Re: Criteria for GCC 4.0

[Joe Buck <Joe dot Buck at synopsys dot COM>]

On Tue, Jun 01, 2004 at 12:51:25PM -0400, Scott Robert Ladd wrote:
> Indeed, the new C++ front end has caused great stress for those trying 
> to build distributions with GCC 3.4. I've been working with this for 
> Gentoo's distro, which still has a few packages that fail with GCC 3.4. 
> This was a major change that should have been communicated more clearly.

We've been talking about the new parser a lot for at least a year before
3.4.0 went out, and it was at the top of the "changes" document long
before there was a release.

Nevertheless, GCC gets far less attention than the Linux kernel gets, so
I'm not surprised that people were surprised.

> Documentation should not be a criteria for a new version number.
> 
> *However*, the current documentation is problematic at best, and 
> downright misleading, absent, and impenetrable at worst. The GCC 
> economic model needs to be changed such that documentation is a priority.

It's not possible to change the economic model.  In free software,
whatever contributors are willing to contribute and the maintainers are
willing to accept becomes the priority.  We've forced contributors to
provide at least minimal documentation by requiring that any patch that
adds a new option includes documentation for that option, and that the
tree-ssa stuff couldn't enter mainline without certain documents being
written.  Using tricks like that, we can force developers to provide at
least some documentation if they want their changes in.  But that, alone,
will not produce highly usable documentation.

It is possible for a sufficiently motivated volunteer to make a
difference, though, by leading an effort to restructure the
documentation and write new documentation.

> After reading the above, I am reminded of past conversations, and I tend 
> to agree that this release needs to be 4.0, not 3.5. However, almost 
> nothing mentioned above (besides the GDB requirement) is specific to C 
> developers, who appear to be a dominant factor (for obvious and valid 
> reasons). If the criteria is based on the mass of users -- most of whom 
> use C or C++ -- then I can see why they are resisting a bump to 4.0.

I anticipate that C++ code will see substantial improvement with tree-ssa.
We have a number of cases that already show large improvement.