Add support for lazy word generation

[tgross35]

No description provided.

[codecov]

Codecov Report

Attention: 14 lines in your changes are missing coverage. Please review.

Comparison is base (0e1f7c0) 77.19% compared to head (fc3fa0e) 77.04%.

Files	Patch %	Lines
zspell/src/dict/rule.rs	57.14%	12 Missing ⚠️
zspell/src/dict/tests_rule.rs	94.44%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #92      +/-   ##
==========================================
- Coverage   77.19%   77.04%   -0.16%     
==========================================
  Files          27       27              
  Lines        3464     3502      +38     
==========================================
+ Hits         2674     2698      +24     
- Misses        790      804      +14     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tgross35]

I need to take a look at how hunspell does this - we need a way to take the rules and apply them to only entries in the hashmap that were from the dictionary.

Maybe we should iterate hashmap for entries that have Dictionary as source. Then we could change Source::Dict to also hold the Vec<Arc<AfxRule>> of the rules attached to that entry. Then we find these entries and test if our given word matches.

[tgross35]

Hm. Maybe it would be more efficient if instead we stored a vector or hashset of all AffixRules. Then we iterate those rules and apply them, and see if the word exists in the

We might even be able to optimize this if we store it in a vector, dedup it. Then if we sort it by something useful, we could use a binary search to locate applicable rules.

As part of this, it could make sense to reverse AfxRulePattern and AfxRule so patterns point to rules rather than the other way around. Or a bidirectional reference where AfxRulePattern holds Arc<AfxRule> and AfxRule holds a Vec<Weak<AfxRulePattern>>. Or just store the needed data inline, AfxRule doesn't store much aside from the patterns.