Re: Emacs 21 on AIX 4.3

[Eli Zaretskii]

[Note that I moved this discussion to emacs-devel.  I think
emacs-pretesters is not appropriate for it anymore.]

On Mon, 25 Jun 2001, Tak Ota wrote:

> Regarding the GUI and w32xxx stuff, I consider this cygwin port to be
> another Emacs port to a variant of unix system since the cygwin layer
> pretty much can hide the underlying Windows.  Therefore I am ignoring
> w32xxx files all together.  To answer your question I personally think
> Cygwin port should not rely on the native Windows GUI.  The port
> should be a legitimate X client since cygwin provides a port of X
> server by itself.

I'd advise against such a categorical approach.  Dumping the
Windows-specific code means you lose all the experience that was
earned by hard labor over the past 5 years of NTEmacs maintenance.
That experience should not be discarded too easily; it is what makes
NTEmacs a much better Windows citizen than XEmacs.

With all due respect to Cygwin, it still doesn't solve some of the
idiosyncrasies of Windows.  Some of those problems cannot be solved at
all on a library level (e.g., text vs binary I/O), because only the
application knows what's right in each case.

In other words, Windows isn't Unix even when using the Cygwin
runtime.  You can see the evidence of this in Emacs-related news
groups and on the Cygwin mailing list: people keep describing problems
with small incompatibilities that cannot be easily solved.

One particularly nasty problem with dumping Windows code is that you
lose interoperability with anything but Cygwin applications.  That is
IMHO a very bad idea; the XEmacs experience shows that many people
avoid the Cygwin build because they need to run non-Cygwin
applications.  I think Emacs should not repeat that mistake.

I also don't see why the Cygwin build should force the use of the
ported Xlib.  That will almost certainly make the Cygwin port
significantly slower than if it used the native GUI toolkit.  Why
should Cygwin users be punished like that?

To summarize: I think someone _does_ have to walk through the Windows
code in NTEmacs and decide in each case what should be retained in the
Cygwin port.  Building with Cygwin as if you were on Unix might be the
easy way out, but IMHO it's the wrong way.  Emacs should support
Cygwin, but not at the expense of being useful in conjunction with
other Windows software.