fix: fixing asymmetric diamond dependency problem

[divdavem]

This is a big refactoring of the algorithm so that the asymmetric diamond dependency problem is fixed:

const a = writable(0);
const b = derived(a, (a) => `b${a}`);
const c = derived([a, b], ([a, b]) => `${a}${b}`);
const values = [];
const unsubscribe = c.subscribe((value) => values.push(value));
a.set(1);
console.log(values); // displays ["0b0","1b0","1b1"], should display ["0b0","1b1"]

For info, the asymmetric diamond dependency bug is also present in svelte stores:
https://svelte.dev/repl/445983f3fdfe4db9ab860d88642f1989?version=3.48.0

A possible workaround (both in ngx-tansu and in svelte stores) is to transform it into the symmetric diamond dependency problem:

const a = writable(0);
const abis = derived(a, (a) => a); // add a useless level of derived store to work around the issue
const b = derived(a, (a) => `b${a}`);
const c = derived([abis, b], ([a, b]) => `${a}${b}`);
const values = [];
const unsubscribe = c.subscribe((value) => values.push(value));
a.set(1);
console.log(values); // correctly displays ["0b0","1b1"]

But it is better to have a proper fix. That's why I am opening this PR!

This PR makes invalidate calls recursive so that the whole tree of dependent stores is prevented from updating while values are changed. As a consequence, there is a need of the reverse revalidate operation (equivalent to next with the same value, but avoiding re-computations) in case the value does not change.
I have not measured the impact on performance in case the tree of dependencies is big. The invalidate listener functions should be kept as small and quick as possible (just switching a boolean and propagating the info).

This PR adds several tests for different cases in addition to the asymmetric diamond dependency case.

[codecov]

Codecov Report

Merging #31 (9039b73) into master (ae35b18) will increase coverage by 0.29%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #31      +/-   ##
==========================================
+ Coverage   98.46%   98.75%   +0.29%     
==========================================
  Files           1        1              
  Lines         130      161      +31     
  Branches       24       28       +4     
==========================================
+ Hits          128      159      +31     
  Partials        2        2              

Impacted Files	Coverage Δ
src/index.ts	98.75% <100.00%> (+0.29%)	⬆️

---

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 ae35b18...9039b73. Read the comment docs.

[divdavem]

@pkozlowski-opensource Feel free to have a look at this PR and to give some feedback if you have time.

[fbasso]

It is fine on my side, but probably need another look from another reviewer.

@maxokorokov, can you have a look ?

[pkozlowski-opensource]

@divdavem oh, cool to see that you are digging into the batched and glitch-free propagation algorithm! I don't think I will have time to do any meaningful review so I don't want to block progress here.

[divdavem]

@fbasso Thank you for your review!

[divdavem]

@divdavem oh, cool to see that you are digging into the batched and glitch-free propagation algorithm! I don't think I will have time to do any meaningful review so I don't want to block progress here.

@pkozlowski-opensource Thank you for your comment! It is great to hear from you! I understand you are very busy!
By the way, did you receive my e-mail about npm rights to publish a new version of ngx-tansu?

[maxokorokov]

LGTM, apart from invalidate/revalidate naming.

I find revalidate confusing, seems like uninvalidate would fit better, but I don't think it's a word. And I don't have better options.

freeze/unfreeze, pause/resume, disable/enable, I don't know...

[divdavem]

@maxokorokov Thank you for your review!
I don't know either what names would be the best. I will integrate this PR as it is, then we can rename the functions in a separate PR.