You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Tobias Ringström <to...@ringstrom.mine.nu> on 2004/03/26 12:09:23 UTC

Using pax instead of cpio in dist.sh

Jostein, is it OK by you if I merge r9079 from trunk to 1.0.x, or would 
you rather prefer to do it yourself?  It needs approval from you if I 
understand HACKING correctly.  The log message is below.

/Tobias

------------------------------------------------------------------------
r9079 | bliss | 2004-03-16 15:57:11 +0100 (Tue, 16 Mar 2004) | 7 lines

Use the POSIX archiver pax instead of cpio to produce tarballs.
GNU cpio is broken because it creates incorrect timestamps. See
http://www.contactor.se/~dast/svn/archive-2004-03/0845.shtml

* dist.sh
   Use pax instead of cpio to produce the tarballs.
------------------------------------------------------------------------


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Using pax instead of cpio in dist.sh

Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Ben Reser wrote:
> I just haven't gotten around to producing tarballs this way to post to
> have people test them for problems.  If someone else doesn't get around
> to it, I'll do this tomorrow.  Once we've done that and caught most of
> the major platforms I'll be happy to see that r9079 gets merged.  Here's
> my list of platforms I want to see tested:
> 
> Linux

Works. (RHL9, FC1)

> Solaris

Works. (Solaris 8 Sparc)

> Windows (WinZip - though I believe it still won't work, it'd be nice to
>          know)

Does not work, at least the version I have.  It was pointed out in some 
other mail that it would be nice to have a zip file for the Windows 
users where all source files have CRLF line endings.  Adding some or all 
of the components that are required for a Windows build would also be nice.

/Tobias

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Tobias Ringstrom <to...@ringstrom.mine.nu>.
Ben Reser wrote:

>>The only reason to test pax would be to test for bugs in the one that 
>>will be used to produce the official tarball.  If five different pax 
>>implementations and versions will be used, they should either all be 
>>tested, or none of them needs to be tested.  Pax produces USTAR files. 
>>That is the purpose of pax.  There is no reason to test the pax in RHL9 
>>if you will use OS X to produce the tarballs.  That was all I meant. 
>>That said, I think we can trust pax to be correct, and we don't need to 
>>test it.  I'm not against testing it, of course.  I just wanted to 
>>explain why I though that it was unneccessary to test it.
> 
> Given that we've now tried 3 different tar implementations looking for
> one that produces portable and correct tar archives and so far we
> haven't found one...  I'm not quite as confident of this as you are.

The best thing would be to have one dedicated release build machine to 
make sure that our releases are of a consistent (high) quality.  So far 
we've have new dist bugs with almost every release.  :-(

Will you really send out new test tarballs each time you build the 
release on a new machine?  In case it's not possible to have a dedicated 
release build machine, I was trying to suggest that the probablity of a 
pax bug is low enough that we can risk a release, and if there is a pax 
bug on that build machine, we can quickly release another release. 
Release numbers are relatively cheap, and the probability is low.

/Tobias


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Ben Reser <be...@reser.org>.
On Fri, Apr 02, 2004 at 10:57:04PM -0800, Ben Reser wrote:
> *cough* I think you're a little too confident here... I tried pax
> tonight:
> 
> $ gzip -dc subversion-1.0.1.tar.gz | tar xvf -
> tar: This does not look like a tar archive
> tar: Skipping to next header
> tar: Archive contains obsolescent base-64 headers
> tar: Error exit delayed from previous errors
> 
> file doesn't even think it is a tar file:
> $ file subversion-1.0.1.tar
> subversion-1.0.1.tar: data
> 
> I'll try it on a different machine later.  I tried this on amd64 and
> it's possible pax doesn't work righ ton this platform for whatever
> reason (i.e. a bug).

Okay found the issue.  It was a minor error in the dist.sh script that
probably wouldn't show up in most peoples environments.  It was
capturing the output of the cd command.  Since I have $CDPATH set the cd
was ambiguous so bash spat out what dir it really changed to.  Commit
to fix coming up on trunk soon.  

I'll post a new thread with URLs to test tarballs.

-- 
Ben Reser <be...@reser.org>
http://ben.reser.org

"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Ben Reser <be...@reser.org>.
On Tue, Mar 30, 2004 at 12:00:43AM +0200, Tobias Ringstrom wrote:
> What I was trying to say was that we don't need to test if the USTAR 
> format works.  We did that when we started to use cpio instead of GNU 
> tar to create the tarballs.  We will just use another tool to create the 
> same thing -- hopefully a tool with no bugs.

