... [mailto:posixtest=
-dis...@li...] On Behalf Of Erik Hamera
> Sent: 2006=E5=B9=B410=E6=9C=8825=E6=97=A5 20:29
> To: pos...@li...
> Subject: [posixtest-discuss] Cumulative patch for aio_suspend and aio_w=
ritetestsuites
>=20
> Hello,
> I have repaired some aio_suspend() testsuites, which fails always on:
> "aio_suspend() should set errno to EAGAIN: 4 (Interrupted system call)"=
,
> because signal, which is used for notify interrupts the aio_suspend()
> function. I rewrote these for operating without this signal.
>=20
> The aio_suspend/9-1.c testsuite cannot pass on my machine, because it
> suspends for at least one jiffie and my AMD Athlon 64 3500+ can done th=
e
> operation before suspended testsuite gets context back. My patch is
> -#define BUF_SIZE       1024*1024
> +#define BUF_SIZE       1024*1024*8
> I don't know how make this better.
>=20
On Wed, Nov 22, 2006 at 02:23:11PM +0800, Fu, Elva wrote:
> The aio_write/4-1.c testsuite cannot pass when is run from loader, but =
can
> pass alone. The rlimit has been set back to -1 (no limit) between aio_w=
rite()
> and printf ("Test PASSED\n");.
>=20
> Thank you all for working on testsuites and have a nice day
>=20
> Yokotashi
>=20