gudusoft.gsqlparser.sqlenv.unicode

Class GspCaseData

Object

gudusoft.gsqlparser.sqlenv.unicode.GspCaseData

public final class GspCaseData
extends Object

GSP-pinned Unicode case operations — the JVM-independent fold engine of the persistent canonical-identity program (docs/designs/persistent-canonical-identity-implementation-plan.md §3).

All data comes from the committed binary resource gsp-case-data.bin, generated from vendored UCD "16.0.0" files. Nothing here consults String.toUpperCase/toLowerCase or any other JVM Unicode table, so the results are identical on every JVM — the whole point: JDK 8 ships Unicode 6.2 while JDK 21 ships 15.x, and identifier identity must not depend on which one happens to run the export.

The operations, deliberately nothing else:

• toUpperFull(java.lang.String)/toLowerFull(java.lang.String) — FULL case mappings (length may change: ß→SS), plus Final_Sigma as the single context rule, implemented per the UCD definition (NOT any JDK's quirk — the JDKs disagree with each other on Case_Ignorable contexts, so "match the JVM" is not even well-defined). Locale-conditional rules (tr/az/lt) are excluded by design: no database folds identifiers by session locale, and including them corrupts catalog identity.
• simpleFoldCollapse(java.lang.String) — CaseFolding.txt C+S simple fold, code-point-wise. Successor of the char-wise caseCollapse (fixes supplementary planes).
• toUpperSimple(java.lang.String)/toLowerSimple(java.lang.String) — SIMPLE case mappings (UnicodeData.txt fields 12/13), one code point to one, no context: the mappings Oracle (unquoted names) and MySQL (table, column and routine names) actually apply. Derived from the same resource plus a small checked override table, so they add no resource and change none of the operations above.
• toLowerAscii(java.lang.String) and clipUtf8Bytes(java.lang.String, int) — PostgreSQL's identifier downcasing in a UTF8 database and its NAMEDATALEN truncation; no Unicode data.

Unpaired surrogates pass through unchanged in every operation — identity must be total on arbitrary UTF-16, and substitution would merge distinct malformed inputs. Inputs the operation does not change are returned as the SAME instance, including surrogate-bearing ones (zero allocation).

Thread-safe: one immutable table holder, safely published via the class initializer, whose body is straight-line code delegating to ordinary methods. (Historical note: this class's first obfuscated build failed verification because proguard.cfg once disabled preverification while emitting v52 class files — that affected EVERY branchy method product-wide and was fixed by re-enabling preverification; the branch-free initializer is retained as cheap hygiene.) Deployment notes: the resource is looked up relative to this class, so package RELOCATION (consumer-side shading) relocates the lookup with it, but the relocator must also move the resource directory; GraalVM native-image users must register gudusoft/gsqlparser/sqlenv/unicode/gsp-case-data.bin as a resource. Missing/corrupt resource fails fast with ExceptionInInitializerError on first use.

Field Summary

Fields

Modifier and Type	Field and Description
static int	ALGORITHM_REV Bumped on ANY behavioral change of these operations.
static String	UNICODE_VERSION Exact UCD version the tables are generated from.

Method Summary

Modifier and Type	Method and Description
static String	clipUtf8Bytes(String s, int maxBytes) The longest prefix of s whose UTF-8 encoding has at most maxBytes bytes, cut on a whole code point (PostgreSQL's pg_mbcliplen in a UTF8 database).
static String	simpleFoldCollapse(String s) CaseFolding C+S simple fold applied per code point — the canonical-equality collapse for INSENSITIVE + no-fold cells.
static String	toLowerAscii(String s) ASCII-only lowercase: A..Z become a..z, every other code unit is kept.
static String	toLowerFull(String s) Full lowercase mapping with the UCD Final_Sigma rule: Σ lowers to ς iff preceded by a Cased code point (skipping Case_Ignorable) and NOT followed by one (skipping Case_Ignorable).
static String	toLowerSimple(String s) Simple lowercase mapping, code point by code point: each code point becomes its Simple_Lowercase_Mapping (UnicodeData.txt field 13) or stays itself.
static String	toUpperFull(String s) Full uppercase mapping (unconditional UCD data; length may grow, e.g.
static String	toUpperSimple(String s) Simple uppercase mapping, code point by code point: each code point becomes its Simple_Uppercase_Mapping (UnicodeData.txt field 12) or stays itself.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

UNICODE_VERSION

public static final String UNICODE_VERSION

Exact UCD version the tables are generated from. Part of every policyId.

See Also:

Constant Field Values

ALGORITHM_REV

public static final int ALGORITHM_REV

Bumped on ANY behavioral change of these operations. Part of every policyId.

See Also:

Constant Field Values

Method Detail

toUpperFull

public static String toUpperFull(String s)

Full uppercase mapping (unconditional UCD data; length may grow, e.g. ß→SS).

toLowerFull

public static String toLowerFull(String s)

Full lowercase mapping with the UCD Final_Sigma rule: Σ lowers to ς iff preceded by a Cased code point (skipping Case_Ignorable) and NOT followed by one (skipping Case_Ignorable). Everything else uses unconditional data.

simpleFoldCollapse

public static String simpleFoldCollapse(String s)

CaseFolding C+S simple fold applied per code point — the canonical-equality collapse for INSENSITIVE + no-fold cells. Code-point-wise, so supplementary-plane pairs fold correctly (unlike the char-wise legacy collapse).

toUpperSimple

public static String toUpperSimple(String s)

Simple uppercase mapping, code point by code point: each code point becomes its Simple_Uppercase_Mapping (UnicodeData.txt field 12) or stays itself. One code point in, one code point out, no context, no locale (ß is kept, unlike toUpperFull(java.lang.String)).

This is the per-character mapping Oracle applies to an unquoted name (straße is stored as STRAßE) and MySQL applies when it compares routine names (ı and i are one function). It is the SQL-LR UPPER_SIMPLE operation (lineage-spec/identifier-rules/FORMAT.md section 6).

toLowerSimple

public static String toLowerSimple(String s)

Simple lowercase mapping, code point by code point: each code point becomes its Simple_Lowercase_Mapping (UnicodeData.txt field 13) or stays itself. No Final_Sigma context (Σ always becomes σ), and İ becomes i.

This is MySQL's one-to-one lowercase mapping for table names under lower_case_table_names 1 and 2 and for column names. It is the SQL-LR LOWER_SIMPLE operation.

toLowerAscii

public static String toLowerAscii(String s)

ASCII-only lowercase: A..Z become a..z, every other code unit is kept. PostgreSQL's downcase_identifier in a multibyte server encoding (UTF8): ÄBC is stored as Äbc. Consults no Unicode data. Returns the same instance when nothing changes.

clipUtf8Bytes

public static String clipUtf8Bytes(String s,
                                   int maxBytes)

The longest prefix of s whose UTF-8 encoding has at most maxBytes bytes, cut on a whole code point (PostgreSQL's pg_mbcliplen in a UTF8 database). Returns the same instance when it already fits. An unpaired surrogate counts as three bytes (its CESU form): it never occurs in a name a server stored, and counting it keeps the operation total on arbitrary UTF-16.

Parameters:

s - the text; null returns null

maxBytes - the byte limit; zero or negative means no limit