syntax: Fix encoding and decoding spans

[erickt]

The protocol for serialize::{En,De}code doesn't allow for two integers to be serialized next to each other.

Closes#31025.

cc @michaelwoerister

[rust-highfive]

r? @sfackler

(rust_highfive has picked a reviewer for you, use r? to override)

[sfackler]

Large numbers are always somewhat sketchy to work with in JSON since they'll so often round trip through a double. Should we maybe encode with the packed number in a string?

[erickt]

@sfackler: True, although libserialize, rustc-serialize, and serde all are able to properly parse out u64 numbers, we can't assume other json parsers can read them. From what I understand, we only have to worry about dropping values if we ever get to 2^53 + 1. All other integers below that are representible in an f64. So I'm not sure how often this will come up in practice :)

[sfackler]

I've seen it hit in systems that use randomly generated 64 bit identifiers, for example.

In this case, I think it'd require a high span index greater than 2^21, which is roughly 2 million. It seems not wildly unreasonable to have 2 megabytes of source code for a crate, but I'm not sure.

[michaelwoerister]

How bad would it be for JSON to encode spans as structs? Because for the binary encoding that is used for crate metadata, there's no more overhead for these. The main reason for encoding it as two separate values was to allow for the variable-length integer format to kick in properly.

[erickt]

@michaelwoerister: Yeah, serializing them into structs would work.

We can save this for another time, but we could also optimize the rbml tuple encoding (and tuple structs) to match that of structs by leaving out the tag and the length encoding. We'd lose some safety, but it might shave some more bytes off our metadata.

[michaelwoerister]

@erickt No need to touch RBML, I think. The bulk of spans stored in metadata is within opaque sections where we don't have any of this overhead anymore since #30565. So I'd say, just do the clean solution and encode spans as structs.

[erickt]

@sfackler / @michaelwoerister: I've updated the patch to serialize Spans as a struct. Tests passed locally, and the size of the libraries seems to be identical to before this patch.

[michaelwoerister]

The field index should be 1 here.

[michaelwoerister]

r=me if you fix those two errors.

[erickt]

Thanks @michaelwoerister!

[michaelwoerister]

@bors: r+

[bors]

📌 Commit 1dc7eb8 has been approved by michaelwoerister

[bors]

⌛ Testing commit 1dc7eb8 with merge 9ae76b3...

[bors]

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-64-debug-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-x-android-t, auto-linux-cross-opt, auto-linux-musl-64-opt, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-win-gnu-32-nopt-t, auto-win-gnu-32-opt, auto-win-gnu-64-nopt-t, auto-win-gnu-64-opt, auto-win-msvc-32-opt, auto-win-msvc-64-opt