|
|
Boost Testing : |
From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2006-01-03 03:23:46
Jim Douglas writes:
> Let me summarise a few points...
>
> I think all the tests so far have eliminated bzip2 and point the finger
> at tar.
>
> Aleksy, What version of tar are you using?
tar (GNU tar) 1.13.25
> How do you check that it is unpacking correctly? Are you comparing
> 'before' and 'after' trees?
Yes.
> Do you have the ability to generate an md5 sum for the tar and
> tar.bz2 files?
For the latter, sure:
http://engineering.meta-comm.com/boost/snapshot/boost-CVS-HEAD.md5
It's going to be re-calculated as a part of the tarball build
procedure. As for the former, there is no intermediate/uncompressed
tar file created on our side; the Boost tree is fed to tar to create a
tar.bz2 archive in one pass.
>
> The problem that John and I face is that tar will unpack the file
> without producing an error message but we have no way of validating the
> tree that is produced. I guess what we need is a text file produced by
> 'ls' or 'find' that we can use to validate the unpacking process.
We can surely provide you with one for the time-stamped archive that I
posted a link to earlier; here:
http://engineering.meta-comm.com/boost/snapshot/boost-CVS-HEAD-2005-12-16-17-43.txt
>
> How do you account for the fact that WinZip (9.0) will not recognise
> the tar file? I have never experienced problems with WinZip
> before. I don't think we can dismiss this factor out of hand.
Well, like I said, the archive unpacks correctly if you use the native
tools, so IMO it doesn't really matter what WinZip thinks of it. It
could be a format/compatibility issue, a bug in WinZip, or anything in
between. Plus people do experience issues with tar files and WinZip.
>
> Upgrading the version of tar under QNX is no trivial matter. I can put
> in a support request to QNX and order them to fix it ASAP but I have to
> be sure of my facts i.e. I have to be fully confident that the tar file
> you have produced is valid.
We understand that, but consider our side of the story:
1) People have been running regressions off the tarballs produced on
the same machine using the same tools and the same build procedure
for about 1,5 years by now and, truncated tarball issue aside
(http://article.gmane.org/gmane.comp.lib.boost.devel/117143),
there've been no issues with the tar archive unpacking incorrectly.
2) The same exact file that you are having problems with unpacks
correctly using a "proven" version of tar on at least two platforms
(Windows+Cygwin and Mac OS X).
3) There are several reports on the net indicating that 1.13 tar
release that you are using is in fact unreliable.
This is no proof by any means, but IMO it's indicative enough to make
the theory worth verifying. Would it be possible for you to download
the GNU tar 1.13.25 sources and just build it locally (without
upgrading the system)?
>
> Using anonymous CVS is my preferred option but right now I understand
> that it is out of service. See -
>
> http://sourceforge.net/docs/A04/en/#top
It seems to say so; however, as of right now, a comparison of the
sources obtained through anonymous vs. developer access reveals the
lap of only a couple of hours between two, which is what I would
normally expect, so the synchronization does in fact work. FWIW.
-- Aleksey Gurtovoy MetaCommunications Engineering