Add set/reset/flip functions to change sequences

[Izaron]

Neither boost::dynamic_bitset nor vanilla std::bitset have the fill() function, and I propose to introduce this.
The point of the function is being used instead of loops with calling the set(pos) function.

Let's consider an example where we want either to set 1024 consecutive bits or to reset them (taking as the fact that we don't change the whole bitset, only its part).
Using a cycle we do it using, well, 1024 operations. Though we could have done it using maximum 1024/64+1=17 operations (considering the blocks as 64-bit integers) - setting the whole blocks between the first and the last bit, and optionally setting a suffix or a prefix of the first/last block respectively.

I also updated docs, tests. Yeah, the algorithm can look a bit complicated, and there are 2 helper functions added, but there's nothing hard at all.

Any suggestions, notes on possible performance improvement, arguments meaning (we change here [first; last], but can change [first;last) as it's more STL-like style); maybe naming too? I guess fill(size_type first, size_type last, bool val = true); could be named set(size_type first, size_type last, bool val = true); as well.

[codecov]

Codecov Report

Merging #27 into develop will increase coverage by 1.35%.
The diff coverage is 90.56%.

@@             Coverage Diff             @@
##           develop      #27      +/-   ##
===========================================
+ Coverage    75.41%   76.76%   +1.35%     
===========================================
  Files            5        5              
  Lines          541      594      +53     
  Branches       198      210      +12     
===========================================
+ Hits           408      456      +48     
- Misses          23       24       +1     
- Partials       110      114       +4

Impacted Files	Coverage Δ
include/boost/dynamic_bitset/dynamic_bitset.hpp	76.24% <90.56%> (+1.48%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b944aa3...f536fcd. Read the comment docs.

[jeking3]

This change is fine, however I think calling it set() makes more sense, and adding reset(f, l) and perhaps flip(f, l) so that all the bit manipulators can operate either on a single bit or a range of bits would complete the desired goal to modify sequences efficiently. As for the meaning of last as you discussed, I'm wondering if first_bit, num_bits makes sense instead of first_bit, last_bit. Then all instantiations of set forward to one implementation that would hopefully be optimal for all cases: set() -> set(0, num_bits), and set(bit) -> set(bit, 1). Just a thought - not sure it would work out that way but worth thinking about.

[Izaron]

@jeking3 I like your thoughts and agree with most of them.

Using relative last bit positions instead of absolute is done in std::string and it's logical to do the same here.

I decided to not call set(bit) -> set(but, 1) and similar, because the common case (any borders) takes some calculations and is more slower than just changing a bit, or setting all the bits to 1. Moreover current set functions are tiny.

But I have a serious doubt connected with implicit conversions from int to bool.
If a programmer paid no attention to warnings and/or had to code really fast, he could write this:

int okay = 42;
/* Transforming 'okay' */
db.set(pos, okay);

The last line was supposed to be db.set(pos, okay != 0) according to the current state, but having a new function dynamic_bitset& set(size_t n, size_t len, bool value = true) makes it into db.set(pos, okay, true) implicitly, which seems to me a serious thing...

Other things seem fine to me, I'd make reset() and flip() with tests as well, after some time.

[Izaron]

So, I've added some commits and there are three new range functions... With tests, docs, without code doubling. Anyone, feel free to comment them!

[jeking3]

This has led to ambiguity. See rdkit/rdkit#2128. We need to resolve this. I will open an issue.

[jeking3]

This needs to be resolved by Nov 7 or it will be reverted for the Boost 1.69.0 beta.