@cosyte/terminology 0.1.5 → 0.1.7

package/CHANGELOG.md

@@ -1,5 +1,49 @@
 # Changelog
 
+## 0.1.7
+
+### Patch Changes
+
+- c688d91: Faster value set membership and translation, and leaner CSV, GEM and UCUM reading. Every answer,
+  loaded model, warning and error is unchanged, and no signature changes. One thing is deliberately
+  different: within one `loadGems` result, the entries that carry the same five flag digits now share
+  one frozen `flags` object rather than each holding its own. Every field reads the same; only
+  `entryA.flags === entryB.flags` is now `true` for two such entries. If you told entries apart by
+  the identity of their `flags`, compare the entries themselves, or their fields, instead.
+
+  - `validateCodeInValueSet` builds a loaded release's hierarchy index on the first membership
+    question over a filter and reuses it, rather than walking the whole release for every code. A code
+    system whose `concepts` you assembled by hand is still indexed afresh on every call, so a change
+    you make to it between two calls is read.
+  - `translate` finds a loaded ConceptMap's elements through an index of each group by source code,
+    built on the first translation through the group, rather than reading every element for each code.
+    A map you assembled by hand is still read element by element.
+  - `loadCodeSystem` reads a CSV release row by row and builds only the configured columns, so a long
+    column you did not configure costs no memory. The strings a loaded model keeps are copied out of
+    the release text, so the model does not keep that text alive after the load; the same holds for
+    `parseCsv` and `loadGems`.
+  - `loadGems` reads each line in place instead of splitting the file into lines and each line into
+    fields, and decodes each distinct flag field once per load (the shared `flags` object above).
+  - `validateUcum`, `ucumEqual` and `parseUcum` match a unit against the atoms and prefixes that start
+    with the same character, from an index of the unit table built once, rather than against the whole
+    table.
+
+  The indexes are kept with the loaded release or map and go when it does. A source coding with no
+  `code` is still matched against a loaded map's elements that have none, as before.
+
+  The package's npm keywords also gain `snomed`, `icd10` and `cpt`, so a search on those terms finds
+  it.
+
+## 0.1.6
+
+### Patch Changes
+
+- 43eea2f: An internal tidy with no change to any result, warning, or diagnostic. The value set diagnostic factories now take the path every caller already gives them, a few expressions became statements so a coverage note can name the one arm no input reaches, and those arms now carry that note.
+- 43eea2f: A CSV source whose `columns.properties` list has a hole now fails with the typed `TERM_CODESYSTEM_MALFORMED` fatal for a missing property column, the same as an explicit `undefined` entry, instead of a `TypeError` from inside the loader.
+- 0cb5b88: Doc comments, which your editor shows on hover and the generated API reference prints, now match the code. `resolveSystem` lists the forms it accepts in the order it tries them, and no longer says it accepts a `v2-0203` mnemonic on its own (it does not; pass the table's OID or URI). `SYSTEM_IDENTITIES` and `resolveSystem` no longer say every accepted mnemonic comes from HL7 Table 0396: aliases such as `LOINC` and `CPT` are accepted too but are not Table 0396 codes. `NdcStatus` says that `loadRxNormGraph` reads no NDC history, so every NDC it resolves is `active`. `RxNormGraph.concepts` says a concept is typed by its first defining atom. The complex-map outcome types and `MAP_CATEGORIES.AMBIGUOUS` no longer say an ambiguous or context-dependent category stops a group on its own: the group's rules decide. Links that printed as a source path, such as `../crosswalk/gems.invertGem`, now link the symbol. No code changes.
+- e16570e: An RRF or fixed-width source whose property list has an entry that is not an object (a hole, `undefined`, or `null`) now fails with the typed `TERM_CODESYSTEM_MALFORMED` fatal, as a CSV source with a hole in its property columns does, instead of a `TypeError` from inside the loader.
+- 43eea2f: `validateCodeInValueSet` no longer admits a code through a `compose` component that selects nothing. A component with no `system`, `concept`, `filter`, or referenced value set (the loader accepts one with an empty `valueSet` list) matched every code, so an include admitted any code and an exclude removed every code, while `expand` gave the same component no members. Membership now agrees with `expand`: such a component matches no code, and a `valueSet` list that yields no reference is an intersection with nothing.
+
 ## 0.1.5
 
 ### Patch Changes

package/README.md

@@ -379,6 +379,9 @@ Questions, bug reports, and an answer the engine got wrong go in
 system, map, or value set that reproduces it. Use synthetic data only: a code in patient context can
 be PHI. [SUPPORT.md](https://github.com/cosyte/.github/blob/main/SUPPORT.md) says where to ask what.
 
+Need a terminology integration built, or hosted and run for you? See our
+[integration services](https://cosyte.com/integration-services).
+
 ## Security
 
 Report a vulnerability privately, never in a public issue, and with no real patient data in it: