Re: before checking in

[Eli Zaretskii]

On Thu, 25 Oct 2001, John Wiegley wrote:

> I work on a C++ compiler with so many #ifdef's, I calculated we could
> produce several hundred different compilers from our one source base.
> And yet, it IS possible to enforce the rule that whatever is in CVS
> should be able to build and pass regressions.

It's possible, but requires a lot of resources and time that most (if 
not all) of Emacs developers cannot afford.

> Developers shouldn't have to test every configuration; that's what
> beta and pretesting are for.  But the code should build and run for
> the person making the change before it's checked in.

I don't think we disagree on this one; your original request sounded much 
more broad than this.

Still, there could be exceptions, like if a change is made which cannot 
be tested (or even compiled) on the developer's platform.  As a rule, 
this should be avoided, but exceptions are possible, in practice.

People who use CVS should be prepared to deal with problems that stem 
from this.  By contrast, people who build official releases, do not (and 
should not) expect such problems.

> It's just that whenever I see missing syntactic characters in a C file
> *checked in*, it makes me wonder.

If the relevant fragment is under an ifdef that didn't fire on the 
developer's system, it's not unheard of.