[llvm] r210414 - Do materialize for floating point

[Duncan P. N. Exon Smith]

> These are mostly my opinion with little evidence to back them up but there's a couple fairly weak reasons. The first is that I'd like to avoid the scenario where MIPS patches that need wider review regularly find it hard to attract reviewers because they are difficult to spot among the more mundane patches. To put it another way, my concern is that CC'ing llvm-commits on everything may encourage many people to think of '[mips]' as a spam tag.

An alternative way to draw attention to MIPS patches that need wider
review is to say exactly that in the review thread -- and if there's
someone specific whose eyes you'd like, CC them.  If it's more of a
directional question, a post to llvmdev can help too.

> The second is that CC'ing llvm-commits seems to make people nervous, occasionally to the point of asking for a pre-pre-commit review. Interestingly, the current approach doesn't seem to have the same effect even though the outcome is pretty much the same. It seems that people don't mind admitting mistakes too much, but they like to know what those mistakes are before they reveal them to the world.

I think the right answer to "pre-pre-commit" review is "no" -- that's the
answer I received when I asked for it.  Forcing me to post my patches on
the list without prior internal peer review drew me into the wider
community, and expanded my peer group from "Apple LLVM developers" to
"LLVM developers".  Public peer review isn't just about visibility -- it's
also about engagement.

Of course, I was nervous about it, and considered my patches with more
care as a result -- probably not a bad outcome either.