Re: Build System Changes

2007/2/14, Robert Schwebel <[email protected]>:
> On Mon, Feb 12, 2007 at 06:19:09PM +0600, Aleksey Demakov wrote:
> > What is exactly the need to require the latest versions of autotools?
>
> The latest is the greatest. They contain less bugs. The current automake
> files in dotgnu clearly show that people try to workaround bugs which
> obviously are not even documented enough that anybody remembers why the
> workarounds are there :-)

The grim reality is that the latest may contain less old and well known
bugs but every now and then it contains new bugs waiting for incautious
developer in the dark.

E.g. in gcc 4.1 they broke __builtin_frame_address() that worked perfectly
well in 3.x and 4.0. And we had to work around this problem.

> Golden Rule of Open Source (TM): If something doesn't work and you are
> using an old version, use the latest and greatest and your problems
> might magically go away.

The Golden Rule of software in general - free, open source. and proprietary
alike is:

If it ain't broken, don't fix it.

> > I'd would rather stick to a more conservative approach of changing
> > version requirements. Like if it becomes a really big maintenance
> > burden to support a particular old version we might move the
> > prerequisite version just past it.
>
> Do you have concrete arguments against the version set I mentioned? I
> mean - I have no problem with another one if it is well argued, but it
> has to be defined somehow, and if nobody sees real issues with the
> latest and greatest - why not use it?

The same way I may ask you if you have any arguments *for* the
version you mentioned. And the "latest and greatest" argument
does not qualify.

It is purely a matter of personal taste whether you prefer "latest and
gretest" or "old and proven."  And I am not going to discuss iif any one
of these approaches is inherently better than the other. This would
be an absolutely pointless discussion.

An argument that would qualify for me will go like this:

the version x.x.x has added a feature y or fixed a bug z that solves
the folowing existing problem in pnet.

I agree that we may scrap the support for that old version of autotools
you found those ugly workarounds for. This is absolutely ok with me.
But asking everyone to upgrade to the latest version is too much.

I know that in corporate environments it is common to use a single
toolset for entire development team. There is a good reason for it.
Maintaining a multitude of configurations on many machines and
resolving incompatibilities may become a serious hassle.

On the other hand, they have admins to take care of upgrades, QA
departments with a host of different systems to test on.

I do not have neither personal admin nor personal QA team. If after
a long day I spared an hour or two for JIT hacking late in the night and
all of a sudden I find out that I have to upgarde my tool chain for no
apparent reason this hardly would make me happy. I have too many
other things to do apart from tracking the latest developemets in
autotools land, thank you.

Regards,
Aleksey