[META] Symbol Table

[MarcusDenker]

The issue tracker entry https://github.com/pharo-project/pharo/issues/8242 has some discussion how symbol interning was slowing down code loading.

The situation needs some careful analysis.

• How slow does interning get when we create a lot of new symbols on code load?
• Can we find places like the one in #8242 where we create symbols but do not need them automatically?
• Can we at least find places that create lots of symbols?
• The code for the symbol table is very old and has not seen any analysis. We know that other systems have a more clever implementation. What are the problems with the current approach?

[MarcusDenker]

It seems that a lot of symbols are small... and would be perfectly ok to be immediate objects

• no interning for most cases needed (identical by construction)
• we would just have a Symboltable for long symbols, this would be quite small and thus very fast to check

It would be interesting to have some statistics...

Marcus

[MarcusDenker]

Here is a list of issues related to Symbols and the SymbolTable

• [x] #compactSymbolTable leads to hash collisions #11319
• [ ] Speed up Symbol finding #12940
• [ ] Symbol>>#intern: of WideString: do we need the #isOctetString check? #11306
• [ ] String>>asSymbol triggers self error: 'There is no free space in this collection!' #6524
• [x] #hasPackageMatchingExtensionName: uses symbol concatenation #8242