Re: preliminary java.nio with MapedByteBuffer channel

[Per Bothner <per at bothner dot com>]

Tom Tromey wrote:

> Per> Also, note how the natFileChannelImpl.cc #includes
> Per> natFileChannel{Posix,Win32,Ecos}.cc depending on
> Per> #defines in platform.h, without any symlinks.
>
> I really prefer the symlinks.  We already use it in several places, it
> is our standard approach for this sort of thing.

I really dislike the symlinks.  They provide no functionality beyond cpp 
(at least when it comes to C or C++ code); they are a needless 
complication in the build procedure (you have to look in configure.in) 
to figure out what gets linked to what); it's a mechanism that gdb and 
other tools don't know about; they violate the conventions that source 
files are in $srcdir; they are contrary to the autoconf model.  Worst of 
all, they assume a hard bifurcation (or trifurcation) of the target 
space which is inflexible and liable to change.  E.g. on CygWin and 
similar targets we might use Posix functions in some places and Win32 
functions in others, so the target isn't Posix *or* Win32 but *both*. 
Also, I can imagine can also see Ecos being (or becoming) very 
Posix-like depending how it is configured.
--
	--Per Bothner
per@bothner.com   http://per.bothner.com/