kicad-developers team mailing list archive

[Simon Richter <Simon.Richter@xxxxxxxxxx>]

Hi Tom,

On 21.01.2016 20:00, Tomasz Wlostowski wrote:

> Currently Kicad supports only Intel-based platforms and the patch is
> intended to make them build reliably.

Well, in Debian we ship a few more[1].

> What happens if Boost devs
> decide to change the ABI of the context library once again, like they
> did between 1.55 and 1.56? Should we #ifdef all our code to support
> every Boost configuration on the planet?

We send them a strongly worded letter.

So far, we've added #ifdef'd blocks whenever supporting an older version
of a library required it and was deemed worthwhile. Cleaning these up as
we raise the minimum required version of dependent libraries is usually
straightforward.

> We can't be responsible for every single build on every single
> architecture and OS.

Right, but I fear that patch is bringing us in the direction of more,
not less responsibility.

The patch is useful for users compiling from source because it absolves
them from looking for a non-broken version of Boost, but with my Debian
hat on, I must say I don't like it. It is fairly easy to step on the
toes of the Boost maintainers to keep their package in shape, and it is
likewise easy to simply declare a build dependency on a known good
version, but if e.g. the ARM assembler code had a bug, then we'd find
ourselves at the beginning of a long email exchange where we try to find
someone willing and able to investigate it.

So I'd prefer to use this code only to replace known bad versions of
Boost, with a large warning during the build process that we are doing
so. That means even more #ifdefs, but it should reduce the maintenance
effort in the long run, because it is clearly marked when the code can
be removed again.

   Simon

[1] http://buildd.debian.org/kicad

Attachment: signature.asc
Description: OpenPGP digital signature