Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!thunder.mcrcim.mcgill.edu!snorkelwacker.mit.edu!apple!usc!cs.utexas.edu!nthropy!andy
From: andy@nthropy.UUCP (Andy Lowe)
Newsgroups: comp.windows.x
Subject: right shifting negative numbers considered harmful
Message-ID: <9101231726.AA05052@nth.com>
Date: 23 Jan 91 17:26:28 GMT
Lines: 30
Posted-Date: Wed, 23 Jan 91 11:26:28 CST
To: comp-windows-x@cs.utexas.edu


Dear X server hackers,

My copy of the ANSI C spec says that when right shifting E1 >> E2, "If
E1 has a signed type, whether the right shift is logical (fill with
zeros) or arithmetic (fill with a copy of the sign bit) is
implementation-defined." (section 3.3.7)

Consider the following fragment of code from the MIARCOCTANTSHIFT macro
in mizerarc.h  

...
	    d = b - (a >> 1) - d + (k3 >> 3); \
...

this is called out from the cfb arc drawing routines.  Note that
variable a can be negative.

I have the misfortune to be running on an architecture without an
arithmetic right shift instruction, and so my compiler has exercised
it's "implementation-defined" option.  

In the interests of portability, can someone arrange to fix this
apparent ANSI non-compliance in the standard distribution?  I'm using
R4.  Perhaps it is already fixed in R5.  Does anybody know?

many thanks.

Andy Lowe
andy@nth.com