Re: GNOME 2.2 screenshots

Havoc Pennington wrote:

Tom Tromey <tromey redhat com> writes:
Havoc> Ideally the automake-1.6 package would not even contain a thing
Havoc> called simply "automake" since it just encourages bugs of the
Havoc> form "I got the wrong automake"

I don't think we'll do this.  Instead just put AUTOMAKE_OPTIONS = 1.6
into the top-level of any converted package.  Then you'll
get an error if you accidentally use 1.4.

That makes sense.

Havoc> If we can't do parallel install, we can't migrate to new
Havoc> automake module-by-module. That means that as soon as we move
Havoc> to 1.6 for HEAD, then people can't work on the stable branch
Havoc> anymore, unless they have two totally different build prefixes,
Havoc> with their own copies of autotools.

I don't understand.  Is this somehow related to the automatic rebuild

No. It's simply that we have to move all thousand-plus people building
out of CVS, and the dozens of modules they're building, as a whole.
Because without parallel install, people can only have one version of
automake installed. We can't change just one module to 1.6, because
users would then need to switch automakes, build that module, and
switch automakes back.  This is why we are still using 1.4, because
changing all this at once is just too hard.

If we could switch each module on its own time, we would have dumped
1.4 ages and ages ago. Because switching would be really easy, since
we wouldn't have to cause any pain/breakage in order to do it; we'd
just say "you have to install all the automakes" and then things would
Just Work for people and the packages could use a mix of automakes.

Maybe it's too late to fix; aside from renaming automake to
automake-1.4, there's the /usr/share/aclocal conflict issue, and all
that. I dunno.
I have been building the core gnome 2.0 modules with automake 1.6 for quite a while. I had to make adjustments to a number of modules in order to get them to build correctly. These changes were often real bugs. One example was bad use of AM_CONDITIONAL. Some's had the following in them:
   if test1; then
       AM_CONDITIONAL(FOO, test2)

With automake 1.6, this raises an error, but doesn't with 1.4. It is a real error with both automakes, as if test1 is false, then neither FOO_TRUE and FOO_FALSE get defined to '#', so both the true and false paths in the Makefiles get activated (which could give unpredictable results).

There were a number of other issues like this that were bugs that only got reported in 1.6. Given that most of the core modules can already be built out of CVS with 1.6 (not necessarily make distcheck'd), I think you are overstating the problems.


Email: james daa com au              | 2003 Call for Papers out
WWW: |

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]