[racket-dev] Migrating the bug database to GitHub

From: Eli Barzilay (eli at barzilay.org)
Date: Mon Mar 19 01:41:55 EDT 2012

Three hours ago, Sam Tobin-Hochstadt wrote:
> On Sat, Mar 17, 2012 at 10:55 AM, Eli Barzilay <eli at barzilay.org> wrote:
> > The real issue is whether it's really alright with you...
> > Currently, something that I do and I'm sure others do it to, is
> > keep the bug in my mailbox with any followup discussions.  In some
> > cases the followups contain enough information so I'll keep only
> > that and not the original.  With the github thing, if you get to
> > deal with a bug several days after it was posted, it will be a
> > good idea to check the bug since there could have been
> > clarifications that you're unaware of.
> This is regularly the case with bugs in Gnats currently, although
> probably less often than on GitHub.  If you don't cc
> bug-notification, then only the assignee and the reporter see the
> email, which is *fewer* people than on GitHub.

Right -- and "less often" is the key point here, since the convention
is to use reply-all.

> This is also more significant for contributors not among the people
> who are on the bug-notification list.  Right now, they won't be
> notified at all about changes to gnats bugs unless they're the
> submitter.  With GitHub, all commenters are treated the same.

That part is very easy to fix: change `bug-notification' to a public
mailing list that anyone can subscribe to.  (But that's obviously too
much since you'd get everything from it, obviously making the github
workflow better for such people.)

> > And since attachments were mentioned: a possible situation is that
> > someone posts a bug, the bug czar asks for some clarification, and
> > the email reply has a screenshot which GH ignores.  In that case
> > you will need to get that attachment directly from one of the
> > people.  The outcome of this is that it's better to leave stepper
> > bugs for you to deal with instead, and the "bug czar" role is
> > minimized to just assigning bugs or maybe not even that and it
> > gets eliminated.
> This is not how I envision the process working.  Instead, I hope
> that people use services like `gist.github.com' and `imgur.com' to
> post large documents and images, and embed links in bug emails.
> This makes the online record much more useful -- gnats storage of
> email attachments is very difficult to work with.  I hope to provide
> command line tools and/or DrRacket tools to make this easy.

Yeah, I considered some of these: gist and imgur have easy APIs,
alternatively, we can setup a simple server for uploading random files
(also easy to avoid spam: make sure that contents have links from bug

The main flaw in all of this, which is how the above scenario plays
out, is a submitter that just replies to an email with an attachment,
since going through such tools is inherently an added hassle.

This could be resolved by making the notification emails reply-able,
and have a script that identifies attachments and saves them on
whatever.  Parsing emails is a bunch of work though.

> As to the "bug czar" role (which I currently have), I think of my
> role as to facilitate people working on software, including fixing
> bugs.  I've planned this move to GitHub because I think it will help
> both me and other people developing Racket with doing this.  As part
> of that, I still plan to triage every bug to someone, and to be
> responsible for contacting bug reporters.  I don't think this is a
> role that will go away -- bugs need human supervision, and I'm
> planning to continue providing it.

(If you plan to interact with submitters, then you should really be
clear on how to avoid such problems.)

          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!

Posted on the dev mailing list.