launchpad-dev team mailing list archive

[Jonathan Lange <jml@xxxxxxxxxxxxx>]

On Tue, Jan 5, 2010 at 3:54 AM, Gary Poster <gary.poster@xxxxxxxxxxxxx> wrote:
> Thank you for the survey, Jono.
>

My pleasure.

...
> """
>  * Making a new branch is new excrutiatingly slow. It took me 10 minutes to do
>   one yesterday as I was short of real RAM and rf-branch seems RAM hungry.
>   It's been like this ever since buildout and it drives me *bonkers*.
> """
>
> Huh.  My perceptions have been that making a branch is now much faster than when I started, which I have attributed to bzr improvements and shared eggs.  People who feel the way this quote describes should maybe talk to me.
>
> My first guess is that you are working in some way that does not let you share eggs across branches.  *If* this is the case, then my observation would be that rocketfuel-* is the blessed and supported way for building branches, and if you are doing something else, I certainly understand and sometimes do the same, but we need to either acknowledge that you are on an "expert" path and are responsible for your own safety, or that we need to have explicit support for a non rocketfuel-* approach.  Either way, IMO the idea should be that there is a path (or two) that is supported, documented, and improved; if you stray off the path for your own reasons, that's your business: the maintenance of your path is your own responsibility (until we all decide to add it to the "supported" list).
>
> On the other hand, maybe people with this experience are in fact sharing eggs.  In that case, we should see what the problem is.  I believe and hope that this experience is unusual.
>

Personally, I use:
  bzr cbranch stable foo
  cd foo
  ./utilities/link-external-sourcecode ~/src/launchpad/devel
  make build

That is, I use sanctioned scripts and routines, and I share eggs, but
I don't use the rocketfuel-* scripts. My reasons are:
  - as someone working on Launchpad / Bazaar integration, I wanted to
be using Bazaar as much as possible
  - I use lightweight checkouts, and I don't know that
rocketfuel-branch supports that
  - If ever someone makes a Bazaar GUI or Emacs mode that I prefer to
the command line, I want to use that.

> """
>  * It seems silly to constantly consult the zcml for how urls, view classes,
>   and templates are mapped. It would be nice if view names and template names
>   were mapped by convention instead of by configuration, but that wouldn't
>   help with urls. Maybe we just need a script to discover the mappings more
>   quickly.
> """
>
> FWIW, this is the kind of thing that Grok technology can provide.  The discussion of the pros and cons of this particular idea within our own code base has not yet been run to ground, to my knowledge.  I am open to this perspective.

Where's the discussion taking place?

> """
>  * ./bin/lint has a lot of false positives mainly because it can't resolve a
>   bunch of lazr modules.
> """
>
> I agree that this is annoying.  I thought I knew how to fix it but someone else had tried what I had in mind and it didn't work.  I don't know what to do about this ATM, other than more digging.

I don't see a bug about this on
https://bugs.edge.launchpad.net/launchpad-foundations/+bugs?field.tag=build-infrastructure

Personally, I think we should ditch pylint, or at least restrict it to
checking formatting issues only. Pyflakes is much faster, catches
almost everything relevant, and produces far fewer false positives.

jml