bug#22506: make change-history now doesn't work at all

bug#22506: make change-history now doesn't work at all

[Glenn Morris]

Package: emacs
Version: 25.0.90

I don't know how y'all plan to address the issue of running make
change-history on the release branch (bug#21998), but now it doesn't
work on either emacs-25 or master. On both branches,
change-history-nocommit fails with "fatal: Invalid revision range". Eg:

  GEN      ChangeLog
fatal: Invalid revision range
3f43da09956b663d92f578e41bcacb8742bb6bba..0156b79ea8fd473c550abdb8d3e3b4eacaaab289
gitlog-to-changelog: error closing pipe from git log --log-size
'--pretty=format:%H:%ct  %an  <%ae>%n%n%B'
3f43da09956b663d92f578e41bcacb8742bb6bba..0156b79ea8fd473c550abdb8d3e3b4eacaaab289
make[1]: *** [ChangeLog] Error 128

bug#22506: make change-history now doesn't work at all

[Paul Eggert]

[-- Attachment #1: Type: text/plain, Size: 1994 bytes --]

Glenn Morris wrote:
> Package: emacs
> Version: 25.0.90
>
> I don't know how y'all plan to address the issue of running make
> change-history on the release branch (bug#21998), but now it doesn't
> work on either emacs-25 or master. On both branches,
> change-history-nocommit fails with "fatal: Invalid revision range".

I reproduced the problem and installed the attached patches to the emacs-25 and 
master branches to try to fix it.

In the master branch this problem was introduced by my commit 
ffbf163ab5ced1bc464a0034e6abc9a41f5e09c4 dated Sat Jan 30 14:45:53 2016 -0800. I 
did this commit by hand, not by using 'make change-history', and evidently I 
messed it up.

In the emacs-25 branch this problem was introduced by Nicolas Petton's commit 
e6b7b6d89ff9288a49099f041752908b5eb9613e dated Sat Jan 30 14:39:23 2016 +0100, 
which updated ChangeLog.2's end commit from 
8a8613bcf4227dfe46a694b761e9575bdf6ca2ce (by me dated Sat Nov 7 23:52:17 2015 
-0800) to d5195155f9a297dc45a40c9b7175715ffe4f1612 (a commit ID that does not 
exist in the Emacs repository). My guess is that Nicolas ran 'make 
change-history' on a private branch and then merged the result into emacs-25. 
This has never worked, and it's why Makefile.in tries to check for this sort of 
thing. I vaguely recall telling Nicolas to ignore the checks and do the right 
thing, but it's not at all clear what the right thing is and there should be 
little wonder that it went awry.

I guess from now on we should take those checks more seriously.

We also need to improve the procedure for keeping track of history, as the 
current procedure does not work well at all. We're better off than we were when 
ChangeLog files were maintained by hand in parallel with commit messages, but 
the by-hand maintenance of ChangeLog.2 is still too much trouble. I've proposed 
a way to improve this but to be honest I'd rather just dump the by-hand 
maintenance entirely and this has meant I'm not motivated to work on the problem.

[-- Attachment #2: 0001-Fix-ChangeLog.2-commit-ID.patch --]
[-- Type: text/x-diff, Size: 703 bytes --]

From f6c1d42be0798c6dae9a5ab65cc1e0506a444e37 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sun, 31 Jan 2016 12:18:03 -0800
Subject: [PATCH] ; Fix ChangeLog.2 commit ID.

---
 ChangeLog.2 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ChangeLog.2 b/ChangeLog.2
index 27c6ce3..5921546 100644
--- a/ChangeLog.2
+++ b/ChangeLog.2
@@ -26338,7 +26338,7 @@
 
 This file records repository revisions from
 commit 9d56a21e6a696ad19ac65c4b405aeca44785884a (exclusive) to
-commit d5195155f9a297dc45a40c9b7175715ffe4f1612 (inclusive).
+commit e6b7b6d89ff9288a49099f041752908b5eb9613e (inclusive).
 See ChangeLog.1 for earlier changes.
 
 ;; Local Variables:
-- 
2.5.0


[-- Attachment #3: 0001-Fix-ChangeLog.2-commit-ID.patch --]
[-- Type: text/x-diff, Size: 703 bytes --]

From d8d5f401aeb371019ae6b1b8089250d9d7fb3240 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sun, 31 Jan 2016 12:37:11 -0800
Subject: [PATCH] ; Fix ChangeLog.2 commit ID.

---
 ChangeLog.2 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ChangeLog.2 b/ChangeLog.2
index c5cba03..12f9834 100644
--- a/ChangeLog.2
+++ b/ChangeLog.2
@@ -28545,7 +28545,7 @@
 
 This file records repository revisions from
 commit 9d56a21e6a696ad19ac65c4b405aeca44785884a (exclusive) to
-commit 3f43da09956b663d92f578e41bcacb8742bb6bba (inclusive).
+commit ffbf163ab5ced1bc464a0034e6abc9a41f5e09c4 (inclusive).
 See ChangeLog.1 for earlier changes.
 
 ;; Local Variables:
-- 
2.5.0

bug#22506: make change-history now doesn't work at all

[Nicolas Petton]

[-- Attachment #1: Type: text/plain, Size: 416 bytes --]

Paul Eggert <eggert@cs.ucla.edu> writes:
> My guess is that Nicolas ran 'make change-history' on a private branch
> and then merged the result into emacs-25.  This has never worked, and
> it's why Makefile.in tries to check for this sort of thing.

No, I changed the branch check on emacs-25 to check for emacs-25 instead
of master, and then ran make change-history-nocommit.

