[llvm-dev] RFC: [SmallVector] Adding SVec<T> and Vec<T> convenience wrappers.

[Chris Tetreault via llvm-dev]

I appreciate the work you’ve done on this. My impression is that the thought process that goes into picking the size parameter is something like “Uhh… Uhh… I guess 4 is good?”, so doing work to formalize this is good. I also think that people are likely reluctant to use `SmallVector<T, 0>`.  That said, I don’t think adding new types is useful.

I think `SmallVector` should just get a default size that is computed as `SVec`’s default size parameter is. Honestly, I’m unconcerned that anybody would “forget” to populate the size parameter. I really feel like any value that isn’t backed by hard data is unlikely to be better than the result of ` calculateSVecInlineElements<T>()`, and should be questioned in code review given the new heuristic you’ve added.

In short, I think that `SmallVector` should get a default size, and that the documentation should become something to the effect of “SmallVector should be preferred to std::vector unless you have a good reason to use the stl container. (api interop?). The default size for SmallVector has been chosen to be reasonable for most use cases. If you have evidence that shows the default is not optimal for your use case you can populate it with a more suitable value.”

Again, thanks for working on this. I’m looking forward to the day when I don’t have to pick a number out of my backside to put in that parameter.

Thanks,
   Christopher Tetreault

From: llvm-dev <llvm-dev-bounces at lists.llvm.org> On Behalf Of Sean Silva via llvm-dev
Sent: Friday, November 13, 2020 2:06 PM
To: llvm-dev <llvm-dev at lists.llvm.org>; Duncan P. N. Exon Smith <dexonsmith at apple.com>; Mehdi AMINI <joker.eph at gmail.com>; Reid Kleckner <rnk at google.com>
Subject: [EXT] [llvm-dev] RFC: [SmallVector] Adding SVec<T> and Vec<T> convenience wrappers.

We've pretty happy now with a patch that adds two wrappers around SmallVector that make it 1) more convenient to use and 2) will tend to mitigate misuse of SmallVector. We think it's ready for wider discussion: https://reviews.llvm.org/D90884

SVec<T> is a convenience alias for SmallVector<T, N> with N chosen automatically to keep its size under 64 Bytes (that heuristic is easy to change though). The recommendation in the patch is to use this "on the stack, where a "small" number of elements are expected".

Vec<T> is a convenience alias for SmallVector<T, 0>. It lets us get the (little-known?) benefits of SmallVector even when it has no inline elements (see https://llvm.org/docs/ProgrammersManual.html#llvm-adt-smallvector-h). The recommendation in the patch is to use this when the SmallVector is on the heap.

A lot of this is boiled out from the discussion in https://groups.google.com/g/llvm-dev/c/q1OyHZy8KVc/m/1l_AasOLBAAJ?pli=1

The goals here are twofold:

1. convenience: not having to read/write "N", or do an extra edit/recompile cycle if you forgot it

2. avoiding pathological cases: The choice of N is usually semi-arbitrary in our experience, and if one isn't careful, can result in sizeof(SmallVector) becoming huge, especially in the case of nested SmallVectors. This patch avoids pathological cases in two ways:
  A. SVec<T>'s heuristic keeps sizeof(SVec<T>) bounded, which prevents pathological size amplifications like in `SmallVector<struct {SmallVector<T, 4> a, b; }, 4>`, where the small sizes effectively multiply together. Of course, users can always write SmallVector<T, N> explicitly to bypass this, but the need for that seems rare.
  B. SmallVector<T, 0> feels "weird to write" for most folks, even though it is frequently the right choice. Vec<T> mitigates that by "looking natural".

I'm surfacing this as an RFC to get feedback on a couple higher-level points:
- does everybody agree that SVec<T> and Vec<T> are useful to have?
- get wider consensus around suggesting these as "defaults" (see my updates to ProgrammersManual.rst in the patch)
- how much we want to bulk migrate code vs let it organically grow. Replacing SmallVector<T, 0> with Vec<T> should be completely mechanical. Replacing SmallVector<T, N> for general N would be a lot more work.
- of course: naming. SVector/Vector were floated in the patch as well and seem ok. SmallVec was rejected as it was a prefix of SmallVector (messes up autocomplete).

Looking forward to a world with fewer guessed SmallVector sizes,

-- Sean Silva
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20201117/e2b10916/attachment.html>