*cough* I think you're a little too confident here... I tried pax
tonight:

$ gzip -dc subversion-1.0.1.tar.gz | tar xvf -
tar: This does not look like a tar archive
tar: Skipping to next header
tar: Archive contains obsolescent base-64 headers
tar: Error exit delayed from previous errors

file doesn't even think it is a tar file:
$ file subversion-1.0.1.tar
subversion-1.0.1.tar: data

I'll try it on a different machine later.  I tried this on amd64 and
it's possible pax doesn't work righ ton this platform for whatever
reason (i.e. a bug).

> The only reason to test pax would be to test for bugs in the one that 
> will be used to produce the official tarball.  If five different pax 
> implementations and versions will be used, they should either all be 
> tested, or none of them needs to be tested.  Pax produces USTAR files. 
> That is the purpose of pax.  There is no reason to test the pax in RHL9 
> if you will use OS X to produce the tarballs.  That was all I meant. 
> That said, I think we can trust pax to be correct, and we don't need to 
> test it.  I'm not against testing it, of course.  I just wanted to 
> explain why I though that it was unneccessary to test it.

Given that we've now tried 3 different tar implementations looking for
one that produces portable and correct tar archives and so far we
haven't found one...  I'm not quite as confident of this as you are.

> The decision (or even discussion) that we shouldn't have a dedicated RM 
> did not happen publicly, and the outcome is not publicly available 
> except in the form of little hints, so who can be blamed for missing 
> that point?

Well I don't think any decision has been made.  Justin and I both have
that as a goal.  If it can be attained remains to be seen.

-- 
Ben Reser <be...@reser.org>
http://ben.reser.org

"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Tobias Ringstrom <to...@ringstrom.mine.nu>.
Ben Reser wrote:
> On Mon, Mar 29, 2004 at 11:07:22AM -0800, Justin Erenkrantz wrote:
> 
>>I think you've missed the point that we shouldn't have an 'official' build 
>>machine nor a dedicated RM.  Note that if I do a roll, it'll likely either 
>>use OS X's pax or Solaris 9's pax.
> 
> This is how I feel as well.  If i do a release it'll be with a GNU pax
> on Mandrake.

What I was trying to say was that we don't need to test if the USTAR 
format works.  We did that when we started to use cpio instead of GNU 
tar to create the tarballs.  We will just use another tool to create the 
same thing -- hopefully a tool with no bugs.

The only reason to test pax would be to test for bugs in the one that 
will be used to produce the official tarball.  If five different pax 
implementations and versions will be used, they should either all be 
tested, or none of them needs to be tested.  Pax produces USTAR files. 
That is the purpose of pax.  There is no reason to test the pax in RHL9 
if you will use OS X to produce the tarballs.  That was all I meant. 
That said, I think we can trust pax to be correct, and we don't need to 
test it.  I'm not against testing it, of course.  I just wanted to 
explain why I though that it was unneccessary to test it.

The decision (or even discussion) that we shouldn't have a dedicated RM 
did not happen publicly, and the outcome is not publicly available 
except in the form of little hints, so who can be blamed for missing 
that point?

/Tobias

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Ben Reser <be...@reser.org>.
On Mon, Mar 29, 2004 at 11:07:22AM -0800, Justin Erenkrantz wrote:
> I think you've missed the point that we shouldn't have an 'official' build 
> machine nor a dedicated RM.  Note that if I do a roll, it'll likely either 
> use OS X's pax or Solaris 9's pax.

This is how I feel as well.  If i do a release it'll be with a GNU pax
on Mandrake.

> But, I'm way behind on email and need to get myself out of that hole before 
> I can be of further constructive help to this discussion.

And ironically I'm in the same position...

-- 
Ben Reser <be...@reser.org>
http://ben.reser.org

"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Monday, March 29, 2004 8:51 PM +0200 Tobias Ringström 
<to...@ringstrom.mine.nu> wrote:

> The best thing would of course be to run trunk's dist.sh on the machine
> that you guys plan to roll the release with.  We already tested that the
> USTAR format was OK when we switched from tar to cpio.  If we are to look
> for pax bugs, we should look at the official build machine's pax.

I think you've missed the point that we shouldn't have an 'official' build 
machine nor a dedicated RM.  Note that if I do a roll, it'll likely either 
use OS X's pax or Solaris 9's pax.

But, I'm way behind on email and need to get myself out of that hole before 
I can be of further constructive help to this discussion.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org


Re: Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Justin Erenkrantz wrote:
> Agreed.  If someone posts a link to such a tarball created with pax, I 
> can catch everything but HP/UX, AIX, and Win32.

The best thing would of course be to run trunk's dist.sh on the machine 
that you guys plan to roll the release with.  We already tested that the 
USTAR format was OK when we switched from tar to cpio.  If we are to 
look for pax bugs, we should look at the official build machine's pax.

