Picking platforms

Whenever you write Open Source Software, you have to make a choice of what platforms you are going to support.

Generally you want to make your program as portable as possible and cover as many platforms, distros and weird computers as possible.

But making your program run on everything is hard work very hard work.

For instance, did you know that:

sizeof(void*) != sizeof(void * const)

is legal in a ISO-C compliant environment ?

Varnish runs on a Nokia N900 but I am not going to go out of my way to make sure that is always the case.

To make sense for Varnish, a platform has to be able to deliver, both in terms of performance, but also in terms of the APIs we use to get that performance.

In the FreeBSD project where I grew up, we ended up instituting platform-tiers, in an effort to document which platforms we cared about and which we did love quite as much.

If we did the same for Varnish, the result would look something like:

A - Platforms we care about

We care about these platforms because our users use them and because they deliver a lot of bang for the buck with Varnish.

These platforms are in our "tinderbox" tests, we use them ourselves and they pass all regression tests all the time. Platform specific bug reports gets acted on.

FreeBSD

Linux

Obviously you can forget about running Varnish on your WRT54G but if you have a real computer, you can expect Varnish to work "ok or better" on any distro that has a package available.

B - Platforms we try not to break

We try not to break these platforms, because they basically work, possibly with some footnotes or minor limitations, and they have an active userbase.

We may or may not test on these platforms on a regular basis, or we may rely on contributors to alert us to problems. Platform specific bug reports without patches will likely live a quiet life.

Mac OS/X

Solaris.

Yes, we'd like to bump Solaris to tier-A but I have to say that the uncertainty about the future for OpenSolaris, and lack of time to care and feed the somewhat altmodishe socket-API on Solaris, does keep the enthusiasm bounded.

NetBSD, AIX and HP-UX are conceivably candidates for this level, but so far I have not heard much, if any, user interest.

C - Platforms we tolerate

We tolerate any other platform, as long as the burden of doing so is proportional to the benefit to the Varnish community.

Do not file bug reports specific to these platforms without attaching a patch that solves the problem, we will just close it.

For now, anything else goes here, certainly the N900 and the WRT54G.

I'm afraid I have to put OpenBSD here for now, it is seriously behind on socket APIs and working around those issues is just not worth the effort.

If people send us a small non-intrusive patches that makes Varnish run on these platforms, we'll take it.

If they send us patches that reorganizes everything, hurts code readability, quality or just generally do not satisfy our taste, they get told that thanks, but no thanks.

Is that it ? Abandon all hope etc. ?

These tiers are not static, if for some reason Varnish suddenly becomes a mandatory accessory to some technically sensible platform, (zOS anyone ?) that platform will get upgraded. If the pessimists are right about Oracles intentions, Solaris may get demoted.

Until next time,

Poul-Henning, 2010-08-03