SCENARIOS-oracle-splitter.md

Oracle Splitter Production Coverage

Phase 1: Ordinary SQL Boundaries

• Empty input returns no segments.
• Whitespace-only input returns no segments.
• Comment-only input returns no segments.
• A single semicolon-terminated SQL statement returns one segment without the delimiter.
• A single SQL statement without a semicolon returns one segment.
• Multiple semicolon-terminated SQL statements return separate segments.
• Mixed terminated and unterminated SQL statements return separate segments.
• Consecutive semicolons do not produce empty segments.
• Leading whitespace before SQL is preserved in segment text and range.
• Inter-statement whitespace is attached to the following segment unless it is consumed as a SQL*Plus line-command delimiter.
• Trailing whitespace after the final statement is preserved in the final segment.
• CRLF line endings do not corrupt boundaries.
• UTF-8 comments do not corrupt byte ranges.
• Arithmetic division does not act as a slash delimiter.
• Schema-qualified and dblink-qualified names do not affect splitting.

Phase 2: Literals, Identifiers, Comments, and Hints

• Semicolons inside single-quoted string literals are ignored.
• Slashes inside single-quoted string literals are ignored.
• Escaped single quotes using doubled quotes are handled.
• National character literals N'...' are handled.
• Oracle q'[ ... ]' quoting is handled.
• Oracle q'{ ... }' quoting is handled.
• Oracle q'( ... )' quoting is handled.
• Oracle q'< ... >' quoting is handled.
• Oracle custom q'!...!' quoting is handled.
• Semicolons inside double-quoted identifiers are ignored.
• Slashes inside double-quoted identifiers are ignored.
• Line comments containing semicolons are ignored.
• Line comments containing slash-only visual markers are ignored.
• Block comments containing semicolons are ignored.
• Block comments containing slash-only lines are ignored.
• Nested block comments are handled.
• Hints /*+ ... */ containing delimiters are ignored for splitting.

Phase 3: SQL*Plus Slash and Buffer Commands

• A slash alone on a line flushes the current SQL segment.
• A slash line with surrounding horizontal whitespace flushes the current SQL segment.
• A slash at EOF flushes the current SQL segment.
• A slash followed by non-whitespace on the same line is not a delimiter.
• A slash preceded by non-whitespace on the same line is not a delimiter.
• RUN alone on a line flushes the current SQL segment.
• R alone on a line flushes the current SQL segment.
• Slash commands with leading command lines before SQL do not produce empty segments.
• Multiple slash commands in a row do not produce empty segments.
• Slash commands after PL/SQL units do not appear in segment text.

Phase 4: Anonymous PL/SQL Blocks

• BEGIN ... END; is returned as one segment.
• DECLARE ... BEGIN ... END; is returned as one segment.
• Labeled anonymous blocks are returned as one segment.
• Nested anonymous blocks do not split at inner semicolons.
• Exception handlers do not split the outer block.
• IF ... END IF; stays inside the block.
• CASE ... END CASE; stays inside the block.
• LOOP ... END LOOP; stays inside the block.
• WHILE ... LOOP ... END LOOP; stays inside the block.
• FOR ... LOOP ... END LOOP; stays inside the block.
• FORALL statements stay inside the block.
• Dynamic SQL strings containing semicolons stay inside the block.

Phase 5: Stored PL/SQL Units

• CREATE PROCEDURE ... END; / is one segment.
• CREATE OR REPLACE PROCEDURE ... END; / is one segment.
• CREATE EDITIONABLE PROCEDURE ... END; / is one segment.
• CREATE NONEDITIONABLE PROCEDURE ... END; / is one segment.
• CREATE FUNCTION ... END; / is one segment.
• Function bodies with arithmetic division do not split at /.
• CREATE PACKAGE ... END; / is one segment.
• CREATE PACKAGE BODY ... END; / is one segment.
• Package body nested procedure bodies do not terminate the package.
• Package body nested function bodies do not terminate the package.
• CREATE TRIGGER ... BEGIN ... END; / is one segment.
• Compound triggers are one segment.
• CREATE TYPE BODY ... END; / is one segment.
• Type body member function bodies do not terminate the type body.
• Stored units followed by ordinary SQL split correctly.

Phase 6: SQL*Plus Line Commands

• SET command lines are returned as line segments.
• SHOW command lines are returned as line segments.
• PROMPT command lines are returned as line segments, including embedded semicolons.
• SPOOL command lines are returned as line segments.
• COLUMN command lines are returned as line segments.
• BREAK command lines are returned as line segments.
• COMPUTE command lines are returned as line segments.
• TTITLE and BTITLE command lines are returned as line segments.
• REPHEADER and REPFOOTER command lines are returned as line segments.
• DEFINE and UNDEFINE command lines are returned as line segments.
• ACCEPT command lines are returned as line segments.
• VARIABLE and PRINT command lines are returned as line segments.
• EXECUTE command lines are returned as line segments.
• CONNECT and DISCONNECT command lines are returned as line segments.
• EXIT and QUIT command lines are returned as line segments.
• WHENEVER SQLERROR command lines are returned as line segments.
• WHENEVER OSERROR command lines are returned as line segments.
• @, @@, and START script invocation lines are returned as line segments.
• REM and REMARK command lines are returned as line segments.
• HOST and ! shell command lines are returned as line segments.
• Buffer manipulation commands LIST, APPEND, CHANGE, DEL, INPUT, SAVE, GET, and EDIT are returned as line segments.
• SQL*Plus command words inside SQL are not treated as line commands.
• SQL*Plus command words inside PL/SQL are not treated as line commands.

Phase 7: Soft-Fail and Safety

• Unterminated single-quoted strings do not panic.
• Unterminated q-quotes do not panic.
• Unterminated double-quoted identifiers do not panic.
• Unterminated block comments do not panic.
• Missing PL/SQL END returns a best-effort trailing segment.
• Truncated CREATE PROCEDURE returns a best-effort trailing segment.
• Binary-ish bytes do not panic.
• Every returned segment has a valid non-overlapping byte range.
• Every returned segment text equals input[ByteStart:ByteEnd].
• Reconstructing returned ranges includes SQL*Plus command lines as their own segments.

Phase 8: Facade Integration

• oracle.Parse parses splitter output from ordinary SQL scripts.
• oracle.Parse parses splitter output from slash-terminated PL/SQL scripts.
• oracle.Parse returns parser errors for SQL*Plus command segments instead of silently skipping them.
• oracle.Parse preserves original byte offsets in AST locations.
• oracle.Parse reports start line and column based on the original script.
• oracle.Parse handles multiple ordinary SQL statements around slash-delimited PL/SQL blocks.