Re: [platform-swt-dev] the build

[Havoc Pennington <hp@xxxxxxxxxx>]

Mike_Wilson@xxxxxxx writes: 
> Having said that, there is _no_better_tool_ for building the java code for 
> Eclipse than Eclipse itself. Period. Asking someone to build all of 
> Eclipse using an external command line driven tool is pretty much 
> equivalent to asking them to paint an image by manually typing in the XBM 
> data. Wanting to "rebuild the whole thing with [y]our compiler and tools" 
> doesn't make sense if the thing itself is the thing which should do the 
> building.

If Eclipse is a GUI tool only, there's no way we can use
automation. As I said, I'm all for self-hosting and using the Eclipse
build system for the build; but it should work out-of-box by having
all the info to build already in CVS, and it should be possible to do
in a command line or scripted fashion.  There is no reason the build
system needs to be GUI-only and require manual intervention to set up.
 
Purely from a process standpoint, automation is hugely important.  If
I have to rely on a human sitting there clicking buttons to get a
rebuild, then if I patch Eclipse, I can have bugs introduced because
the human clicked the wrong button or clicked different things. If I
have a fully automated rebuild, then I know with near-certainty that
any new bugs are from my patch, or from changing the toolchain. This
allows me to make patches with much more confidence (not to mention
much more quickly).

The same human error arises in that I can check out Eclipse from CVS
on two different machines, do the manual intervention to get it
building, and do that intervention differently each time. Then I could
have two slightly different results.

There's also the issue that in deploying Eclipse, one would doubtless
choose the Sun or IBM or gcj Java implementations to do so, and want
to support only that Java implementation. So that means someone
deploying it should be building all code with said implementation.

If nothing else, I'm lazy and don't feel like going through all the
steps I listed when I could just type a one-line command instead. ;-)


All that said, I'm not really that interested in arguing all day about
whether building it is or isn't a good idea - maybe you could convince
me, I'm open-minded - my point was more that typical Linux open source
contributors will expect to be able to build it, and will find it
bizarre and astonishing that they can't.

So, you can try to evangelize all however-many open source developers
that they don't really want to do this, or you can just give up and go
with the flow, so to speak. ;-) I'm just trying to offer advice about
the developers that I'm familiar with - I'm just the messenger here...

It's up to you to decide how important it is to appeal to these
developers, etc. - I'm not trying to tell you what to work on, just
give you some information from my perspective that might inform that
decision.

Havoc