Introduce custom hash table data structures.

[chandlerc]

The hash table design is heavily based on Abseil's "Swiss Tables" design. It uses an array of bytes storing metadata about each entry and an array of entries where each is a pair of key and value. The metadata byte consists of 7-bits of hash of the key (distinct from the bits used to index the table), and one bit indicating the presence of a special entry -- either empty or deleted.

TODO: document the design and PR more fully

[josh11b]

I wanted to give feedback on map.h early, before you update set.h

[chandlerc]

Thanks, definitely useful to get map.h tidied up before I port it to set.h. I think I've responded to all the comments there.

[josh11b]

Sending what comments I have so far.

[chandlerc]

Thanks for all the feedback so far, PTAL.

[josh11b]

Finished a pass through raw_hashtable.h

[chandlerc]

PTAL!

Beyond inline replies, also was able to simplify the specialization strategy for the table type and improved some other comments spotted as I was working.

Also, think this is far enough along in review I'm moving it out of draft.

[chandlerc]

PTAL!

Beyond inline replies, also was able to simplify the specialization strategy for the table type and improved some other comments spotted as I was working.

Also, think this is far enough along in review I'm moving it out of draft.

[chandlerc]

So, I'm most of the way through adding really nice support for handling stateful keys like indices into vectors and such. It turned out to be necessary to even move the toolchain over to these data structures as it also lets us use something other than operator== for equality, and we need that for APFloat keys. There are a few interesting things that emerged:

• I've ended up removing unnecessary functionality that also makes the stateful case unimplementable.

• I've found and cleaned up a number of innocuous inconsistencies in the code.

• I've been able to implement the test you asked for where we control the hashing and force every key to collide. I can implement even more of these if desired.

Wanted to both let you know @josh11b about my progress there, and also ask whether you'd like me to merge this into this code review or keep it as a follow-up. I can easily manage either way.

[josh11b]

So, I'm most of the way through adding really nice support for handling stateful keys like indices into vectors and such. It turned out to be necessary to even move the toolchain over to these data structures as it also lets us use something other than operator== for equality, and we need that for APFloat keys. There are a few interesting things that emerged:

• I've ended up removing unnecessary functionality that also makes the stateful case unimplementable.
• I've found and cleaned up a number of innocuous inconsistencies in the code.
• I've been able to implement the test you asked for where we control the hashing and force every key to collide. I can implement even more of these if desired.

Wanted to both let you know @josh11b about my progress there, and also ask whether you'd like me to merge this into this code review or keep it as a follow-up. I can easily manage either way.

I'm fine either way. If it simplifies the files haven't reviewed yet, it would be a benefit.