Tuesday, 21 May 2013

on build systems

So i'm kind of baffled by gradle.

"power and flexibility of ant" with [enforced] "conventions of maven".

Sounds like it cherry picked the two worst parts of both outside of using XML!

Actually it looks ok enough for simple projects, but then again pretty much every tool is because solving simple problems is always ... simple. However I think the decision to go with implementing it in a scripting language is just going to lead to some pretty nasty long-term maintenance problems.

The only valid argument for something like ant is that the configuration files are machine readable (even if they aren't human readable!), which can lead to tooling support (ok, ant isn't very machine readable anyway, i'm just stating that it could be valid if they did it right). So it's kind of strange that gradle eschews that for something which is about as parseable as a batch file.

Of course it's the flash new kid on the block so it will go through a rapid adoption phase, but like every other tool before it cracks will then start to appear.

I'm also a little baffled by the claim that somehow groovy is just java and so it's easier for java developers. Doesn't look anything like java to me. At all. Actually even if it were true, i think that would be a problem not a benefit. Java is just not the right language to use for the problems that build systems solve.

At least it's better than ant, but that's a pretty low bar. At best ant isn't much better than a 'build-all.sh' file, and demonstrably worse in many ways.


I've put a few hours into getting somewhere on the java automake stuff. However I seem to have got stuck in an extended discussion on how a zip file works. The java build process is so simple I don't think anyone who is only familiar with C can grasp it.

I guess the main impression I get is that there isn't a particularly strong desire for simplicity vs 'the way we do it', which is a bit frustrating. If I end up with something I wouldn't want to use myself there doesn't seem much point. And given that in the intersection of the sets of 'i write java' and 'i want to use makefiles' and 'using automake isn't utterly and completely out of the question' i'm probably one of about a dozen unique and beautiful snowflakes, there isn't much hope if i'm not interested myself. Actually i may not use it anyway.

So although earlier I was more optimistic now i'm not sure where it's headed. I have some fragments which do part of the job but given the difficult i've had in explaining this simple external stuff i'm not sure I'm mentally up to trying to create and then explain any code inside automake.in. I'm not really that thrilled with the idea of trying to provide a complete patch anyway.

Most (big) projects seem to want every potential contributor to kow-tow to the whims of some god-like maintainer as if it's you the one who should feel privileged that they should deign to even entertain the idea of you doing free work for them. I'm ashamed this is exactly how we did things in Evolution and now regret it. There's quite a difference between a casual contribution and a long-term maintainer. I have no idea if automake is like that, but my patience threshold is pretty low these days so it wouldn't have to be for me to suddenly not to give a shit (i get paid to put up with crap, it's not something I need to volunteer for).

No comments: