Document interned string location

[Caellian]

Documentation doesn't specify whether interned string will be moved.

I'm using StringInterner at the moment, but this clarification would be useful for all implementations.

I want to produce Cow<'inerner_owner, str> from strings and am wondering whether inserting new strings can cause their layout to shift.

Example:

• ["a", ..., "hello", "there", "world"], insert "hello world"
• ["a", ..., "hello world", "there"]

I'm assuming this isn't the case, but a backend that's optimized for minimal memory footprint could implement such optimizations, so it would be nice for docs to specify whether string position is persistent.

[Robbepop]

Hi, this is a good question and indeed we are missing docs for this. From what I remember this use case is exactly what the BucketBackend has been designed for.

[Caellian]

I'm repurposing this question into a task to:

• Note in documentation that: "returned symbols are stable across insertion, but resolved string references might become invalid for some implementations"
• Note whether resolved strings are stable in comparison table.

I'll do this shortly in a related PR.

[Caellian]

While responding I realized that besides documentation, this can be communicated with lifetimes as well.

#76 would allow constructing Cow<'i, str> via Cow::Borrowed(StringInterner::resolve) directly.

I also realized that StringInterner can't be used to construct Cow<'i, str> that are valid between insertions because an insertion might cause buffer: String to move if it can't grow in place.

Currently, none of the backends support this kind of use and #76 is basically the prerequisite.

I want something along the lines of string_cache (global static cache; refcounted freeing; double/triple indirection), but with more explicit control over when the cache is freed of values (i.e. no global mutex).