gdb_7_3-branch not buildable?

gdb_7_3-branch not buildable?

[Jim Meyering]

Tom Tromey mentioned that the git mirror has lost sync with the cvs one,
which was causing a build failure, so I went to compare them.
I did the following on sourceware.org:
(these snippets may not be 100% accurate, because they're
partially reconstructed after the fact, but you get the idea)

# First get a cvs dir:
cd /tmp
mkdir .j && cd .j
(
  wget ftp://sourceware.org/pub/gdb/snapshots/current/gdb-weekly-CVS-7.3.50.20110412.tar.bz2
  bzip2 -dc gdb-weekly-CVS-7.3.50.20110412.tar.bz2 | tar xf -
  mv src gdb-cvs
  cd gdb-cvs
  cvs up -r gdb_7_3-branch
)

# Now a git dir:
(
  git clone /git/gdb.git && cd gdb &&
    git br --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
)

# Compare them:
diff -u -r -x.git -xCVS gdb gdb-cvs > k

$ diffstat k
 gdb-cvs/ChangeLog                               |   71 ------
 gdb-cvs/Makefile.in                             |    2
 gdb-cvs/Makefile.tpl                            |    2
 gdb-cvs/bfd/ChangeLog                           |    4
 gdb-cvs/bfd/elfxx-ia64.c                        |    1
 gdb-cvs/bfd/version.h                           |    2
 gdb-cvs/configure                               |  283 ++++++++++++++----------
 gdb-cvs/configure.ac                            |  282 ++++++++++++++---------
 gdb-cvs/contrib                                 |only
 gdb-cvs/gdb/ada-lang.c                          |   90 ++-----
 gdb-cvs/gdb/arm-linux-tdep.c                    |   38 +--
 gdb-cvs/gdb/arm-tdep.c                          |  112 +++------
 gdb-cvs/gdb/arm-tdep.h                          |    3
 gdb-cvs/gdb/breakpoint.h                        |    4
 gdb-cvs/gdb/doc/ChangeLog                       |   10
 gdb-cvs/gdb/doc/gdb.texinfo                     |    8
 gdb-cvs/gdb/doc/stabs.texinfo                   |    2
 gdb-cvs/gdb/gdb.1                               |    2
 gdb-cvs/gdb/testsuite/gdb.ada/arrayptr.exp      |    9
 gdb-cvs/gdb/testsuite/gdb.ada/arrayptr/foo.adb  |    7
 gdb-cvs/gdb/testsuite/gdb.cp/cpexprs.exp        |    7
 gdb-cvs/gdb/utils.c                             |    6
 gdb-cvs/sim/bfin/ChangeLog                      |    5
 gdb-cvs/sim/bfin/dv-bfin_otp.c                  |    7
 gdb-cvs/sim/common/ChangeLog                    |   25 --
 gdb-cvs/sim/common/dv-glue.c                    |   41 ---
 gdb-cvs/sim/common/hw-alloc.c                   |    3
 gdb-cvs/sim/common/hw-base.c                    |   30 +-
 gdb-cvs/sim/common/hw-device.h                  |    9
 gdb-cvs/sim/common/hw-events.c                  |    6
 gdb-cvs/sim/common/hw-handles.c                 |    6
 gdb-cvs/sim/common/hw-instances.c               |    3
 gdb-cvs/sim/common/hw-instances.h               |    3
 gdb-cvs/sim/common/hw-main.h                    |    6
 gdb-cvs/sim/common/hw-ports.c                   |    9
 gdb-cvs/sim/common/hw-ports.h                   |    3
 gdb-cvs/sim/common/hw-properties.c              |    3
 gdb-cvs/sim/common/hw-properties.h              |   15 -
 gdb-cvs/sim/common/hw-tree.c                    |  100 +++-----
 gdb/.gitignore                                  |only
 gdb/gdb/testsuite/gdb.ada/mi_catch_ex.exp       |only
 gdb/gdb/testsuite/gdb.ada/mi_catch_ex/foo.adb   |only
 gdb/gdb/testsuite/gdb.arch/thumb-singlestep.S   |only
 gdb/gdb/testsuite/gdb.arch/thumb-singlestep.exp |only
 44 files changed, 536 insertions(+), 683 deletions(-)

That's many more differences than I expected,
so maybe something is going wrong with the mirroring
mechanism (note that git's top-level ChangeLog has four
commits from april 4 that are not in git-cvs's ChangeLog file).

Re: gdb_7_3-branch not buildable?

