[llvm-commits] [llvm] r173270 - in /llvm/trunk: include/llvm/MC/MCAsmInfo.h lib/MC/MCAsmInfo.cpp lib/Target/MSP430/MCTargetDesc/MSP430MCAsmInfo.cpp lib/Target/Mips/MCTargetDesc/MipsMCAsmInfo.cpp lib/Target/NVPTX/MCTargetDesc/NVPTXMCAsmInfo.cpp lib/Target/PowerPC/MCTargetDesc/PPCMCAsmInfo.cpp lib/Target/Sparc/MCTargetDesc/SparcMCAsmInfo.cpp

[Eli Bendersky]

On Wed, Feb 6, 2013 at 2:16 PM, Jim Grosbach <grosbach at apple.com> wrote:
> Hi Eli,
>
> I really don't like this patch. This imposes additional burden on every target due to the odd requirements of one configuration of X86. That's really backwards. Can you elaborate on why you believe this is worth it?
>

Hi Jim,

This version was committed after a longish back-and-forth between
Nadav, Evan and myself (start
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20130114/162561.html
and look around to dig up all of it). We were considering a couple of
approaches, but eventually Evan asked to write it this way for the
sake of readability, since the alternative is an unsightly
special-casing of 0 which makes code harder to understand. I agreed
with him, and went ahead to change the patch to look like that.

FWIW, IMHO "imposes additional burden on every target due to the odd
requirements of one configuration of X86" can be said about a lot of
target-specific properties that currently exist. Assumptions have to
broken all the time because they don't hold with certain targets.
Pointer size != stack slot size is just another such assumption.

Eli