It is slightly slower, but correctnes is, of course, more important:
```bash
benchmark old ns/op new ns/op delta
BenchmarkSortByWeightAndReverse-4 367 645 +75.75%
benchmark old allocs new allocs delta
BenchmarkSortByWeightAndReverse-4 2 2 +0.00%
benchmark old bytes new bytes delta
BenchmarkSortByWeightAndReverse-4 64 64 +0.00%
```
Running the same benchmark without any cache (i.e. resorting the slice on every iteration) and then compare it to the current version shows that it still is plenty worth it:
```bash
▶ benchcmp 2.bench 1.bench
benchmark old ns/op new ns/op delta
BenchmarkSortByWeightAndReverse-4 1358757 645 -99.95%
benchmark old allocs new allocs delta
BenchmarkSortByWeightAndReverse-4 17159 2 -99.99%
benchmark old bytes new bytes delta
BenchmarkSortByWeightAndReverse-4 274573 64 -99.98%
```
Closes#5239
As an example:
```html
{{ $pages := .Site.RegularPages | lang.Merge $frSite.RegularPages | lang.Merge $enSite.RegularPages }}
```
Will "fill in the gaps" in the current site with, from left to right, content from the French site, and lastly the English.
Fixes#4463
This closes#98, even if this commit does not do full content text search.
We may revisit that problem in the future, but that deserves its own issue.
Fixes#98
The custom sort functions used from the templates had some subtle data race- and related issues,
especially when used in the single page template.
This commit fixes this by making copies and protect the read and writes with a RWMutex.
The results are cached (it will typically be invoked *number of pages* times with exactly the same data).
This is, not surprisingly, also faster:
```
benchmark old ns/op new ns/op delta
BenchmarkSortByWeightAndReverse 14228 491 -96.55%
benchmark old allocs new allocs delta
BenchmarkSortByWeightAndReverse 1 0 -100.00%
benchmark old bytes new bytes delta
BenchmarkSortByWeightAndReverse 32 0 -100.00%
```
Fixes#1293
