libsql-sqlite3/ext/jni/src/org/sqlite/jni/test-script-interpreter.md
The purpose of the Test Script Interpreter is to read and interpret script files that contain SQL commands and desired results. The interpreter will check results and report an discrepencies found.
The test script files are ASCII text files. The filename always ends with ".test". Each script is evaluated independently; context does not carry forward from one script to the next. So, for example, the --null command run in one test script does not cause any changes in the behavior of subsequent test scripts. All open database connections are closed at the end of each test script. All database files created by a test script are deleted when the script finishes.
The test script is read line by line, where a line is a sequence of characters that runs up to the next '\n' (0x0a) character or until the end of the file. There is never a need to read ahead past the end of the current line.
If any line contains the string " MODULE_NAME:" (with a space before the initial "M") or "MIXED_MODULE_NAME:" then that test script is incompatible with this spec. Processing of the test script should end immediately. There is no need to read any more of the file. In verbose mode, the interpreter might choose to emit an informational messages saying that the test script was abandoned due to an incompatible module type.
If any line contains the string "SCRIPT_MODULE_NAME:" then the input script is known to be of the correct type for this specification and processing may continue. The "MODULE_NAME" checking in steps 2 and 3 may optionally be discontinued after sighting a "SCRIPT_MODULE_NAME".
If any line contains "REQUIRED_PROPERTIES:" and that substring is followed by any non-whitespace text, then the script is not compatible with this spec. Processing should stop immediately. In verbose mode, the interpreter might choose to emit an information message saying that the test script was abandoned due to unsupported requirement properties.
If any line begins with the "|" (0x7c) character, that indicates that the input script is not compatible with this specification. Processing of the script should stop immediately. In verbose mode, the interpreter might choose to emit an informational message indicating that the test script was abandoned because it contained "a dbtotxt format database specification".
Any line that begins with "#" is a C-preprocessor line. The interpreter described by this spec does not know how to deal with C-preprocessor lines. Hence, processing should be abandoned. In verbose mode, the interpreter might emit an informational message similar to "script NAME abandoned due to C-preprocessor line: ..."
If a line begins with exactly two minus signs followed by a lowercase letter, that is a command. Process commands as described below.
All other lines should be accumulated into the "input buffer". The various commands will have access to this input buffer. Some commands will reset the buffer.
The initial state of the interpreter at the start of processing each script is as if the following command sequence had been run:
--close all --db 0 --new test.db --null nil
In words, all database connections are closed except for connection 0 (the
default) which is open on an empty database named "test.db". The string
"nil" is displayed for NULL column values.
The only context carried forward after the evaluation of one test script
into the evaluation of the next test script is the count of the number of
tests run and the number of failures seen.
## Commands:
Each command looks like an SQL comment. The command begins at the left
margin (no leading space) and starts with exactly 2 minus signs ("-").
The command name consists of lowercase letters and maybe a "-" or two.
Some commands have arguments.
The arguments are separated from the command name by one or more spaces.
Commands have access to the input buffer and might reset the input buffer.
The command can also optionally read (and consume) additional text from
script that comes after the command.
Unknown or unrecognized commands indicate that the script contains features
that are not (yet) supported by this specification. Processing of the
script should terminate immediately. When this happens and when the
interpreter is in a "verbose" mode, the interpreter might choose to emit
an informational message along the lines of "test script NAME abandoned
due to unsupported command: --whatever".
The initial implemention will only recognize a few commands. Other
commands may be added later. The following is the initial set of
commands:
### The --testcase command
Every test case starts with a --testcase command. The --testcase
command resets both the "input buffer" and the "result buffer". The
argument to the --testcase command is the name of the test case. That
test case name is used for logging and debugging and when printing
errors. The input buffer is set to the body of the test case.
### The --result command
The --result command tries to execute the text in the input buffer as SQL.
For each row of result coming out of this SQL, the text of that result is
appended to the "result buffer". If a result row contains multiple columns,
the columns are processed from left to right. For each column, text is
appended to the result buffer according to the following rules:
* If the result buffer already contains some text, append a space.
(In this way, all column values and all row values are separated from
each other by a single space.)
* If sqlite3_column_text() returns NULL, then append "nil" - or
some other text that is specified by the --null command - and skip
all subsequent rules.
* If sqlite3_column_text() is an empty string, append `{}` to the
result buffer and skip all subsequent rules.
* If sqlite3_column_text() does not contain any special
characters, append it to the result buffer without any
formatting and skip all subsequent rules. Special characters are:
0x00 to 0x20 (inclusive), double-quote (0x22), backslash (0x5c),
curly braces (0x7b and 0x7d).
* If sqlite3_column_text() does not contains curly braces, then put
the text inside of `{...}` and append it and skip all subsequent rules.
* Append the text within double-quotes (`"..."`) and within the text
escape '"' and '\\' by prepending a single '\\' and escape any
control characters (characters less than 0x20) using octal notation:
'\\NNN'.
If an error is encountered while running the SQL, then append the
symbolic C-preprocessor name for the error
code (ex: "SQLITE_CONSTRAINT") as if it were a column value. Then append
the error message text as if it where a column value. Then stop processing.
After the SQL text has been run, compare the content of the result buffer
against the argument to the --result command and report a testing error if
there are any differences.
The --result command resets the input buffer, but it does not reset
the result buffer. This distinction does not matter for the --result
command itself, but it is important for related commands like --glob
and --notglob. Sometimes test cases will contains a bunch of SQL
followed by multiple --glob and/or --notglob statements. All of the
globs should be evaluted agains the result buffer correct, but the SQL
should only be run once. This is accomplished by resetting the input
buffer but not the result buffer.
### The --glob command
The --glob command works just like --result except that the argument to
--glob is interpreted as a TEST-GLOB pattern and the results are compared
using that glob pattern rather than using strcmp(). Other than that,
the two operate the same.
The TEST-GLOB pattern is slightly different for a standard GLOB:
* The '*' character matches zero or more characters.
* The '?' character matches any single character
* The '[...]' character sequence machines a single character
in between the brackets.
* The '#' character matches one or more digits (This is the main
difference between standard unix-glob and TEST-GLOB. unix-glob
does not have this feature. It was added to because it comes
up a lot during SQLite testing.)
### The --notglob command
The --notglob command works just like --glob except that it reports an
error if the GLOB does match, rather than if the GLOB does not matches.
### The --oom command
This command is to be used for out-of-memory testing. It means that
OOM errors should be simulated to ensure that SQLite is able to deal with
them. This command can be silently ignored for now. We might add support
for this later.
### The --tableresult command
The --tableresult command works like --glob except that the GLOB pattern
to be matched is taken from subsequent lines of the input script up to
the next --end. Every span of one or more whitespace characters in this
pattern text is collapsed into a single space (0x20).
Leading and trailing whitespace are removed from the pattern.
The --end that ends the GLOB pattern is not part of the GLOB pattern, but
the --end is consumed from the script input.
### The --new and --open commands
The --new and --open commands cause a database file to be opened.
The name of the file is the argument to the command. The --new command
opens an initially empty database (it deletes the file before opening it)
whereas the --open command opens an existing database if it already
exists.
### The --db command
The script interpreter can have up to 7 different SQLite database
connections open at a time. The --db command is used to switch between
them. The argument to --db is an integer between 0 and 6 that selects
which database connection to use moving forward.
### The --close command
The --close command causes an existing database connection to close.
This command is a no-op if the database connection is not currently
open. There can be up to 7 different database connections, numbered 0
through 6. The number of the database connection to close is an
argument to the --close command, which will fail if an out-of-range
value is provided. Or if the argument to --close is "all" then all
open database connections are closed. If passed no argument, the
currently-active database is assumed.
### The --null command
The NULL command changes the text that is used to represent SQL NULL
values in the result buffer.
### The --run command
The --run command executes text in the input buffer as if it where SQL.
However, nothing is added to the result buffer. Any output from the SQL
is silently ignored. Errors in the SQL are silently ignored.
The --run command normally executes the SQL in the current database
connection. However, if --run has an argument that is an integer between
0 and 6 then the SQL is run in the alternative database connection specified
by that argument.
### The --json and --json-block commands
The --json and --json-block commands work like --result and --tableresult,
respectively. The difference is that column values are appended to the
result buffer literally, without ever enclosing the values in `{...}` or
`"..."` and without escaping any characters in the column value and comparison
is always an exact strcmp() not a GLOB.
### The --print command
The --print command emits both its arguments and its body (if any) to
stdout, indenting each line of output.
### The --column-names command
The --column-names command requires 0 or 1 as an argument, to disable
resp. enable it, and modifies SQL execution to include column names
in output. When this option is on, each column value emitted gets
prefixed by its column name, with a single space between them.