Re: PATCH: prompt escape tests

Re: PATCH: prompt escape tests

[Sven Wischnowsky]

Peter Stephenson wrote:

> If I remember, there was a failed test reported by Sven to do with
> different system behaviour which hasn't been fixed up yet.

I wanted to mention that again anyway... A collegue noticed this for
some newer Solaris version, too: nowadays some systems don't update
the atime inode field anymore when a file is read, which makes the -N
condition test fail.

Bye
 Sven


--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de

Re: PATCH: prompt escape tests

[Tanaka Akira]

In article <200002020834.JAA09905@beta.informatik.hu-berlin.de>,
  Sven Wischnowsky <wischnow@informatik.hu-berlin.de> writes:

> I wanted to mention that again anyway... A collegue noticed this for
> some newer Solaris version, too: nowadays some systems don't update
> the atime inode field anymore when a file is read, which makes the -N
> condition test fail.

Z:akr@thorax% uname
Linux
Z:akr@thorax% df .
Filesystem           1k-blocks      Used Available Use% Mounted on
crane:/ldl3            3939604   1660770   2274895  42% /amd/crane/root/ldl3
Z:akr@thorax% make check
...
Test ./07cond.ztst failed: bad status 1, expected 0 from:
  cat unmodified
  touch newnewnew
  [[ -N newnewnew && ! -N unmodified ]]
Was testing: -N cond
./07cond.ztst: test failed.
...

This problem is not Solaris specific and can be reproduced on Linux.
I think it is caused by NFS.
-- 
Tanaka Akira

Re: PATCH: prompt escape tests

[Dan Nelson]

In the last episode (Feb 03), Tanaka Akira said:
> In article <200002020834.JAA09905@beta.informatik.hu-berlin.de>,
>   Sven Wischnowsky <wischnow@informatik.hu-berlin.de> writes:
> 
> > I wanted to mention that again anyway... A collegue noticed this
> > for some newer Solaris version, too: nowadays some systems don't
> > update the atime inode field anymore when a file is read, which
> > makes the -N condition test fail.
>
> This problem is not Solaris specific and can be reproduced on Linux.
> I think it is caused by NFS.

I think it's a POSIX requirement that atime must be updated.
 
Lots of unixes do let let you disable atime updates though (usually by
mount -o noatime).  It's useful for news servers or FTP archives that
do a lot of reading and don't really care about the atime field.  I
looked on my Solaris 2.6 box and there doesn't seem to be a way to
disable atime updates.  7 or 8 might very well be able to, though.

-- 
	Dan Nelson
	dnelson@emsphone.com

atime (Re: PATCH: prompt escape tests)

[Tanaka Akira]

In article <20000202105104.D31919@dan.emsphone.com>,
  Dan Nelson <dnelson@emsphone.com> writes:

> I think it's a POSIX requirement that atime must be updated.

Hm.

> Lots of unixes do let let you disable atime updates though (usually by
> mount -o noatime).  It's useful for news servers or FTP archives that
> do a lot of reading and don't really care about the atime field.  I
> looked on my Solaris 2.6 box and there doesn't seem to be a way to
> disable atime updates.  7 or 8 might very well be able to, though.

It's server side config and not related to this problem.

Linux produces NFS requests by `touch t':

02:26:54.406581 thorax.jaist.ac.jp.1258203975 > crane.nfs: 112 lookup fh 32,10/568960 "t" (ttl 62, id 8982)
02:26:54.408238 thorax.jaist.ac.jp.1258203976 > crane.nfs: 144 create fh 32,10/568960 "t" (ttl 62, id 8983)
02:26:54.437421 thorax.jaist.ac.jp.1258203977 > crane.nfs: 136 setattr fh 32,10/569010 (ttl 62, id 8986)

and by `cat t':

02:27:08.065326 thorax.jaist.ac.jp.1258203978 > crane.nfs: 112 lookup fh 32,10/568960 "t" (ttl 62, id 9002)
02:27:08.067635 thorax.jaist.ac.jp.1258203979 > crane.nfs: 104 getattr fh 32,10/569010 (ttl 62, id 9003)

Some requests (especialy requests by `cat t') are omitted by cache.
Anyway, the server cannot updates atime since the client doesn't send
read requests.
-- 
Tanaka Akira