edf2csv 0.5.94 → 0.5.96

package/CHANGELOG.md

@@ -3,6 +3,58 @@
 Notable changes to edf2csv. Versions follow [semantic versioning](https://semver.org); while the
 major version is 0, a minor bump may contain breaking changes.
 
+## 0.5.96
+
+### Fixed: api.md said `readAnnotations` returns three counts, then named four
+
+0.5.58 added `negativeDurations` to the list and left the number in front of it. That is the
+mistake 0.5.62 fixed one page over — "This code covers three conditions" against a list of five
+— on a sentence a reader checks against the list in the same breath.
+
+The pronoun after it had slipped the same way: "That last one is why `duration` being `null` is
+not by itself the same as the file giving no duration" was written when `unreadableDurations`
+was last in the sentence, and by then it named `negativeDurations`, which is not what the
+paragraph goes on to describe. It names the count now rather than pointing at a position.
+
+The number is counted from `readAnnotations`'s own return type, so the page has to agree with
+the function rather than with itself, and every count it returns has to be named — a number is
+only useful with the list.
+
+## 0.5.95
+
+### Fixed: api.md's cheap timing recipe mistimes every record after a gap, and says the conversion uses it
+
+The page argues, correctly and at length, that `index * recordDuration` is not where a record
+sits — a discontinuous file puts its records where its timekeeping annotations say, and a
+continuous one need not start at zero. It then offers `readOrigin()` as "the cheap version",
+shows
+
+```js
+const origin = (await file.readOrigin()) ?? 0;
+const recordStart = origin + (batch.firstRecordIndex + r) * recordDuration;
+```
+
+and closes: "That is what the conversion itself does, which is why its `time_s` and its
+`annotations.csv` agree."
+
+The conversion does that for `EDF+C`. For `EDF+D` it reads every record's own start time. On
+`discontinuous.edf`, whose records sit at 0, 1 and 10 seconds, the recipe puts the third at 2 —
+nine seconds from where the file says it is and from where `convert()` writes it. A reader who
+took the shortcut because the page said it was the same thing lines every record after a gap up
+against the wrong samples, and the sentence promising agreement with `annotations.csv` is
+exactly the promise it breaks.
+
+Scoped to continuous recordings now, with the check to make first and the arithmetic's answer on
+that fixture spelled out. The test runs both and requires them to differ, so the example cannot
+quietly stop being a counterexample.
+
+### Fixed: the same page still described the gain-of-zero rule 0.5.83 replaced
+
+"A gain of zero is different: the mapping is defined but flat, so `physicalMin` is returned and
+written normally." That is now true of one of the two ways to get a gain of zero. The other —
+a span too small to represent, which underflows — returns `NaN` and leaves the cells empty, and
+raises `UNUSABLE_PHYSICAL_RANGE`, which the sentence beside it did not list either.
+
 ## 0.5.94
 
 ### Fixed: two conversions with identical output disagreed about whether they were whole

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "edf2csv",
-  "version": "0.5.94",
+  "version": "0.5.96",
   "description": "Convert EDF, EDF+ and BDF biosignal recordings (European Data Format) to CSV from the command line. Local, streaming, and never resamples or alters units.",
   "keywords": [
     "edf",