[Andreas Schwab]

Jim Meyering <jim@meyering.net> writes:

> That's many more differences than I expected,
> so maybe something is going wrong with the mirroring
> mechanism (note that git's top-level ChangeLog has four
> commits from april 4 that are not in git-cvs's ChangeLog file).

git-cvsimport can easily be confused by a branch that is created from a
point back in time.  That's a fundamental restriction of cvsps, because
it only tells you about the revision of files that changed on the
branch, but you never know the revision of the files that are unchanged.
To fix that you have to manually reset the branch to the true
branchpoint.  See <git://repo.or.cz/src> for a more precise rendition of
the src repository.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: gdb_7_3-branch not buildable?

[Joel Brobecker]

> git-cvsimport can easily be confused by a branch that is created from a
> point back in time.  That's a fundamental restriction of cvsps, because
> it only tells you about the revision of files that changed on the
> branch, but you never know the revision of the files that are unchanged.
> To fix that you have to manually reset the branch to the true
> branchpoint.  See <git://repo.or.cz/src> for a more precise rendition of
> the src repository.

How about we just delete the 7.3 branch, and recreated from
the current sources? The fact that I made a mistake on the date
as well as the fact that back-dated branchpoints confuse git
are two arguments in favor of starting over...

-- 
Joel

Re: gdb_7_3-branch not buildable?

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

Joel> How about we just delete the 7.3 branch, and recreated from
Joel> the current sources?

We've already put stuff on CVS trunk that is probably not 7.3 material.

Can we just fix sourceware's gdb.git along the same lines as Andreas's
repository?  Andreas, could you say what you did to make yours better?

Tom

Re: gdb_7_3-branch not buildable?

[Andreas Schwab]

Tom Tromey <tromey@redhat.com> writes:

> Andreas, could you say what you did to make yours better?

Handcraft.  These are things that something based on cvsps cannot do
better.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: gdb_7_3-branch not buildable?

[Joel Brobecker]

> > Andreas, could you say what you did to make yours better?
> 
> Handcraft.  These are things that something based on cvsps cannot do
> better.

Can we take your branch and force-push it? If we could do that
without confusing cvsps, that'd be awesome. git really has become
a significant time saver, and not having a correct mirror of the
branch would mean a lot of work on my end.

-- 
Joel

Re: gdb_7_3-branch not buildable?

[Andreas Schwab]

Joel Brobecker <brobecker@adacore.com> writes:

>> > Andreas, could you say what you did to make yours better?
>> 
>> Handcraft.  These are things that something based on cvsps cannot do
>> better.
>
> Can we take your branch and force-push it?

The repositories are not compatible, mine is mirroring the whole src
repo, not just the gdb subset.

> If we could do that without confusing cvsps, that'd be awesome.

You can just reset the branch to the gdb-7.3 branchpoint, and
git-cvsimport will happily continue from that point.

(The gdb_7_3-2011-04-01-branchpoint tag is missing in the git repo,
since it was created in the past.  It should be added to
c73dde705c3b163b5101ed4cc4506cef6cc58905.)

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: gdb_7_3-branch not buildable?

[Jim Meyering]

Joel Brobecker wrote:
>> > Andreas, could you say what you did to make yours better?
>>
>> Handcraft.  These are things that something based on cvsps cannot do
>> better.
>
> Can we take your branch and force-push it? If we could do that
> without confusing cvsps, that'd be awesome. git really has become
> a significant time saver, and not having a correct mirror of the
> branch would mean a lot of work on my end.

Hi Joel,

I've adjusted git's gdb_7_3-branch so that it is much closer to cvs now.
Diffs below.

For the record, I've done the following in a local, private clone.

  # Rebase to proper branch point:
  git tag gdb_7_3-2011-04-01-branchpoint c73dde705c3b163b5101ed4cc4506cef6cc58905
  git branch --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
  git rebase --onto gdb_7_3-2011-04-01-branchpoint master v7.3

  # Update the mirror repo used to sync from cvs:
  git push ~/mirror-git-to-cvs/repo/src +v7.3:gdb_7_3-branch
  git push ~/mirror-git-to-cvs/repo/src tag gdb_7_3-2011-04-01-branchpoint

In a few minutes the twice-hourly sync-from-cvs script
will run and push that result to the public git repository.

