chore!: bump lru-cache to 10.x & increase node engine

[H4ad]

Trying to improve/solve the issue with dedupe on cli: npm/cli#7350

I don't know if it's a good idea since semver is being used by a lot of projects and bumping the version of node-lru-cache requires the bump of engines.node to >=16 , dropping a lot of node versions at once.

[wraithgar]

This would need to be part of #501 since changing engines is a semver major.

[domoritz]

Node 16 is not supported anymore so maybe this doesn't need to be a breaking change?

[wraithgar]

Changing the engines to drop support for any node version, supported or not, is a breaking change for a package.

[domoritz]

Okay, a major version bump makes sense then.

[mbtools]

This package doesn't use any fancy features of lru-cache. It's just caching up to 1000 parses.

How about we get rid of the dependency here and replace it with a simple class based on Map?

classLRUCache{constructor(max=0){if(!Number.isInteger(max)||max<0)thrownewTypeError('max must be a nonnegative integer');this.max=max;this.map=newMap();}get(key){constvalue=this.map.get(key);if(value===undefined)returnundefined;else{// Remove the key from the map and add it to the endthis.map.delete(key);this.map.set(key,value);returnvalue;}}has(key){returnthis.map.has(key);}delete(key){if(this.map.has(key)){this.map.delete(key);returntrue;}elsereturnfalse;}set(key,value){constdeleted=this.delete(key);if(!deleted&&value!==undefined){// If cache is full, delete the least recently used itemif(this.map.size>=this.max){constfirstKey=this.map.keys().next().value;this.delete(firstKey);}this.map.set(key,value);}returnthis;}clear(){this.map.clear();}capacity(){returnthis.max;}size(){returnthis.map.size;}*entries(){for(const[key,value]ofthis.map)yield[key,value];}}exportdefaultLRUCache;

I'm happy to provide a PR with 100% test coverage if you like.

PS: You could also argue whether LRU is actually the best caching strategy. FIFO might be much faster for the majority of cases. But that's a different story.

[wraithgar]

I'm happy to provide a PR with 100% test coverage if you like.

On one hand it does seem like a simple Map cache is all we need and should be easy to implement.

On the other hand "why don't you just" and "how hard can it be?" are usually hard lessons once one goes down the path of answering them.

If it's not too much trouble, you can make that PR. I think that will inform a lot of our decision better than an academic discussion would.

[H4ad]

Closing in favor of #697