Bug/problem with 1.6.0 on UnixWare

Bug/problem with 1.6.0 on UnixWare

[Boyd Lynn Gerber]

Hello,

I am having the same problem on UnixWare as I have with the SCO OpenServer 
6. trying to build git-1.6.0.

Here is the error.

    CC shell.o
    AR compat/lib.a
    LINK git-shell
Undefined                       first referenced
symbol                              in file
hexval_table                        abspath.o
null_sha1                           abspath.o
trust_executable_bit                abspath.o
has_symlinks                        abspath.o
UX:ld: ERROR: Symbol referencing errors. No output written to git-shell
gmake: *** [git-shell] Error 1

I need some assistance to track this down.  I did not have any error with 
the tests I did during the testing.  I did not test often on all OS's.

Thanks,

--
Boyd Gerber <gerberb@zenez.com>
ZENEZ	1042 East Fort Union #135, Midvale Utah  84047

Re: Bug/problem with 1.6.0 on UnixWare

[Mike Ralphson]

2008/8/21 Boyd Lynn Gerber <gerberb@zenez.com>:
> I am having the same problem on UnixWare as I have with the SCO OpenServer
> 6. trying to build git-1.6.0.
>
> Here is the error.
>
>    CC shell.o
>    AR compat/lib.a
>    LINK git-shell
> Undefined                       first referenced
> symbol                              in file
> hexval_table                        abspath.o
> null_sha1                           abspath.o
> trust_executable_bit                abspath.o
> has_symlinks                        abspath.o
> UX:ld: ERROR: Symbol referencing errors. No output written to git-shell
> gmake: *** [git-shell] Error 1

I think you've hit the problem described here:
http://article.gmane.org/gmane.comp.version-control.git/92945/match=git+shell

Junio's revert of the offending commit is I think in the 'next' branch.

You could also look at reverting 5b8e6f85f97 locally?

Cheers, Mike

Re: Bug/problem with 1.6.0 on UnixWare

[Boyd Lynn Gerber]

On Thu, 21 Aug 2008, Mike Ralphson wrote:
> 2008/8/21 Boyd Lynn Gerber <gerberb@zenez.com>:
> > I am having the same problem on UnixWare as I have with the SCO OpenServer
> > 6. trying to build git-1.6.0.
> >
> > Here is the error.
> >
> >    CC shell.o
> >    AR compat/lib.a
> >    LINK git-shell
> > Undefined                       first referenced
> > symbol                              in file
> > hexval_table                        abspath.o
> > null_sha1                           abspath.o
> > trust_executable_bit                abspath.o
> > has_symlinks                        abspath.o
> > UX:ld: ERROR: Symbol referencing errors. No output written to git-shell
> > gmake: *** [git-shell] Error 1
> 
> I think you've hit the problem described here:
> http://article.gmane.org/gmane.comp.version-control.git/92945/match=git+shell
> 
> Junio's revert of the offending commit is I think in the 'next' branch.

Yep, that sounds about right.  It has bitten me on the 12 platforms, I 
submitted to get git working on them.  I will look at reverting it.  

I just tried to get this working and now I am being flooded with emails 
from people that are trying git for the first time, and asking...

How I could your recommend such a broken SCM?  I told them to got back to 
the 1.5.X but being new to git and then not being able to build and use it 
is causing a lot of flack.  I am getting emails now about my recommending 
git, as it is broken.  I should not have announce to the various lists 
about how great git is and that they should dump their older SCM's in 
favor of git.  Really bad timing on my part.  I just hope these people 
will give git an other try.

Thanks,

--
Boyd Gerber <gerberb@zenez.com>
ZENEZ	1042 East Fort Union #135, Midvale Utah  84047

Re: Bug/problem with 1.6.0 on UnixWare

[Mike Ralphson]

2008/8/21 Boyd Lynn Gerber <gerberb@zenez.com>:
> Yep, that sounds about right.  It has bitten me on the 12 platforms, I
> submitted to get git working on them.  I will look at reverting it.
>
> I just tried to get this working and now I am being flooded with emails
> from people that are trying git for the first time, and asking...
>
> How I could your recommend such a broken SCM?  I told them to got back to
> the 1.5.X but being new to git and then not being able to build and use it
> is causing a lot of flack.  I am getting emails now about my recommending
> git, as it is broken.  I should not have announce to the various lists
> about how great git is and that they should dump their older SCM's in
> favor of git.  Really bad timing on my part.  I just hope these people
> will give git an other try.

If it's possible, you might want to look at automatically building and
running the git testsuite on some or all of your SCO etc platforms.
You're very welcome to push tags describing the state of the build and
tests to http://repo.or.cz/w/git/gitbuild.git - there's a few scripts
there in a side branch.

Personally, I'd like it if Junio could check to see if the 'satellite'
platforms are happily building and passing the last rc before he tags
a final release (with no additional code changes in it!), but that may
be too much of the tail wagging the dog.

Even though 1.6 builds and passes here, I probably won't push it out
in my environment until 1.6.0.2 or so.

Mike

Re: Bug/problem with 1.6.0 on UnixWare

[Jeff King]

On Thu, Aug 21, 2008 at 01:41:07PM +0100, Mike Ralphson wrote:

> If it's possible, you might want to look at automatically building and
> running the git testsuite on some or all of your SCO etc platforms.
> You're very welcome to push tags describing the state of the build and
> tests to http://repo.or.cz/w/git/gitbuild.git - there's a few scripts
> there in a side branch.

BTW, on my todo list is pushing my tags there. I am building FreeBSD
nightly, and I plan to add Solaris 2.8. I just haven't gotten around to
making the build work yet...:)

> Personally, I'd like it if Junio could check to see if the 'satellite'
> platforms are happily building and passing the last rc before he tags
> a final release (with no additional code changes in it!), but that may
> be too much of the tail wagging the dog.

I think his procedure is more reactive. He tags 1.6.0-rc*, and he
expects us to complain to him if it is broken. The real deficiency here
is that the broken platforms were not tested.

-Peff

Re: Bug/problem with 1.6.0 on UnixWare

[Mike Ralphson]

2008/8/21 Jeff King <peff@peff.net>:
> On Thu, Aug 21, 2008 at 01:41:07PM +0100, Mike Ralphson wrote:
>
>> You're very welcome to push tags describing the state of the build and
>> tests to http://repo.or.cz/w/git/gitbuild.git
>
> BTW, on my todo list is pushing my tags there. I am building FreeBSD
> nightly, and I plan to add Solaris 2.8. I just haven't gotten around to
> making the build work yet...:)

Great stuff. Though I think it might provoke some patches to improve
the display of multiple long tag names for the same commit in gitweb!

>> Personally, I'd like it if Junio could check to see if the 'satellite'
>> platforms are happily building and passing the last rc before he tags
>> a final release (with no additional code changes in it!), but that may
>> be too much of the tail wagging the dog.
>
> I think his procedure is more reactive. He tags 1.6.0-rc*, and he
> expects us to complain to him if it is broken.

Granted, but Junio seems to do releases at the weekend, and sometimes
they aren't the same content as the previous rc. Though I build and
test 7 days a week, I'd rarely look at the test results over the
weekend. And I was on vacation when the last minute AIX breakage went
into 1.5.6 - not that I would necessarily have been able to fix that
one.

Mike

Re: Bug/problem with 1.6.0 on UnixWare

[Jeff King]

On Thu, Aug 21, 2008 at 03:38:32PM +0100, Mike Ralphson wrote:

> Great stuff. Though I think it might provoke some patches to improve
> the display of multiple long tag names for the same commit in gitweb!

Heh. Can you pass along the final version of your build script? I want
to make sure we are doing the same things.

> Granted, but Junio seems to do releases at the weekend, and sometimes
> they aren't the same content as the previous rc. Though I build and
> test 7 days a week, I'd rarely look at the test results over the
> weekend. And I was on vacation when the last minute AIX breakage went
> into 1.5.6 - not that I would necessarily have been able to fix that
> one.

Yes, in theory, last minute code changes could cause a portability
breakage. However, any changes from the last -rc to the release version
tend to be extremely conservative (as judged by Junio), so in practice
I don't think it happens. For example, I think this git-shell bug made
it into master over a month before release.

-Peff

Re: Bug/problem with 1.6.0 on UnixWare

[Mike Ralphson]

2008/8/21 Jeff King <peff@peff.net>:
> On Thu, Aug 21, 2008 at 03:38:32PM +0100, Mike Ralphson wrote:
>
>> Great stuff. Though I think it might provoke some patches to improve
>> the display of multiple long tag names for the same commit in gitweb!
>
> Heh. Can you pass along the final version of your build script? I want
> to make sure we are doing the same things.

http://repo.or.cz/w/git/gitbuild.git?a=blob;f=gitbuild.sh;h=d06fbef3c7aa54ea3beab4b3eed4d73387f3b136;hb=2da51f87b94ef564691a656f91b97472bf6d8c53

The setting of the PATH and GIT_SKIP_TESTS are obviously env. specific.

At the moment there's still a bit of manual faffing around when next
is rewound after a release.

> Yes, in theory, last minute code changes could cause a portability
> breakage. However, any changes from the last -rc to the release version
> tend to be extremely conservative (as judged by Junio), so in practice
> I don't think it happens. For example, I think this git-shell bug made
> it into master over a month before release.

True, this instance isn't a last minute breakage.

Mike