This worked without any problem.

Nico

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 512 bytes --]

bug#22506: make change-history now doesn't work at all

[Paul Eggert]

Nicolas Petton wrote:
> This worked without any problem.

It worked for you, and it probably would have worked again for you in your 
environment. But it didn't work for the rest of us because we're using the 
public Emacs repository and its commit IDs are different from those in your 
repository. (Welcome to the wonderful world of distributed Git development....)

bug#22506: make change-history now doesn't work at all

[Eli Zaretskii]

> From: Paul Eggert <eggert@cs.ucla.edu>
> Date: Sun, 31 Jan 2016 13:50:20 -0800
> 
> Nicolas Petton wrote:
> > This worked without any problem.
> 
> It worked for you, and it probably would have worked again for you in your 
> environment. But it didn't work for the rest of us because we're using the 
> public Emacs repository and its commit IDs are different from those in your 
> repository.

??? How can the commits differ, assuming Nicolas pushed his to
upstream?

bug#22506: make change-history now doesn't work at all

[Paul Eggert]

Eli Zaretskii wrote:
> ??? How can the commits differ, assuming Nicolas pushed his to
> upstream?

I don't know. Maybe he rebased his changes before pushing. Or maybe he did a 
merge before pushing. I suppose it could be other things.

bug#22506: make change-history now doesn't work at all

[Paul Eggert]

I assume this particular bug (the one that blocks a release) has been fixed, so 
I'll close the bug report. If I haven't actually fixed it please feel free to 
reopen it. We do have the more-general problem that ChangeLog maintenance is a 
pain, but that should be a different bug as it does not block a release.

bug#22506: make change-history now doesn't work at all

[Nicolas Petton]

[-- Attachment #1: Type: text/plain, Size: 466 bytes --]

Eli Zaretskii <eliz@gnu.org> writes:

>> It worked for you, and it probably would have worked again for you in your 
>> environment. But it didn't work for the rest of us because we're using the 
>> public Emacs repository and its commit IDs are different from those in your 
>> repository.
>
> ??? How can the commits differ, assuming Nicolas pushed his to
> upstream?

Indeed, I'm really confused, I pushed my commits to emacs-25 after
that...

Nico

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 512 bytes --]

bug#22506: make change-history now doesn't work at all

[Nicolas Petton]

[-- Attachment #1: Type: text/plain, Size: 416 bytes --]

Paul Eggert <eggert@cs.ucla.edu> writes:

> Eli Zaretskii wrote:
>> ??? How can the commits differ, assuming Nicolas pushed his to
>> upstream?
>
> I don't know. Maybe he rebased his changes before pushing. Or maybe he did a 
> merge before pushing. I suppose it could be other things.

You are right, I did rebase one commit that happened in between in
emacs-25. That's most probably the reason!

Nico

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 512 bytes --]

bug#22506: make change-history now doesn't work at all

[Nicolas Petton]

[-- Attachment #1: Type: text/plain, Size: 524 bytes --]

Paul Eggert <eggert@cs.ucla.edu> writes:

> I assume this particular bug (the one that blocks a release) has been fixed, so 
> I'll close the bug report. If I haven't actually fixed it please feel free to 
> reopen it. We do have the more-general problem that ChangeLog maintenance is a 
> pain, but that should be a different bug as it does not block a
> release.

How should I generate the changelog files next time?  IIUC I need to
make sure no commit happens while doing so, so I won't have to rebase.

Nico

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 512 bytes --]

bug#22506: make change-history now doesn't work at all

[Paul Eggert]

On 02/01/2016 12:43 AM, Nicolas Petton wrote:
> How should I generate the changelog files next time?  IIUC I need to
> make sure no commit happens while doing so, so I won't have to rebase.
>

You might try something like this:

git checkout emacs-25
git pull
git diff # This should output nothing.
make change-history
git diff HEAD^ # This should output what you want.
git push

bug#22506: make change-history now doesn't work at all

[Nicolas Petton]

[-- Attachment #1: Type: text/plain, Size: 493 bytes --]

Paul Eggert <eggert@cs.ucla.edu> writes:

> On 02/01/2016 12:43 AM, Nicolas Petton wrote:
>> How should I generate the changelog files next time?  IIUC I need to
>> make sure no commit happens while doing so, so I won't have to rebase.
>>
>
> You might try something like this:
>
> git checkout emacs-25
> git pull
> git diff # This should output nothing.
> make change-history
> git diff HEAD^ # This should output what you want.
> git push

I'll do that, and make sure I don't rebase.

Nico

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 512 bytes --]

bug#22506: make change-history now doesn't work at all

[Glenn Morris]

Paul Eggert wrote:

> We also need to improve the procedure for keeping track of history, as
> the current procedure does not work well at all. We're better off than
> we were when ChangeLog files were maintained by hand in parallel with
> commit messages

It's going pretty much as I expected. The overall quality of logs has
gone down, and correcting them has largely been lumped on the plate of
the person making the release. The fact that merging ChangeLog from
release branch to master has actually become harder than it was before
is an unexpected bonus though. ;)

> I've proposed a way to improve this but to be honest I'd rather just
> dump the by-hand maintenance entirely

I've also felt like suggesting dropping the ChangeLog completely,
because the project as a whole doesn't seem interested in such things.
But at a minimum, there needs to be a way to correct legally relevant
information: wrong author, missing tiny change.

I would once again suggest git notes (need to correct a commit log? add
a note with the fixed version) rather than keeping a generated file in
the repo, but I'm also unmotivated to work on it, so it's just hot air.