improve perf for string with both strings and chars as argument

[KristofferC]

Before:

julia> @btime string("foo", 'b', "ar")
  443.838 ns (8 allocations: 288 bytes)

After

julia> @btime string("Foo", 'b', "ar")
  110.836 ns (3 allocations: 192 bytes)

Ref #27030 (comment)

[KristofferC]

@nanosoldier runbenchmarks("shootout" || "string" || "problem", vs=":master")

[KristofferC]

This is still 3x slower than 0.6 though (instead of 12).

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here. cc @ararslan

[mbauman]

Why not String(take!(io))?

[ararslan]

This consumes the data rather than making a copy.

[mbauman]

In this case, take! behaves the same way and doesn't make a copy. Ah, but I do understand now, the issue is that take! also is preparing the IOBuffer for additional writes, and allocates a new buffer to replace the one that becomes the string.

In Kristoffer's @btime string("foo", 'b', "ar"), this incurs a 30% penalty. Alright, carry on.

[JeffBezanson]

I wish take! didn't do that. In the vast majority (all?) of cases, an IOBuffer is not used after take! is called, so we should at least optimize for that case.

[KristofferC]

Opened #27741