From: N8TM AT aol DOT com
Subject: Re: i586-linux-pc -> i386-mingw32 [One man's porting experience]
1 Aug 1998 18:42:47 -0700
Message-ID: <983c4aac.35c32699.cygnus.gnu-win32@aol.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
To: Brown AT sw DOT mke DOT etn DOT com
Cc: gnu-win32 AT cygnus DOT com, egcs-bugs AT cygnus DOT com

In a message dated 7/31/98 8:14:23 PM Pacific Daylight Time,
Brown AT sw DOT mke DOT etn DOT com writes:
 I attempted to cross-compile egcs (1.0.2 with patches)
> error when it came time to
>  compile objc

Some of these problems with the objc build have gone away in the current egcs
snapshot, but some remain with certain combinations of OS/make versions.  For
example, on Irix 6.4, configuring with CC = 'cc -n32' works because that
matches the assumption made in the build stages which don't observe the CC
specification.  As late as egcs 1.03a, --with-gnu-as fails in objc on hpux1020
because objc wants to build with /bin/as instead of binutils, but that appears
to have been corrected in the snapshots.  I note when performing 'make
bootstrap' on NT/cygwin32 using gnu make that some recompilations in objc and
f are being skipped, with messages like 'can't find stage1/xgcc' with sections
shortly after where stage1/xgcc executes successfully. In the end, the build
is successful, although the stage to stage comparisons are bogus where there
wasn't actually a rebuild.  It gets worse trying to build on Win95/cygwin32
using the native make, where make (without bootstrap) is clean, but 'make
bootstrap' has errors.  So it can happen that a build without bootstrap gives
better results (but I don't ever test objc; I don't care if it works).  I
haven't dealt with using linux for cross-compilation, but clearly that is
cleaner than building under cygwin32.
-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request AT cygnus DOT com" with one line of text: "help".