Folker Schamel wrote:
> Anders J. Munch:
> > The "make argument" is flawed. Read my 2008-06-10 posting
> > http://article.gmane.org/gmane.comp.version-control.subversion
> .user/77803
>
> Sorry, this makes no sense. The whole idea of the last modified
> timestamp is that it represents the time when this file was
> last modified.
Everyone agrees on that. The point of contention is whether moving
identical file contents from one place to another constitutes
modification.
Or should I say, there are two different traditions, one that says it
does, one that says it doesn't. Neither tradition is more right than
the other, they are just different. And those who see a file as
something that can be moved without changing it, would like to see
Subversion support their tradition better.
> Btw, for build systems of larger software, checksums are no option
> because they are far too slow. Thats the reason why for example
> also scons can use timestamps.
A build system that detects timestamp /change/ does not inherit make's
deficiencies.
The source of make's weaknesses is that it doesn't have its own
database, but relies solely on information that just happens to be
already available in the file system. A build system with memory can
outperform make both on speed and accuracy.
But that's a digression. My key point is this: if you use make, then
make /is/ going to fail you from time to time, and if you don't have
some strategy for coping with that, like say a forced full rebuild
before any release, then you're already in trouble.
It's time to let go of the myth that /usr/bin/make is some sort of
authoritative source on how to deal with timestamps.
regards,
Anders
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-07-15 18:40:51 CEST