==============================================
With that, here is the result of comparing the latest checked-out/cloned
directories from cvs and git:

Only in gdb-cvs: contrib
diff -u -r -x.git -xCVS gdb/gdb/doc/stabs.texinfo gdb-cvs/gdb/doc/stabs.texinfo
--- gdb/gdb/doc/stabs.texinfo	2011-05-02 18:47:22.530555384 +0000
+++ gdb-cvs/gdb/doc/stabs.texinfo	2011-01-05 05:09:54.000000000 +0000
@@ -40,7 +40,7 @@
 @page
 @tex
 \def\$#1${{#1}}  % Kluge: collect RCS revision info without $...$
-\xdef\manvers{\$Revision$}  % For use in headers, footers too
+\xdef\manvers{\$Revision: 2.130 $}  % For use in headers, footers too
 {\parskip=0pt
 \hfill Cygnus Support\par
 \hfill \manvers\par
diff -u -r -x.git -xCVS gdb/gdb/gdb.1 gdb-cvs/gdb/gdb.1
--- gdb/gdb/gdb.1	2011-05-02 18:47:22.581555645 +0000
+++ gdb-cvs/gdb/gdb.1	2011-01-05 05:09:51.000000000 +0000
@@ -1,6 +1,6 @@
 .\" Copyright (C) 1991, 1999, 2010, 2011 Free Software Foundation, Inc.
 .\" See section COPYING for conditions for redistribution
-.\" $Id$
+.\" $Id: gdb.1,v 1.4 1999/01/05 00:50:50 jsm Exp $
 .TH gdb 1 "22may2002" "GNU Tools" "GNU Tools"
 .SH NAME
 gdb \- The GNU Debugger
Only in gdb-cvs/gdb/testsuite/gdb.ada: mi_catch_ex
Only in gdb: .gitignore

Re: gdb_7_3-branch not buildable?

[Andreas Schwab]

Jim Meyering <jim@meyering.net> writes:

>   # Rebase to proper branch point:
>   git tag gdb_7_3-2011-04-01-branchpoint c73dde705c3b163b5101ed4cc4506cef6cc58905
>   git branch --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
>   git rebase --onto gdb_7_3-2011-04-01-branchpoint master v7.3

You should have reset the branch to the branchpoint, not rebase.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: gdb_7_3-branch not buildable?

[Jim Meyering]

Andreas Schwab wrote:
> Jim Meyering <jim@meyering.net> writes:
>
>>   # Rebase to proper branch point:
>>   git tag gdb_7_3-2011-04-01-branchpoint
>> c73dde705c3b163b5101ed4cc4506cef6cc58905
>>   git branch --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
>>   git rebase --onto gdb_7_3-2011-04-01-branchpoint master v7.3
>
> You should have reset the branch to the branchpoint, not rebase.

Why do you say that?
Does the current set-up cause trouble, somehow?
If so, please give details.

That you're saying something now suggests that you solved
this a different way.  Instead of just giving minimalistic
hints that maybe things should be different, why don't you
tell us what you actually did?

Re: gdb_7_3-branch not buildable?

[Andreas Schwab]

Jim Meyering <jim@meyering.net> writes:

> Andreas Schwab wrote:
>> Jim Meyering <jim@meyering.net> writes:
>>
>>>   # Rebase to proper branch point:
>>>   git tag gdb_7_3-2011-04-01-branchpoint
>>> c73dde705c3b163b5101ed4cc4506cef6cc58905
>>>   git branch --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
>>>   git rebase --onto gdb_7_3-2011-04-01-branchpoint master v7.3
>>
>> You should have reset the branch to the branchpoint, not rebase.
>
> Why do you say that?

So that all the commits can be properly reimported, instead of
potentially remaining broken.

> tell us what you actually did?

Reset the branch and rerun cvsimport.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: gdb_7_3-branch not buildable?

[Jim Meyering]

Andreas Schwab wrote:
> Jim Meyering <jim@meyering.net> writes:
>> Andreas Schwab wrote:
>>> Jim Meyering <jim@meyering.net> writes:
>>>
>>>>   # Rebase to proper branch point:
>>>>   git tag gdb_7_3-2011-04-01-branchpoint
>>>> c73dde705c3b163b5101ed4cc4506cef6cc58905
>>>>   git branch --track v7.3 remotes/origin/gdb_7_3-branch && git co v7.3
>>>>   git rebase --onto gdb_7_3-2011-04-01-branchpoint master v7.3
>>>
>>> You should have reset the branch to the branchpoint, not rebase.
>>
>> Why do you say that?
>
> So that all the commits can be properly reimported, instead of
> potentially remaining broken.
>
>> tell us what you actually did?
>
> Reset the branch and rerun cvsimport.

As far as I can see, all of the commits were imported into the
git repository without error.  The only problem is that
the branch tag appeared in CVS a few days after it should have,
and the intervening commits all happened to belong on the branch.
At least those ~20 intervening commits just happened to represent
precisely the differences I was seeing between the git and cvs
versions of the 7.3 branch.

Bottom line, since the desired commits were already there,
and contiguous with the old base of the branch, rebasing
the 7.3 branch to the earlier commit on the trunk did
exactly what was required.  That's why the inter-(cvs-git)-diff
is now effectively empty for the 7.3 branch.