Bazaar-NG Traffic #3 Prerelease

[James Blackwell]

And yes, I'll change the names over to a first name basis. Just don't come
complaining to me when there's four James' two martins and a half-dozen
Robs hanging around. =) 



On Mon, Oct 17, 2005 at 02:30:43AM -0400, James Blackwell wrote:
> Ok guys. Here's the Bazaar Traffic #3 prerelease. This file covers the
> summary of events on the mailing list traffic from approximately 03
> October to 11 October. 
> 
> This time I managed to squash 307 posts down to a paltry 220 lines.
> 
> Without any further ado: 
> 
> 
> = Traffic Notes =
> This traffic covers the Bazaar-NG mailing list traffic from 03 October
> until approximately 11 October. Within this document you will find a
> variety of things: bug fixes, releases, timings and most importantly the
> tireless efforts of over a dozen people into a paltry 221 lines. Enjoy!
> 
> = Executable flagging patch =
> The executable flagging patch was taking this week. The conversation, which
> gently spread open with a short discussion about an optimization that was
> causing a bug in the implementation, blossomed into a rehash about whether
> to track just the executable bit or not. The end result was the same plan
> as originally; namely that only the three executable bits would be
> preserved. The thread closed off with Robert Collins announcing a merge of
> the final version.
> 
> In a later post Gustavo Niemeyer updated the executable flagging so that a
> conflict on merging permissions was impossible. Good work
> 
> = bzr init dirname=
> John Meinel pinged the list asking if anybody was planning on merging a
> fix for 'bzr init dirname'. The name or location of the merge he was
> referring to didn't seem to be available.
> 
> = Don't display logs from merges =
> The thread about whether to have "bzr log" dive into merged logs continued
> this week. Collins reported to Gustavo Niemeyer that his patch for logs no
> longer applied because of new code. After the fix bounced betwixt the two
> for a bit, a patch that everyone liked went in. 
> 
> = Hiding trailing newlines in commit messages =
> Last week Alexander Belchenko supplied a patch that hides the display of
> extraneous newlines in commit messages. Martin Pool seemed to be a bit
> ambivalent about the patch and requested for other opinions. Though the
> general consensus is positive, the list does not mention whether it was
> accepted or rejected.
> 
> = New svn2bzr tool (cont) =
> Last week Niemeyer announced the release of a new svn2bzr. This week, 
> Jelmer Vernooij reported that svn2bzr took quite a while to run. Niemeyer
> asked for Vernooij for more information and provided some pointers on how
> to speed things up a little bit. Niemeyer also suggested avoiding
> compressed svn dumps.
> 
> = Output for missing =
> A thread reminiscent of the log display thread showed up entitled "star or
> parallele merge"(sic) started up that this week. As with the other thread,
> the pivotal question is 'should information for merges from third parties
> be displayed or hidden'. The final decision was to hide merges from third
> parties.
> 
> 
> = Merge is not a solved problem =
> Under the same title as the previous thread, a new conversation started up
> about the unpredictability of merge. Though several people contributed
> constructively to this thread, the most memorable part was written by David
> Allouche: 
> 
>    The bottom line is that, in my understanding, merging is largely
>    an unsolved problem with nasty yet relatively common edge cases,
>    so good diagnostic facilities are required.
> 
> After a little back and forth, Aaron Bentley hit the nail on the head with
> the following comment:
>     
>     Three-way also has silent failures there.  We gotta use something.
>     What do you suggest?
> 
> = Coding Standards =
> The "Re: Nomenclature" thread finished up this week. The developers agreed
> upon coding guidelines for variable names and such. If you hack on Bzr,
> please consider this thread against any patches that you may submit.
> 
> = Uni-coding for newformat =
> A thread opened up about unicoding filenames for inventories. There's not
> much here for endusers, but if you like encoding pesky filenames, then this
> thread should be your cup of tea. You can find it under "Re: newformat
> format change"
> 
> In another thread that later became moot, named "[BUG] bogus name_version
> in bzr.newformat", John Meinel noted that the information for a file was
> being updated unnecessarily. 
> 
> = Using SQLite with Bazaar-NG =
> John Meinel hacked up Bazaar-NG this week to use SQLite and reported
> impressive numbers for conversions (from what?). Though he stopped short of
> advocating using SQLite, he did report that conversions went from 10
> minutes down to 30 seconds. Martin Pool, in replying to a post that yours
> truly didn't receive, suggested that Bazaar-NG could see notable
> improvements over the 5:1 history-to-code storage that Bazaar-NG had at the
> time of the post.
> 
> Though the thread petered out this particular time, don't be surprised if
> this doesn't come up again.
> 
> = Xtla to use Pymacs to support Bazaar-NG =
> Matthieu Moy reported to the list that he was was planning to provide
> Bazaar-NG support for Xtla (Xtla is an emacs module that provides support
> for tla and Bazaar 1.x). Prior to getting started, Moy asked the list for
> opinions on whether pymacs would be a good choice, which the list
> wholeheartedly agreed with. The thread finished up with Martin Pool
> reaffirming that he was very interested in supporting the use of Bazaar-NG
> in a library context.
> 
> = Bazaar-NG may depend upon twisted = 
> Robert Collins was ecstatic that the transports code had been merged into
