This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: floating point war casualty list

Excerpt of message (sent 10 June 2002) by Richard Henderson:
> On Thu, Jun 06, 2002 at 10:08:08AM -0400, Paul Koning wrote:
> > I wonder if a branch is overkill.
> No, this is exactly what a branch is for.

As I see it, a branch allows you to do a significant change without
getting wrapped up in the schedule for the current mainline (or dot
release branch, in some cases, perhaps).  That would apply to large
and complex changes that are created over a significant period of time
-- DFA comes to mind as a classic example.  The large amount of time
may be in the testing as opposed to the coding.

So the question here is: what would that very extensive testing look
like and who might do it?  I have done a bunch -- see the description
in the patch I posted Friday.  I could possibly do more, but I don't
see how it could be a *lot* more given that all that's changing is the
conversion routines.  Maybe you could comment once you look at that

One possibility is that other work may be planned in real.[ch] (by
others; I don't have anything in mind) in which case that would be a
possible reason for branching.

> Note that it is possible to branch individual files in CVS,
> so if you expect to be modifying only real.[ch], that 
> reduces the load when it comes to merging.

I sort of knew that and then forgot.  Thanks for the reminder.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]