[libcxx-commits] [PATCH] D109510: [libc++] string: Allocate exactly when growing from short to long string.

[Arthur O'Dwyer via Phabricator via libcxx-commits]

Quuxplusone added a comment.

> We have measured this overhead to be significant on our (Google) internal workloads, as 64% of global string allocations (and 26% of allocated bytes) are 48 bytes.

And after this patch, what are the numbers? Does it become, for example, "30% of global string allocations are 48 bytes and 34% are 32 bytes"?

Seems plausibly like a good idea to me, but I think your rationale is conspicuously incomplete without those "after" numbers.



================
Comment at: libcxx/include/string:2316
     pointer __old_p = __get_pointer();
+    const bool __was_short = __old_cap + 1 == __min_cap;
     size_type __cap = __old_cap < __ms / 2 - __alignment ?
----------------
FWIW, I'd rather see `bool __was_short = __old_cap < __min_cap;` — no const (for consistency with every other variable in this function even though, sure, we're not planning to modify any of them), and shorter/simpler condition. Assuming the condition `__old_cap < __min_cap` remains correct, that is.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109510/new/

https://reviews.llvm.org/D109510