best | Bayesian estimation supersedes the t test | Data Visualization library

Is there a way to work around this issue without imposing additional constraints on map key and value types?

It doesn't appear doable in safe Rust, at least not with reasonable algorithmic complexity. (See Aiden4's answer for a solution that does it by re-building the whole map.)

But if you're allowed to use unsafe, and if you're determined enough that you want to delve into it, this code could do it:

The functions provided in the Python re module do not optimize based on anchors. In particular, functions that try to apply a regex at every position - .search, .sub, .findall etc. - will do so even when the regex can only possibly match at the beginning. I.e., even without multi-line mode specified, such that ^ can only match at the beginning of the string, the call is not re-routed internally. Thus:

How about this?

Ok, so I don't know if it is the correct answer, but finally changing the settings in Eslint helped me to change the type of function for Components. I added the following rule to my .eslintrc.js file:

If you look closely at the specification of ranges​::​size in [range.prim.size], except when the type of R is the primitive array type, ranges​::​size obtains the size of r by calling the size() member function or passing it into a free function.

And since the parameter type of transform() function is reference, ranges::size(r) cannot be used as a constant expression in the function body, this means we can only get the size of r through the type of R, not the object of R.

However, there are not many standard range types that contain size information, such as primitive arrays, std::array, std::span, and some simple range adaptors. So we can define a function to detect whether R is of these types, and extract the size from its type in a corresponding way.

The TLDR is that loads which miss all levels of the TLB (and so require a page walk) and which are separated by address unknown stores can't execute in parallel, i.e., the loads are serialized and the memory level parallelism (MLP) factor is capped at 1. Effectively, the stores fence the loads, much as lfence would.

The slow version of your insert function results in this scenario, while the other two don't (the store address is known). For large region sizes the memory access pattern dominates, and the performance is almost directly related to the MLP: the fast versions can overlap load misses and get an MLP of about 3, resulting in a 3x speedup (and the narrower reproduction case we discuss below can show more than a 10x difference on Skylake).

The underlying reason seems to be that the Skylake processor tries to maintain page-table coherence, which is not required by the specification but can work around bugs in software.

For those who are interested, we'll dig into the details of what's going on.

I could reproduce the problem immediately on my Skylake i7-6700HQ machine, and by stripping out extraneous parts we can reduce the original hash insert benchmark to this simple loop, which exhibits the same issue:

A major reason to choose to use a sealed class instead of interface would be if there is common property/function that you don't want to be public. All members of an interface are always public.

The restriction that members must be public can be worked around on an interface using extension functions/properties, but only if it doesn't involve storing state non-publicly.

Otherwise, sealed interfaces are more flexible because they allow a subtype to be a subclass of something else, an enum class, or a subtype of multiple sealed interface/class hierarchies.

The non language lawyer answer is that there is a tie breaker rule for exactly this case.

Understanding standard wording well enough to decode it would require a short book chapter. But when deduced T&& vs T& overloads are options being chosen between for an lvalue and everything else ties, the T& wins.

This was done intentionally to (a) make universal references work, while (b) allowing you to overload on lvalue references if you want to handle them seperately.

The tie breaker comes from the template function overload "more specialized" ordering rules. The same reason why T* is preferred over T for pointers, even though both T=Foo* and T=Foo give the same function parameters. A secondary ordering on template parameters occurs, and the fact that T can emulate T* means T* is more specialized (or rather not not, the wording in the standard is awkward). An extra rule stating that T& beats T&& for lvalues is in the same section.

How about something like: