fix(parsing): remove shared scratch tables; parse fully in PHP (#247)

CHANGELOG.md

@@ -7,6 +7,20 @@ ones are marked like "v1.0.0-fork".
 
 ## [Unreleased]
 
+### Fixed
+
+* **Japanese (MeCab) parsing produced no tokens on Windows**: both MeCab output
+  readers split their lines on `PHP_EOL`, but a subprocess's line terminator
+  comes from MeCab, not from the host PHP runs on. Wherever the two disagree —
+  notably Windows, where `PHP_EOL` is `"\r\n"` while the output ends lines with
+  `"\n"` — the entire output stayed a single line and every token collapsed into
+  one, yielding zero usable tokens. Both
+  `JapaneseTextParser::buildTokensFromMecab()` and
+  `MecabParser::parseMecabOutput()` now normalize line endings before splitting,
+  the same way `SqlFileParser` does since #241. This is what failed the Windows
+  PHPUnit jobs. Verified end-to-end against MeCab 0.996 and covered by tests
+  that assert LF, CRLF and CR-only output tokenize identically.
+
 ### Changed
 
 * **Word-status model is now a single source of truth** (#238, Phase 1): the

db/migrations/20260702_120000_drop_persistent_scratch_tables.sql

@@ -0,0 +1,21 @@
+-- Migration: Drop the persistent parse/import scratch tables.
+--
+-- temp_word_occurrences, temp_words and tempexprs used to be persistent,
+-- globally-shared InnoDB tables that every parse/import TRUNCATEd and refilled.
+-- That caused two problems:
+--   * "Tablespace is missing for a table" (InnoDB error 194) crashes on
+--     TRUNCATE when the table's .ibd file went missing (notably on Windows) —
+--     every text import then failed. See issue #247.
+--   * Concurrent parses/imports corrupted each other's rows (no isolation).
+--
+-- They are now created per database connection as CREATE TEMPORARY TABLE at
+-- runtime (see src/Shared/Infrastructure/Database/ScratchTables.php), so the
+-- persistent versions are no longer needed. Dropping them here also repairs
+-- installs whose tablespace was already orphaned (the reported crash).
+--
+-- DROP TABLE IF EXISTS is idempotent and succeeds even when the tablespace is
+-- missing, so it clears the orphaned dictionary entry as well.
+
+DROP TABLE IF EXISTS temp_word_occurrences;
+DROP TABLE IF EXISTS temp_words;
+DROP TABLE IF EXISTS tempexprs;

db/schema/baseline.sql

@@ -112,23 +112,12 @@ CREATE TABLE IF NOT EXISTS word_occurrences (
 )
 ENGINE=InnoDB DEFAULT CHARSET=utf8;
 
-CREATE TABLE IF NOT EXISTS temp_word_occurrences (
-    TiCount smallint(5) unsigned NOT NULL,
-    TiSeID mediumint(8) unsigned NOT NULL,
-    TiOrder smallint(5) unsigned NOT NULL,
-    TiWordCount tinyint(3) unsigned NOT NULL,
-    TiText varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL
-) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;
-
-CREATE TABLE IF NOT EXISTS temp_words (
-    WoText varchar(250) DEFAULT NULL,
-    WoTextLC varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL,
-    WoTranslation varchar(500) NOT NULL DEFAULT '*',
-    WoRomanization varchar(100) DEFAULT NULL,
-    WoSentence varchar(1000) DEFAULT NULL,
-    WoTaglist varchar(255) DEFAULT NULL,
-    PRIMARY KEY(WoTextLC)
-) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;
+-- The parse/import scratch tables (temp_word_occurrences, temp_words, tempexprs)
+-- are no longer persistent shared tables. They are created per database
+-- connection as CREATE TEMPORARY TABLE at runtime (see
+-- src/Shared/Infrastructure/Database/ScratchTables.php). This avoids InnoDB
+-- "Tablespace is missing" (error 194) crashes on TRUNCATE and isolates
+-- concurrent parses/imports from each other. See issue #247.
 
 CREATE TABLE IF NOT EXISTS texts (
     TxID smallint(5) unsigned NOT NULL AUTO_INCREMENT,