/Tobias

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Testing pax on platforms was Re: Using pax instead of cpio in dist.sh

Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Sunday, March 28, 2004 10:23 AM -0800 Ben Reser <be...@reser.org> wrote:

> I just haven't gotten around to producing tarballs this way to post to
> have people test them for problems.  If someone else doesn't get around
> to it, I'll do this tomorrow.  Once we've done that and caught most of
> the major platforms I'll be happy to see that r9079 gets merged.  Here's
> my list of platforms I want to see tested:
>
> Linux
> (Free|Open)BSD
> OS X
> Solaris
> HP/UX
> AIX
> Windows (WinZip - though I believe it still won't work, it'd be nice to
>          know)

Agreed.  If someone posts a link to such a tarball created with pax, I can 
catch everything but HP/UX, AIX, and Win32.

As we've discussed before, Win32 shouldn't be using tarballs in the first 
place.  I'll be shooting to take a first pass at redoing dist.sh this week 
or next.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Using pax instead of cpio in dist.sh

Posted by Ben Reser <be...@reser.org>.
On Fri, Mar 26, 2004 at 01:09:23PM +0100, Tobias_Ringström wrote:
> Jostein, is it OK by you if I merge r9079 from trunk to 1.0.x, or would 
> you rather prefer to do it yourself?  It needs approval from you if I 
> understand HACKING correctly.  The log message is below.
> 
> /Tobias
> 
> ------------------------------------------------------------------------
> r9079 | bliss | 2004-03-16 15:57:11 +0100 (Tue, 16 Mar 2004) | 7 lines
> 
> Use the POSIX archiver pax instead of cpio to produce tarballs.
> GNU cpio is broken because it creates incorrect timestamps. See
> http://www.contactor.se/~dast/svn/archive-2004-03/0845.shtml
> 
> * dist.sh
>   Use pax instead of cpio to produce the tarballs.
> ------------------------------------------------------------------------

I just haven't gotten around to producing tarballs this way to post to
have people test them for problems.  If someone else doesn't get around
to it, I'll do this tomorrow.  Once we've done that and caught most of
the major platforms I'll be happy to see that r9079 gets merged.  Here's
my list of platforms I want to see tested:

Linux
(Free|Open)BSD
OS X
Solaris
HP/UX
AIX
Windows (WinZip - though I believe it still won't work, it'd be nice to
         know)


-- 
Ben Reser <be...@reser.org>
http://ben.reser.org

"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Using pax instead of cpio in dist.sh

Posted by kf...@collab.net.
"Jostein Chr. Andersen" <jo...@josander.net> writes:
> Justin Erenkrantz and Ben Reser are the new RMs. The RM hat was passed to 
> them after my brodband problems and also because their C knowledge 
> (among other things).

I should point out that Justin and Ben have expressed (at least in
private mail, but I don't think they'll mind if I share it here) a
desire to streamline & automate the release process much more than it
has been, so that making a release is not such a big deal -- that is,
so anyone can wear the RM hat for a given release when necessary.

Still probably a good idea for them to vet changes like this one; I
just wanted to convey their long-term goal here.

-Karl

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Using pax instead of cpio in dist.sh

Posted by Tobias Ringström <to...@ringstrom.mine.nu>.
Jostein Chr. Andersen wrote:
> On Friday 26 March 2004 13.09, Tobias Ringström wrote:
> 
>>Jostein, is it OK by you if I merge r9079 from trunk to 1.0.x, or
>>would you rather prefer to do it yourself?  It needs approval from you
>>if I understand HACKING correctly.  The log message is below.
> 
> Justin Erenkrantz and Ben Reser are the new RMs. The RM hat was passed to 
> them after my brodband problems and also because their C knowledge 
> (among other things).

Oh, that info is probably in my huge dev@ backlog.  (I'm soooo happy 
that dev@ was split into dev@ and users@.)  Thanks for letting me know!

/Tobias

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: Using pax instead of cpio in dist.sh

Posted by "Jostein Chr. Andersen" <jo...@josander.net>.
On Friday 26 March 2004 13.09, Tobias Ringström wrote:
> Jostein, is it OK by you if I merge r9079 from trunk to 1.0.x, or
> would you rather prefer to do it yourself?  It needs approval from you
> if I understand HACKING correctly.  The log message is below.

Justin Erenkrantz and Ben Reser are the new RMs. The RM hat was passed to 
them after my brodband problems and also because their C knowledge 
(among other things).

Jostein

-- 
http://www.josander.net/kontakt/ ||
http://www.josander.net/en/contact/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org