Hi Branden, On 11/3/22 03:18, G. Branden Robinson wrote:
I'll get it from git soon. I'm not in a hurry for it ;)A good thing too. Incidentally this is one reason I don't push every day even though I work on groff almost every day. After 24-72 hours I often realize that some change I have made is boneheaded, and I revise it or back it out.
Ahh, I miss Michael so much for that! Not having someone reviewing my patches allows me to commit such crimes. Last month you can find a commit that reverted _the parent_ commit. Just 15 minutes of difference. I won't even link to it here, since it's a bit embarrasing :P
I wonder why so much duplication. The ChangeLog seems like an exact _duplicate_ of the git history.It's a subset. I am continuing the practice that I was trained on when I joined groff development. It wasn't written down anywhere except emails, though. I've since added a HACKING file and documented it. https://git.savannah.gnu.org/cgit/groff.git/tree/HACKING I also suffix a commit message with asides, mentions of miscellaneous or auxiliary changes (usually to the style of text prose), or illustrative exhibits of formatter behavior before and/or after the change. This material I do _not_, as a rule, copy into the commit message.
I couldn't parse the above. I guess the last sentence wanted to say Changelog instead of commit message?
I chopped the Linux man-pages 'Changes' file considerably by having a high level overview of the changes, and deferring to git(1) for further details.There are some groff contributors who are vocally opposed to the practice. After some initial frustration with it, while I found a workflow that suited me (and discovering that Werner is a reliable source of good advice), I have come to see it as useful. Setting aside the stuff that is documented _only_ in commit messages (documentation tweaks, cosmetic code changes), a proper ChangeLog file has virtues. 1. It's available, or is required to made available on request, to all who receive the software. This is explicit in GNU GPLv2, ยง2. "a) You must cause the modified files to carry prominent notices stating that you changed the files and the date of any change." This provision dated to a time when change logs were maintained on a per-file basis, usually in a whopping big comment block at the top of the file. A GNU-style ChangeLog file honors the spirit of the foregoing provision. I seem to remember that during the GPLv3 drafting process, this requirement was uncontroversially revised to reflect common practice even in GNU projects and the supersession of file-based revision control systems (SCCS, RCS) by hierarchical ones ("repositories"). I can't find the updated language in GPLv3 at a quick glance, but I don't regard that as determinative; the GNU GPL was written to codify and promote good _social_ practices of software sharing. We've had 25 years to observe the results when people hew to them only grudgingly. Ideally, people who hate copyleft would stick to their BSD and MIT licenses while chasing their dreams of mixing a little secret, proprietary sauce into free code and thereby living out their days on a yacht sustained by the glorious extraction of monopoly rents. Of course that is a utopian perspective; the modus operandum of a rentier is to find new and vibrant social contracts to join and defect from in swift succession, maximizing gains. If a community doesn't survive such predation, those defectors who have read a little Schumpeter (perhaps at Nth hand) will characterize their actions as "creative destruction".
(1) is not really an issue, since saying "To learn more about changes applied to individual pages, use git(1)" already points to the changes list, so it's publicly available on request. Having it in the tarball itself is just a convenience to those who inspect tarballs, which these days is not as common as it once was, I like to think.
2. It's mutable; commit messages are not, but people make mistakes. I
make mistakes all the time. Typos and even occasional errors of
fact get into my commit messages, and moreso because, unlike some, I
am not committed to terseness above all other virtues. It is easy
to write a useless commit message, and easier still to tell oneself
that one is thereby upholding the elegant traditions of Unix and C.
This is fixed by reviewing changes. However, let's admit it: no-one reviews my patches to the Linux man-pages, unless I consider it important enough to ask, and even then I often get no responses so I just push...
I like to extend on commit messages, writing any details that may seem interesting, but often I make assumptions about what is "common knowledge" that may differ from what others consider common knowledge. And so I may write less clear than others would like.
However, by keeping a list of emails that one can reach in the commit log, if someone doesn't have those assumptions so clear can ask the commit authors and any other emails noted in the email.
If we restrict the legitimate purpose of a ChangeLog entry or commit
message only to the "what" of a change set, then you don't need it:
a diff suffices to tell you _what_ changes were made. Commit
messages, like code comments, should be concerned with "what"
insofar as doing so enables clear expression of the more important
matter of _why_ changes were made. I try to explain the changes I
make, both so I can defend myself in the arguments that occasionally
arise in software development, and more fundamentally so that I can
later _remember_ why I made them.
Human brains can be slick surfaces. I often find I have to modify
an old ChangeLog entry (usually one of my own) to clarify or correct
it, usually to explain what _really_ changed or what _actual_ facts
obtained at the time, instead of the ones I deludedly believed at
the time I first wrote it.
I admit having wanted to rebase some times to ammend a commit message. Not so often though, for a reason: normally, when the commit message is not so good, the diff itself was also not so good, so I find an excuse to make some little change in the code, and then I can add whatever commit message I think was deserved.
My main issue with changelogs is that they are not bound to any specific diff. For finding the related commit messages for a given code, it's as easy as running git-blame(1) and git-log(1) on a given file. For finding the relevant entry in the changelog, that's speleology (since it's written in the same commit, it's not so hard, as git-blame(1) still works, but it's not so ideal).
I will note that I generally don't _retrospectively_ revise others'
ChangeLog entries. (I may help compose them as part of an initial
commit.) If I find commentary is essential for honest communication
with the users, I tend to put it in brackets and sign it with my
initials.
I hope this sheds some light on things. I know you asked for it earlier
and I didn't have the wherewithal to write all this down at the time.
No problem! :)I still have considerable doubts. But yeah, I understand why you would want to edit changelogs, whhich you can't do with commit messages.
Other projects don't necessarily need to follow the same practices; the foregoing is my defense of existing practice in groff, which has the virtue of long continuity. I can say I'm very much looking forward to retiring the current ChangeLog file to ChangeLog.123. Its length could be considered embarrassing. ๐
Heh, in the man-pages we only have one Changes.old: $ wc Changes.old 55425 181405 1627707 Changes.oldIf this file is useful to anyone, I'd like to know how. What kind of public uses that file? I have never opened that file for anything useful.
From your HACKING file, I could be interested in the NEWS file, if I was a casual user/packager, or in the git log, but I don't think I would be interested in reading the ChangeLog file.
Thanks! Alex
Regards, Branden
-- <http://www.alejandro-colomar.es/>
OpenPGP_signature
Description: OpenPGP digital signature