> Bazaar-NG and happily offered to work upon the network support. He offered
> to port an old proof of concept to the new format. The catch was that the
> code depended upon twisted. Surprisingly, the only person that spoke up was
> John Meinel, who agreed.
> 
> = Cleaning up bzr ignore output =
> Michael Ellerman, who addressed the list as "Bazaaristas", opinioned that
> the output from bzr ignore was more confusing than necessary. Ellerman put
> his money where his mouth was at in a time honored way: By providing a
> patch to fix his itch. The thread ended up one person suggesting the help
> should be improved and another person suggesting that Ellerman's assessment
> was correct. The thread seemed to end without common agreement.
> 
> = Ghosts in Bazaar-NG just in time for Halloween =
> Martin Pool announced that the Bazaar-NG upgrade command now converts
> missing revisions to "ghost" revisions. The thread then quickly turned into
> a discussion about supporting the pre-newformat compressed revisions.
> 
> = Bzr in server mode =
> In a foreshadowing move, John Meinel announced a plugin for Bazaar-NG that
> keeps a copy of Bazaar-NG running all the time. One accesses this running
> copy of NG by running a much smaller program written in C that connects to
> the server plugin which performs commands on behalf of the smaller
> frontend. The timings, good on a newer machine, were likely to be quite
> useful for slower machines. The conversation then moved to suggestions for
> the plugin such as possibly using sockets.
> 
> = Weave tests =
> John Meinel apparently owns the Bazaar-NG world when it comes to
> performance testing. This time Meinel compared the time for weave. Though
> weave timing was susceptible to the size of the change, generally speaking
> the time to weave was O(n), where n is the number of lines in the weave.
> Not quite satisfied with himself, he then took the extra step of graphing
> his results at http://bzr.arbash-meinel.com/other/RevisionInsertTime.png
> 
> In a later thread, Meinel compared the Bazaar-NG weave format against CSSC.
> Though Bazaar-NG wins handily, Martin Pool reported that this "is probably
> fain praise; tridge reports that some of the [cssc] code is very inefficient"
> 
> = Prefixed Stores =
> Gustavo Niemeyer supplied a patch to break the three stores into
> subdirectories. By using "the lower 8 bits of an adler32 checksum on the
> fileid", he breaks the store up into smaller chunks, which Robert Collins
> later merged. This patch makes it theoretically possible for Bazaar-NG to
> support very busy, very long lived branches such as python on filesystems
> such as ext3.
> 
> = Compacting Weave files =
> Realizing the the length of weave files mattered, Meinel continued work on
> his bzr-sax branch. He was successful in cutting down the size of weaves
> down quite a bit by being more selective on storing attribute changes. He
> was able to shave down the bzr.newformat branch from 2.6 megabytes down to
> 1.4.
> 
> = Bug in commit =
> Michael Ellerman reported that bzr commit had a bug that involved trying to
> commit a file that was in a subdirectory that was added but not committed.
> Robert Collins came to the rescue with a fix.
> 
> = Tabs in commit messages=
> Alexander Belchenko accidentally used a tab in a commit message and found a
> bug in Bazaar-NG. Harald Meland worked up a fix for it but had difficulties
> getting it submitted via Robert Collins because of a format incompatibility. 
> At the end of the thread Harald provided a traceback for assistance, which
> didn't make it in time for this traffic.
> 
> = Bug in adding then reverting a file =
> In a problem that wasn't resolved by the end of bzr traffic 3, Mark
> Rosenstand indicated that Bazaar-NG was eating data for files that had been
> added but not yet committed if the revert command was run. Over several
> iterations, a fix was suggested, only to be replaced by a more accurate
> fix. 
> 
> = Bzr 0.1rc1 available =
> (note: as of this writing, 0.1.1 or later is already out)
> 
> In an amusing turn of events, the mailserver for the ISP that Robert
> Collins used decided that gnupg signatures were a virus. 
> 
> Martin Pool was kind enough to post on his behalf the url for the download
> and a quick description of the release (new faster format and a couple
> minor bugs that needed to be fixed before 0.1)
> 
> = Ubuntu Below Zero =
> Martin Pool announced Ubuntu Below Zero. This is a hackers meet for people
> that like to work. More details are available at
> http://wiki.ubuntu.com/UbuntuBelowZero
> 
> = Aaron Bentley's branches moved =
> Aaron Bentley moved his branches into a subdirectory. They can now be found
> under http://panoramicfeedback.com/opensource/bzr/
> 
> = Cleared up progress bar =
> Dan Loda provided a fix for the confusing output after attempting a merge.
> Aaron Bentley already had a fix, which Robert Collins took.
> 
> = Goodbye, Gustavo =
> Gustavo sadly announced that he would no longer be working on Bazaar-NG
> full time. His frequent contributions will certainly missed. 
> 
> As a goodbye present, Niemeyer provided a patch for symbolic tags for
> Bazaar-NG, which Robert Collins merged in.
> 
> = Bzr log more useful =
> David Allouche provided a patch which shows the arguments and the current
> working directory of Bazaar-NG. Martin Pool thanked him and merged it.
> 
> In a sign of more good things to come, he then went to submit a merge
> request for test case fixes. Robert Collins asked for for a few changes,
> which David has just completed.
> 
> = Bzrweb is back =
> Chris McCormick announced that bzrweb is back online! At the time of this
> traffic bzrweb worked with 0.0.9, but not 0.1rc1.
>