From: Florent R. <fl...@vi...> - 2010-03-16 11:15:01
|
[ Please don't Cc me, I am already subscribed to the list, thank you. ] Hello, Peter Åstrand <as...@ce...> wrote: > This is all nice and so, but I think we are confusing our users by having at > least three different "forks" of the project: > > * Your version > > * The Sourceforge one > > * The fork by Chris Dew. Oops. I do read the mailing-list, but had forgotten the mail where Chris Dew announced his reluctant forking of pythondialog to github (<e6a...@ma...> from Sept. 16, 2008). My fault for that. 2008 was a bad period, and since there were no follow-ups to Chris' announcement, it went out of my memory. In my enthousiasm updating pythondialog again, I see I also implemented --form myself, which he had already implemented at that time. Point taken. > I think this is partly my fault. I've been a poor maintainer for the last > years, and I can now see that I've not even responded to the request from > Chris Dew to step in as a maintainer. Sorry for that - I haven't kept track of > the mailing list very well. Please always use my direct email for such > important issues. According to Chris' mail, this is indeed the reason why he did the fork to github. > Btw, does anyone have the email adress to Chris Dew? I've failed to find it. I put in Cc the e-mail address he used on Sept. 16, 2008. >> ----> Peter, would you be so kind to give me ownership on PyPI? Thanks >> in advance. Similarly with SourceForge, BTW... > > I don't have access to PyPI. The Package Index Owner is "paulproteus". You'll > need to contact him. But first of all, I think we should think a little bit > about merging our 3 forks... OK. I wrote that because http://pypi.python.org/pypi?%3Aaction=search&term=pythondialog&submit=search listed you as the Author, but the Owner must be "paulproteus" indeed (not contacting him before we get out of the current mess, though). >> The new version is available here: >> >> http://people.via.ecp.fr/~flo/projects/pythondialog/dist/ > > You don't use any kind of version control system? This is really bady, IMO. I > tend to avoid software that lacks proper version control. Ahem. I did try to commit version 2.09 to the svn repos I set up in 2004, but the operation hung. Maybe some dump/reload cycle of the DB would be needed. However: About the "really bady" part: you are free to avoid using my software if you are not confident about its quality. Although concerning pythondialog, I am afraid that if you actually rip out every line I wrote, you won't find much remaining (I would expect: a few lines of the demo from Robb Shecter, and the Xdialog patches). I am a careful and rational person, and for such simple things as pythondialog, version control brings basically nothing to me. I don't make modifications "just to see if it is going to fix $bug", with the intent of rolling back if not. I keep previous versions on my hard drive, and I make backups on offline media. This is basic version control. When I do an involved modification, I make a backup copy beforehand, etc. The software I write, be it public or not, usually has relatively few bugs once I declare it usable. I don't think the pythondialog case makes an exception here. You are free to disagree, but please provide some evidence then. About version control: of course, it is pretty much vital when several people work on the same project. It might be the case here, but ignoring Chris' activity, this has not been the view I had when publishing 2.09. [ However, when only a careful person works on a relatively simple program, version control is useless bloat, frankly. Especially if, a couple of years after being declared as the state-of-the-art of version control software, long-term replacement of the venerable CVS, these "things" are publically qualified as completely obsoleted by the new VCSes, people still using them being said to be "nuts", as seen very recently on lwn.net, in a long discussion mainly about Hg and git. ] > I can see that Chris Dew uses git. This is not my cup of tee either - > I prefer Subversion - but it's way batter than nothing. Personally, I don't care much. If I work alone, I don't need any of these for something simple enough. If not, I am OK with either svn, Hg or git (I don't know Hg nor git, but am ready to learn either of these). > I think that the best solution for pythondialog would be that someone like > Chris Dew took over the Sourceforge project, migrated to Subversion, and then > incorporated the various patches floating around, for example your 2.09 > changes. The version on http://github.com/chrisdew/pythondialog, as well as that on Sourceforge, which has been uploaded to Debian, have seen the ChangeLog file **disappear**. When such an important file manages to slip out of the releases for more than 5 years, please allow me to express my scepticism concerning the actual benefits of "good version control practices" in order to ensure top-quality in small projects. Modern tools don't replace love, care and talent. This has been shown many times in arts, craftman's trade, engineering, building... More prosaically, regarding commit http://github.com/chrisdew/pythondialog/commit/58be2752c85f63da91f6f2cea5190caab23fece7 in particular, I'd like to see a test case where the bug is reproducible with pythondialog 2.09, because: # Close the pipe that we are using to feed dialog's stdin try: p["stdin"].close() + # Doesn't always die on pipe closure + try: + os.kill(p["pid"], signal.SIGTERM) + except: + pass looks to me more like a workaound than a fix of the root cause. In fact, it may well be that this change in version 2.09: * dialog.py(__call_program): new close_fds option causing the child process to close the specified file descriptors before the execve(2) system call. This is useful for instance to have the child close an end of a pipe he isn't going to use. Without that, deadlocks could happen because of the child never seeing EOF from the pipe. as documented in my ChangeLog file, does fix the root cause of the problem. Version control with workarounds, or no version control but fixing of root causes, which one is more "really bady", according to you? > We are still using Xdialog. Others might as well. I think that just removing > Xdialog support is totally wrong. If the goal is to remove it, we should start > by deprecating it in 1 or 2 releases. We can't just suddenly remove it. Okay, okay, as said, if people are still using it and the patches don't impact too much the code, keeping them is fine with me. >> [1] Yeah, I know very well it is the same as 2.9 (in the most >> common/sane versioning scheme), but I started this way years ago, >> and 2.09 / 2.10 / etc. sort much better than 2.9 in lexicographic >> order, which is always pleasant in a directory listing such as: > > I *hate* version numbers like 2.09. It is your entire right, however I am not sure how rational it is. I am definitely *not* confusing 2.1 and 2.10 (in version string context). As said, 2.09 sorts better in directory listings (until, of course, 2.99) and shouldn't confuse anyone, since 09 == 9 in numeric context. I didn't intend to prolonge this scheme ad vitam æternam, but the sorting argument plus remaining consistent with previous numbers made me think that waiting for 2.10 before switching transparently to the more conventional versioning scheme was a perfectly sane course of action. -- Florent |