Tracking issue for {Vec,String}::splice

[alexcrichton]

Tracking issue for rust-lang/rfcs#1432.

First needs to be implemented, then needs to go through FCP!

[aturon]

There's a preliminary implementation here, which could easily be picked up and brought over the finish line!

[clarfonthey]

Because this is implemented, is there anything blocking stabilisation?

[aturon]

@clarcharr Nope! I'll get the wheels in motion:

@rfcbot fcp merge

[rfcbot]

Team member @aturon has proposed to merge this. The next step is review by the rest of the tagged teams:

• @BurntSushi
• @Kimundi
• @alexcrichton
• @aturon
• @brson
• @dtolnay
• @sfackler

Concerns:

• IndexAssign resolved by Tracking issue for {Vec,String}::splice #32310 (comment)
• return value of String::splice resolved by Tracking issue for {Vec,String}::splice #32310 (comment)

Once these reviewers reach consensus, this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

[dtolnay]

Why would you ever use the return value from String::splice?

1. The String and the replacement str are in scope (start of 'a and 'b).
2. Call splice, receive Splice struct.
3. The String is exclusively borrowed and unusable.
4. Process the replaced segment one char at a time.
5. Drop the Splice struct.
6. Drop the String and the replacement str (end of 'a and 'b, must happen after 5).

Given that 'a and 'b are required to outlive Splice, wouldn't it always be simpler and at least as efficient to do this?

1. The String and the replacement str are in scope (start of 'a and 'b).
2. Grab the soon replaced segment &s[range].
3. The String is shared borrowed but otherwise usable.
4. Process the soon replaced segment one char at a time or as an ordinary &str.
5. Call splice which eagerly does the replace and returns ().
6. Drop the String and the replacement str (end of 'a and 'b).

@rfcbot concern return value of String::splice

[dtolnay]

If IndexAssign (or IndexSet or whatever, rust-lang/rfcs#997) existed today, would we be using that instead? Would we want both?

let mut s: String = /* ... */;
s[from..to] = "replacement";

@rfcbot concern IndexAssign

[clarfonthey]

I will say that splice would still be useful for Vec, just not for String in that case, because IndexAssign still won't be as efficient.

Perhaps going forward the features for Vec::splice and String::splice could be split, allowing to stabilise the former will keeping the latter unstable?

[sfackler]

I'm on board with stabilizing the Vec method, but not String for the reasons mentioned above. I think IndexSet is far enough away that I'd be comfortable landing this and then deprecating it in if/when that stabilizes.

[dtolnay]

I agree that IndexSet is too far away to block this.

@rfcbot resolved IndexAssign

[dtolnay]

I filed #44038 to follow up on changing the signature of String::splice.

@rfcbot resolved return value of String::splice

[dtolnay]

The following applies only to Vec::splice.

@rfcbot reviewed

[rfcbot]

🔔 This is now entering its final comment period, as per the review above. 🔔

[kennytm]

IMO if String::splice is not returning the previous data, the method shouldn't be really called splice.

[rfcbot]

The final comment period is now complete.

[dtolnay]

The FCP above applies to Vec::splice. I filed #44643 as the tracking issue for String::splice which remains unstable with its new signature.