Re: master cbef1e9 2/3: ; make change-history-commit

[Glenn Morris]

Paul Eggert wrote:

>> I had assumed that the ChangeLog would not be versioned at all now,
>> but it seems like it is?
>
> No, your assumption is correct.  For example, there is no ChangeLog here:
>
> http://git.savannah.gnu.org/cgit/emacs.git/tree/ChangeLog

I meant: I had assumed that the generated ChangeLog _information_ would
not be versioned now (did you really think I was asking about a file
literally named ChangeLog, and/or am unable to tell whether a file is in
the VCS or not?). One usually does not store generated files in the VCS,
as you know.

>> I assumed that only the corrections would be versioned, and the full
>> file would only appear in tarballs as a generated file.
>
> Stefan indicated long ago that this was too drastic , and that we
> should keep ChangeLog history files in the repository, e.g.,
> lisp/ChangeLog.10 is still in the repository and hasn't changed since
> January 1's copyright-date edits. ChangeLog.1 is another file like
> lisp/ChangeLog.10; i.e., it's another versioned ChangeLog history file
> that you can edit by hand whenever you like.

You seem to be conflating the old, written-by-hand ChangeLogs and the
new, generated ones. Obviously the old ones should not be removed from
VCS, that would be crazy and is not what I meant.

>> I further assumed that the new-style generated log would always be
>> separate from the old-style ChangeLog.1
>
> Yes, that's correct.  The ChangeLog file in the tarball (built by
> make ChangeLog') is disjoint from ChangeLog.1, in that ChangeLog.1
> contains only "older" entries and Changelog contains only "newer"
> entries.

The file ChangeLog.1 in the repo contains a mixture of new, generated
entries applying to the entire Emacs tree, and (before a certain magic
date) older, hand-written entries that only apply to the top-level
(essentially) directory. I think it's pretty confusing to have both
things in the same file.

> Yes, the idea is that you can move the boundary line between "older"
> and "newer" by running "make change-history".  This automated
> procedure moves the boundary to be the present, and hence entries that
> until now would have been put into ChangeLog, are put into ChangeLog.1
> and become part of the manually-edited history, which means that
> ChangeLog will become empty.  This is how you can fix old ChangeLog
> entries: run "make change-history", then edit ChangeLog.1 by hand and
> commit the result.

That's not great, is it?
Suppose we improve/change the ChangeLog generation script, and want to
regenerate the generated ChangeLog. It will be difficult to extract the
various corrections that have been applied by hand to the generated
entries.

I had assumed that "ChangeLog" would be a top-level generated file that
would grow and grow with time, and would always be generated from
scratch whenever needed, with entries always starting from Apr 7 2015.

> effect, we have a hybrid approach, which uses Coreutils style for
> "recent" changelog entries and traditional Emacs style for "older"
> changelog entries.

Obviously, since we didn't want to throw away our old, hand-written
entries. This was a basic requirement and not some kind of selling
point.


But I'm not going to do any work on this myself, so obviously I can't
expect anyone else to.


PS On a semi-related note, you renamed man/ChangeLog to ChangeLog.01
using the justification that erc/ already uses that system. But erc uses
a non-standard scheme where .YY contains the entries for 20YY. So
personally I think erc/ChangeLog.09 is incorrect, since it does not
contain entries for (just) 2009. IMO it would have been better to
combine all the old erc logs into erc/ChangeLog.1, and not to rename
man/ChangeLog.