On Mon, 13 Jul 2015, Curtis Dunham wrote:
On July 13, 2015, 5:29 p.m., Nilay Vaish wrote:
src/sim/serialize.cc, line 472
<http://reviews.gem5.org/r/2930/diff/1/?file=47312#file47312line472>
Please explain the Pixel disagreement more.
Pixel doesn't have an operator<. If we added one, we could do a full
template instantiation with set<> using the existing infra by adding to
the INSTANTIATE_PARAM_TEMPLATES macro. It didn't seem that important
when we only need those two templates to be instantiated -
arrayParam{In,Out} for string sets.
Please add a line or two from here to the code so that the comment becomes
clear.
On July 13, 2015, 5:29 p.m., Nilay Vaish wrote:
src/sim/serialize.cc, lines 219-221
<http://reviews.gem5.org/r/2930/diff/1/?file=47312#file47312line219>
I can we are doing this separate handling of the first element for vector
and list as well. Do we really need it? Would we some problem in creating
tokens?
The checkpoint upgrader adds a bunch of superfluous whitespace as well
and everything seems reasonably robust to that. So I think the answer
Sometime back I was trying to compare a pair of original and upgraded
checkpoints and was really annoyed with the extra space the upgrader adds.
It is simply not possible to do a diff on the checkpoints.
is no, it's not really needed, but that's the way all the other
arrayParamOut methods are written.
OK. I am fine with the current patch. Would it be possible for you to
write another that does away with this special case handling? I am
guessing initially there was some purpose to it, but now that we don't
need it, we should do away with it.
Thanks
Nilay
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
