Use `LocalDefId` as an index in metadata encoding by aerooneqq · Pull Request #163321 · rust-lang/rust

aerooneqq · 2026-09-25T10:54:13Z

The goal of this PR to eliminate encoding of a raw DefIndex, instead we should encode something (LocalDefId) that certainly indicates that index belongs to the local crate. The implementation of DefIndex encoding should always panic to prevent accidental remappings of non-local DefIndexes in #162809.

So the plan is as follows:

Use LocalDefId as a writing index in tables for metadata encoding, use LocalDefId instead of DefIndex everywhere, specialize encode/decode methods to correctly process it,
Replace implementation of encode_def_index in EncodeContext with panic, so we never remap and encode def index,
Make tables' keys two-sided: we can read and write using different index types, in this case we will write using LocalDefId and read using DefIndex (done in Allow using different index types when reading and writing to tables #163447).

Blocked by #162900.

r? @petrochenkov

petrochenkov · 2026-09-25T11:52:04Z

@bors try @rust-timer queue

Create `LocalDefIndex` for metadata encoding

rust-bors · 2026-09-25T12:50:30Z

☀️ Try build successful (CI)
Build commit: 2fa5a47 (2fa5a472bd28122b86b69706e673d2b3704e5c27)
Base parent: 4e701dc (4e701dc6ba63a40c908a173df269a55e1fd6297e)

rust-timer · 2026-09-25T13:48:43Z

Finished benchmarking commit (2fa5a47): comparison URL.

Overall result: ❌ regressions - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.8%	[0.8%, 0.9%]	4
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (secondary -0.5%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.5%	[-0.5%, -0.5%]	1
All ❌✅ (primary)	-	-	0

Cycles

Results (primary -1.6%, secondary 3.8%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	3.8%	[3.8%, 3.8%]	1
Improvements ✅ (primary)	-1.6%	[-1.6%, -1.6%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.6%	[-1.6%, -1.6%]	1

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 488.984s -> 488.935s (-0.01%)
Artifact size: 406.31 MiB -> 406.32 MiB (0.00%)

@petrochenkov

…ata, r=petrochenkov Allow using different index types when reading and writing to tables Make tables of metadata two-sided: one can write with one index type and read with another, as long as both those types are indexes. That will be used in rust-lang#163321 when we will have `LocalDefIndex` or similar type. r? @petrochenkov

@petrochenkov

…ata, r=petrochenkov Allow using different index types when reading and writing to tables Make tables of metadata two-sided: one can write with one index type and read with another, as long as both those types are indexes. That will be used in rust-lang#163321 when we will have `LocalDefIndex` or similar type. r? @petrochenkov

@petrochenkov

…ata, r=petrochenkov Allow using different index types when reading and writing to tables Make tables of metadata two-sided: one can write with one index type and read with another, as long as both those types are indexes. That will be used in rust-lang#163321 when we will have `LocalDefIndex` or similar type. r? @petrochenkov

@petrochenkov

…ata, r=petrochenkov Allow using different index types when reading and writing to tables Make tables of metadata two-sided: one can write with one index type and read with another, as long as both those types are indexes. That will be used in rust-lang#163321 when we will have `LocalDefIndex` or similar type. r? @petrochenkov

aerooneqq · 2026-09-29T11:44:50Z

After some experiments with LocalDefIndex and LocalDefId it seems that LocalDefId is a better option, so now waiting for #162900 to merge and after that this will be ready.

petrochenkov · 2026-10-02T12:08:11Z

@bors try @rust-timer queue

Use `LocalDefId` as an index in metadata encoding

rust-bors · 2026-10-02T13:12:22Z

☀️ Try build successful (CI)
Build commit: 9c851b5 (9c851b5dd100fb5b5547d9c0090ef7f1e97e902e)
Base parent: c165834 (c1658342b0ea7eab61db1960082668c77f553522)

rust-timer · 2026-10-02T15:59:37Z

Finished benchmarking commit (9c851b5): comparison URL.

Overall result: ✅ improvements - no action needed

Benchmarking means the PR may be perf-sensitive. It's automatically marked not fit for rolling up. Overriding is possible but disadvised: it risks changing compiler perf.

@bors rollup=never rustc-perf
@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.5%	[-0.5%, -0.5%]	2
Improvements ✅ (secondary)	-0.4%	[-0.7%, -0.2%]	31
All ❌✅ (primary)	-0.5%	[-0.5%, -0.5%]	2

Max RSS (memory usage)

This perf run didn't have relevant results for this metric.

Cycles

Results (secondary 4.6%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	4.6%	[2.7%, 6.5%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 491.121s -> 491.415s (0.06%)
Artifact size: 408.64 MiB -> 408.67 MiB (0.01%)

aerooneqq · 2026-10-02T16:41:47Z

cc @oli-obk as it is important prerequisite for #162809.

petrochenkov · 2026-10-03T12:15:32Z

    }
 }

+impl<'a, 'tcx> Decodable<MetadataDecodeContext<'a, 'tcx>> for LocalDefId {


Hm, why do local def ids ever need to be decoded from metadata?

Or is this somehow used for decoding external DefIndexes, which are not actually local?
Then it's a hack, that needs at least a good comment explaining why we have to do this.

Ideally, we'd of course decode DefIndexes directly, even if they were decoded as LocalDefIds, but I agree that it may be annoying if they are e.g. use the same structure in compiler/rustc_metadata/src/rmeta/mod.rs.

View changes since the review

When LocalDefId is in the value of some table we need to decode it, we can access tables' keys with different types as we created infrastructure for this, but when the same situation happens in values there is more needed to be done to support it.

Then it's a hack, that needs at least a good comment explaining why we have to do this.

Still needs a comment.

Or is this somehow used for decoding external DefIndexes, which are not actually local?

They are local to the crate being decoded, so those def indices are external for the current crate but local to external, it makes some sense why we decoding local def ids.

petrochenkov · 2026-10-05T14:50:56Z

(I'll finish the review later today, need to go now.)

rustbot assigned petrochenkov Sep 25, 2026

rustbot added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Sep 25, 2026