[OSM-dev] osm2pgsql and homebrew

[Matt Patterson]

Sorry about the delay following up on this: I got sick and massively behind…

Kai Kruger wrote:
> Matt Patterson wrote
>>  I know Kai Kruger packages osm2pgsql for Ubuntu and therefore imposes
>>  'stable' releases, at least as far as APT is concerned.
> 
> Well, they are not really stable releases and the situation isn't directly
> ideal there either. I kind of just periodically take a SVN snapshot and
> package it to the ubuntu ppa repository. It isn't directly a "stable release
> process". Although I mostly only update the packages for the most recent
> developemnt version of Ubuntu. For older / stable versions of Ubuntu, I
> mostly only update the packages if there are known issues or a specific
> reason that an update is necessary (e.g. to transition to 64 bit builds as
> the 32bit unsigned node id space runs out). So those are semi stable. Mostly
> also because there are no "stable releases" and so I don't want to risk
> breaking existing installations.

Ah, I understand now.

> Given the developer situation of osm2pgsql / mod_tile, I suspect we won't
> see a proper release process with forked off stable versions any time soon.
> However, perhaps we can move over to a process a la JOSM? Which seems to
> just "randomly" tag SVN snapshots as stable if there have been no reported
> issues with that snapshot for a while.
> 
> I.e. someone would announce to the dev mailinglist that a certain SVN
> revision should be declared "stable" and then if there are no complaints
> within e.g. a week it gets tagged as such in SVN?

That sounds workable to me. Maybe it would be worth declaring a 'release' once a quarter - judging by the activity in the github mirror of SVN, that'd mean we'd nearly always have an up-to-date release. We could even pick a date, declare a particular revision 'RC' and push that as a release a month later if there weren't any problems reported. 

> Do any of the commercial users of the osm rendering stack have some ideas of
> how to improve the quality control and "release cycles"?  E.g. to try and
> improve the automated testing framework?
> 
> 
> Matt Patterson wrote
>> I've looked over some of the error reports people were sending in for
>> osm2pgsql as part of homebrew and I'm pretty sure that the fix for those
>> is simple - namely, that osm2pgsql either doesn't build with Clang, or its
>> configure script doesn't think it does, and so the packaging script
>> ('formula', if you're unfamiliar with homebrew) needs to declare a
>> dependency on GCC-proper.
> 
> I haven't tried building osm2pgsql on MacOSX, as I am not really familiar
> with the whole development toolchain there, but at least in linux it does
> seem to build fine with clang.

Hmmm. I probably need to go and dig deeper into these build crash reports before I open my mouth then ;-)

> I ran the configure script with "./configure CC=clang CXX=clang++" and both
> configure and make completed without issues.
> 
> If you can help me with more error reports or with guides of how to set
> things up and test on mac OSX, I'd be happy to try and fix any build issues. 

I'll look into it.

> Matt Patterson wrote
>> I will happily take on ensuring that osm2pgsql gets back into Homebrew if
>> there's a sensible way of declaring 'stable' releases - i.e. piggy backing
>> on Kai's work, a bi-weekly tarball uploaded to a known location with a
>> sensible incrementing naming format, or even just a regularly updated
>> 'release' tag in SVN or Git.
> 
> What do other osm2pgsql developers think of this? Can we find a more
> sensible and reliable solution to this issue than what we have now (i.e. no
> policy at all)?

If nothing else, for homebrew I could run a mirror of the official github SVN mirror and tag that with a quarterly cron job…

Matt

-- 
 Matt Patterson | Design & Code
 <matt at reprocessed.org> | http://www.reprocessed.org/