experimental/currency: Implement runtime fallback for missing patterns and remove constructor validation by younies · Pull Request #8066 · unicode-org/icu4x

younies · 2026-06-11T14:06:02Z

This PR enables support for incomplete or custom currency data by implementing runtime fallback for missing patterns in CurrencyEssentials and removing strict constructor validation in formatters.

Previously, we attempted to make positive patterns optional in PatternIndices, but reverted that approach to avoid footprint overhead in standard data. Instead, we keep positive patterns as non-optional u8 indices, resolving fallbacks at datagen time for standard data, while introducing a runtime fallback safety net in the getters to support custom data.

Changelog

Runtime Fallback in CurrencyEssentials:
- Positive pattern getters now return non-optional &DoublePlaceholderPattern. If an index is out of bounds (indicating a missing pattern in custom data), it falls back according to the hierarchy, with FALLBACK_PATTERN ("{1}{0}") as the ultimate fallback.
- Established positive pattern fallback hierarchies:
  - standard_alpha_next_to_number_positive ➔ standard
  - accounting_positive ➔ standard
  - accounting_alpha_next_to_number_positive ➔ standard_alpha_next_to_number ➔ standard
- Negative pattern getters return Option<&DoublePlaceholderPattern>, with accounting_negative falling back to standard_negative if missing (None).
Relaxed Formatter Validation:
- Removed standard pattern validation from CurrencyFormatter and CompactCurrencyFormatter constructors. Formatters now initialize successfully even if the standard pattern is missing, relying on runtime fallback during formatting.
Documentation & Cleanups:
- Documented fallback hierarchies in CurrencyEssentials getters.
- Added comments in datagen explaining datagen-time fallback resolution.
Data:
- Regenerated experimental baked data.

TAG=agy
CONV=de8f6008-79a8-45a5-9a6f-3332bb99904f

…ternIndices Make standard_alpha_next_to_number, accounting_positive, and accounting_alpha_next_to_number_positive optional (Option<u8>) in PatternIndices to support true runtime fallback and avoid datagen-time fallback duplication. Also updated getters in CurrencyEssentials to implement the fallback hierarchies at runtime and added documentation comments for them. Regenerated experimental test data. TAG=agy CONV=7de8b65e-8ff1-43aa-9dbd-1a73ee7b45ff

younies · 2026-06-11T14:07:17Z

+        debug_assert!(
+            (self.indices.standard as usize) < self.patterns.len(),
+            "Standard pattern index {} is out of bounds for patterns of length {}",
+            self.indices.standard,
+            self.patterns.len()
+        );


@robertbastian : https://github.com/unicode-org/icu4x/pull/8050/changes/BASE..36013c38afcdb0e14dbca1c66a9e515400bd8b5d#r3396052964

please see my comment on the other PR

add by @younies #8050 (comment)

robertbastian

no, for several reasons:

an Option<u8> takes more space than a u8
it shifts work that can be done at datagen time to runtime

…alidation Revert the changes that made positive patterns optional in PatternIndices. Positive patterns are now u8 (non-optional) again, and fallback is resolved at datagen time. Added runtime fallback to FALLBACK_PATTERN in CurrencyEssentials getters for custom data support. Removed standard pattern validation from CurrencyFormatter and CompactCurrencyFormatter constructors, allowing them to succeed and use runtime fallback if the pattern is missing. Regenerated experimental baked data. TAG=agy CONV=2329ca53-2c21-4bb8-bc21-4b69db92e9cc

- Remove `.unwrap()` from positive pattern getters in `icu_provider_source` tests, as positive patterns are now non-optional. - Add `#[cfg(feature = "chrono")]` to `test_against_chrono` in `zoneinfo64` to fix compilation failure when `chrono` feature is not enabled. TAG=agy CONV=2329ca53-2c21-4bb8-bc21-4b69db92e9cc

younies · 2026-06-15T13:56:26Z

I have made all pattern indices u8 except for the negative ones. At runtime, we need to know if a negative pattern does not exist so we can fall back to the decimal formatter.

Regarding the struct size, we will address that when we add all the patterns and implement the ULE.

robertbastian · 2026-06-15T15:00:42Z

+// we have the fallback here for users feeding their own data without
+// handling the fallback logic in their data generation.


You should document the invariant that the standard pattern always exists, and phrase any case where it doesn't as a condition that we don't support.

Suggested change

// we have the fallback here for users feeding their own data without

// handling the fallback logic in their data generation.

