bound | Implements a nice helper for fast boundary definitions | REST library

The intent is to disallow redeclaring structured bindings as references. See CWG 2313.

I wish someone can prove me wrong, but this is one of the few corner cases where we meet a limitation of current GHC and we can not get away from using Proxy, Tagged or similar "relics" of the past.

Let's consider a simpler example:

Not a perfect solution, I must admit. But it's simple and comes pretty close.
I create 5 vectors that are gonna span the space of the matrix and create random linear combinations to fill the rest of the matrix. My initial thought was that a trivial solution will be to copy those vectors 20 times.
To improve that, I created linear combinations of them with weights drawn from a uniform distribution, but then the distribution of the entries in the matrix becomes normal because the weighted mean basically causes the central limit theorm to take effect.
A middle point between the trivial approach and the second approach that doesn't work is to use sets of weights that favor one of the vectors over the others. And you can generate these sorts of weight vectors by passing any vector through the softmax function with an appropriately high temperature parameter.
The distribution is almost uniform, but the vectors are still very close to the base vectors. You can play with the temperature parameter to find a sweet spot that suits your purpose.

To measure the branch you are interested in and particularly the scenario when while loop becomes hot, I've used the following benchmark:

The real reason is in V8 memory optimization. When you store integers - it stores the 32 bit number in place, But when you store double-number - it is stored differently (as an object) - so yValues array contains the reference but the actual value stored in heap. So in your example you just used all heap memory. To see the limit, use: console.memory and you'll see something like this:

Cache misses. When N int objects are allocated back-to-back, the memory reserved to hold them tends to be in a contiguous chunk. So crawling over the list in allocation order tends to access the memory holding the ints' values in sequential, contiguous, increasing order too.

Shuffle it, and the access pattern when crawling over the list is randomized too. Cache misses abound, provided there are enough different int objects that they don't all fit in cache.

At r==1, and r==2, CPython happens to treat such small ints as singletons, so, e.g., despite that you have 10 million elements in the list, at r==2 it contains only (at most) 100 distinct int objects. All the data for those fit in cache simultaneously.

Beyond that, though, you're likely to get more, and more, and more distinct int objects. Hardware caches become increasingly useless then when the access pattern is random.

Illustrating:

You're right that a smarter compiler should be able to resolve this unambiguously.

The way Java resolves method invocations is complex. It's defined by the JLS, and I make it 7500 words purely to determine how to resolve a method. Pasted into a text editor, it was 15 pages.

The general approach is:

1. Compile-Time Step 1: Determine Type to Search (no issue here)
2. Compile-Time Step 2: Determine Method Signature
   1. Identify Potentially Applicable Methods
   2. Phase 1: Identify Matching Arity Methods Applicable by Strict Invocation
   3. Phase 2: Identify Matching Arity Methods Applicable by Loose Invocation
   4. Phase 3: Identify Methods Applicable by Variable Arity Invocation
   5. Choosing the Most Specific Method
   6. Method Invocation Type
3. Compile-Time Step 3: Is the Chosen Method Appropriate?

I don't understand anywhere close to all of the details and how it pertains to your specific case. If you care to dive into it then I've already linked the full spec. Hopefully this explanation is good enough for your purposes:

Ambiguousness is determined at step 2.6, but there is still a further appropriateness check at step 3. Your foo method must be failing at step 3. Your bar method never makes it that far because the compiler still considers both methods to be valid possibilities. A human can make the determination that the non-appropriateness resolves the ambiguity, but that's not order the compiler does things. I could only speculate why - performance might be a factor.

Your code is operating at the intersection of generics, overloading and method references, all three of which were introduced at different times; it's not massively surprising to me that the compiler would struggle.

There appears to be something very wrong with the calculated GC CPU time. It's 41010 secs compared to 2737 sec elapsed, which doesn't make sense if you're only running on three capabilities.

This miscalculation means that the calculated MUT CPU time, which is just total CPU time minus INIT, GC, and EXIT time, is actually a large negative number (5073-41010-2 = -35939). This gives a productivity of -35939/5073=-708%. When the MUT seconds are displayed, negative numbers are truncated at zero, to avoid reporting small negative numbers when MUT is very low and there's a clock precision error, which is why the displayed MUT time is 0 instead of -35939.

I don't know why the GC time is so badly miscalculated. My best guess is this. If you're running on Windows, there are known issues with CPU time clock precision, and it's possible that certain unusual patterns of garbage collection timing might result in precision errors occuring in only one direction, slightly overestimating the actual GC time more often than it underestimates it. Over 2.4 million collections (see your GC stats), this difference could accumulate to a huge positive error.

I looked through GitLab issues, and except for the report on general Windows CPU time imprecision and a couple of probably unrelated negative MUT reports here and here, I didn't see anything helpful.

What about this? It is shorter than the full signature: