Hi Alex, At 2022-11-02T13:39:58+0100, Alex Colomar wrote: > Heh, I've been reading the patch, and it very much sounds like Chinese > to me. Good that it was easy for you :)
Not _too_ easy; the second part of the patch was spurious. I managed to confuse myself. (It's _static_ register and string interpolations, i.e., those performed at macro definition time, that are incompatible with `ec`/`eo` bracketing, not dynamic ones. The latter is the common case.) > 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. > 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 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". 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. 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 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. 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 Regards, Branden
signature.asc
Description: PGP signature