// we have the fallback here in case the data struct is corrupted.

robertbastian · 2026-06-15T15:03:36Z

+        debug_assert!(
+            (self.indices.standard as usize) < self.patterns.len(),
+            "Standard pattern index {} is out of bounds for patterns of length {}",
+            self.indices.standard,
+            self.patterns.len()
+        );
        self.patterns
            .get(self.indices.standard as usize)
-            .or_else(|| self.patterns.get(0))
+            .unwrap_or(FALLBACK_PATTERN)


Suggested change

debug_assert!(

(self.indices.standard as usize) < self.patterns.len(),

"Standard pattern index {} is out of bounds for patterns of length {}",

self.indices.standard,

self.patterns.len()

);

self.patterns

.get(self.indices.standard as usize)

.or_else(|| self.patterns.get(0))

.unwrap_or(FALLBACK_PATTERN)

self.patterns

.get(self.indices.standard as usize)

.unwrap_or_else(|| {

debug_assert!(

false,

"Standard pattern index {} is out of bounds for patterns of length {}",

self.indices.standard,

self.patterns.len()

);

// GIGO

FALLBACK_PATTERN

})

robertbastian · 2026-06-15T15:05:29Z

    }

    #[test]
+    #[cfg(feature = "chrono")]


don't make unrelated changes.

robertbastian · 2026-06-15T15:11:14Z

+    /// Even though the baked data handles the fallback at data generation time,
+    /// we have the fallback here for users feeding their own data without
+    /// handling the fallback logic in their data generation.


there is no fallback here

robertbastian · 2026-06-15T15:13:07Z

        self.patterns
            .get(self.indices.standard_alpha_next_to_number as usize)
-            .or_else(|| self.standard_pattern())
+            .unwrap_or_else(|| self.standard_pattern())


Suggested change

.unwrap_or_else(|| self.standard_pattern())

.unwrap_or_else(|| {

debug_assert!(

false,

"Standard-alpha-next-to-number pattern index {} is out of bounds for patterns of length {}",

self.indices.standard,

self.patterns.len()

);

// GIGO

FALLBACK_PATTERN

})

robertbastian · 2026-06-15T15:14:39Z

+    /// Fallback hierarchy:
+    /// `standard_alpha_next_to_number` -> `standard`
+    ///
+    /// Even though the baked data handles the fallback at data generation time,
+    /// we have the fallback here for users feeding their own data without
+    /// handling the fallback logic in their data generation.


don't document like this. we should treat the fallback as a datagen concern, here we just assume we have a value for each type

Suggested change

/// Fallback hierarchy:

/// `standard_alpha_next_to_number` -> `standard`

///

/// Even though the baked data handles the fallback at data generation time,

/// we have the fallback here for users feeding their own data without

/// handling the fallback logic in their data generation.

what will happen if a user fed the data without handling the fallback in their datagen?

if they break a documented invariant on the data struct, that's their problem

robertbastian · 2026-06-15T15:15:26Z


    /// Returns the `standard_alpha_next_to_number` negative pattern if specified, falling back to standard negative.
+    ///
+    /// Fallback hierarchy:


issue: handle the negative fallbacks at datagen time as well. they can still be Option<u8>

younies requested review from a team, Manishearth, robertbastian and sffc as code owners June 11, 2026 14:06

younies removed request for Manishearth and sffc June 11, 2026 14:06

younies mentioned this pull request Jun 11, 2026

currency: Pack distinct patterns in CurrencyEssentialsV1 and support explicit standard negative patterns #8050

Merged

younies commented Jun 11, 2026

View reviewed changes

fix fmt

8b03327

robertbastian requested changes Jun 11, 2026

View reviewed changes

younies added 6 commits June 11, 2026 15:28

fix comment

c5ab59a

fix fmt

00e769e

fix fmt

c5d871b

fix fmt

faeb9cf

younies changed the title ~~currency: Make positive pattern variants optional and implement runtime fallback~~ currency: Implement runtime fallback for missing patterns and remove constructor validation Jun 15, 2026

younies requested a review from robertbastian June 15, 2026 13:56

robertbastian requested changes Jun 15, 2026

View reviewed changes

younies changed the title ~~currency: Implement runtime fallback for missing patterns and remove constructor validation~~ experimental/currency: Implement runtime fallback for missing patterns and remove constructor validation Jun 23, 2026

		// we have the fallback here for users feeding their own data without
		// handling the fallback logic in their data generation.

	// we have the fallback here for users feeding their own data without
	// handling the fallback logic in their data generation.
	// we have the fallback here in case the data struct is corrupted.

Conversation

younies commented Jun 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changelog

Uh oh!

Choose a reason for hiding this comment

Uh oh!

robertbastian Jun 11, 2026 • edited by younies Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

robertbastian left a comment

Choose a reason for hiding this comment

Uh oh!

younies commented Jun 15, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

younies Jun 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

younies commented Jun 11, 2026 •

edited

Loading

robertbastian Jun 11, 2026 •

edited by younies

Loading

younies Jun 23, 2026 •

edited

Loading