Appendix 1: Command Line Options

Appendix 1: Command Line Options

CQL has a variety of command line (CLI) options but many of them are only interesting for cql development. Nonetheless this is a comprehensive list:

NOTE: CQL is often used after the c pre-processor is run so this kind of invocation is typical:

cc -E -x c foo.sql | cql [args]

With No Options

  • emits a usage message

–in file

  • reads the given file for the input instead of stdin
  • the input should probably have already been run through the C pre-processor as above
  • returns non-zero if the file fails to parse

Example:

cql --in test.sql

–sem

  • performs semantic analysis on the input file ONLY
  • the return code is zero if there are no errors

Example:

cql --in sem_test.sql --sem

–ast

  • walks the AST and prints it to stdout in human readable text form
  • may be combined with –sem (semantic info will be included) Example
cql --in sem_test.sql --sem --ast >sem_ast.out

–echo

  • walks the AST and emits the text of a program that would create it
  • this has the effect of “beautifying” badly formatted input or “canonicalizing” it
    • some sensible indenting is added but it might not be the original indenting
    • extraneous whitespace, parens, etc. are removed
  • may be combined with –sem (in which case you see the source after any rewrites for sugar)
  • this also validates that the input can be parsed

Example

cql --in test.sql --echo >test.out  # test.out is "equivalent" to test.sql

–dot

  • prints the internal AST to stdout in DOT format for graph visualization
  • this is really only interesting for small graphs for discussion as it rapidly gets insane

Example:

cql --dot --in dottest.sql

–cg output1 output2 …

  • any number of output files may be needed for a particular result type, two is common
  • the return code is zero if there are no errors
  • any –cg option implies –sem

Example:

cql --in foo.sql --cg foo.h foo.c

–nolines

  • Suppress the # directives for lines. Useful if you need to debug the C code.

Example:

cql --in test.sql --nolines --cg foo.h foo.c

–global_proc name

  • any loose SQL statements not in a stored proc are gathered and put into a procedure of the given name
  • when generating a schema migrate script the global proc name is used as a prefix on all of the artifacts so that there can be several independent migrations linked into a single executable

–compress

  • for use with the C result type, (or any similar types added to the runtime array in your compiler)
  • string literals for the SQL are broken into “fragments” the DML is then represented by an array of fragments
  • since DML is often very similar there is a lot of token sharing possible
  • the original string is recreated at runtime from the fragments and then executed
  • comments show the original string inline for easier debugging and searching

NOTE: different result types require a different number of output files with different meanings

–test

  • some of the output types can include extra diagnostics if --test is included
  • the test output often makes the outputs badly formed so this is generally good for humans only

–dev

  • some codegen features only make sense during development, this enables dev mode to turn those one ** example: explain query plan

–exp

  • runs macro expansion without semantic analysis
  • combined with --echo this is kind of like cc -E it can give you a view of what happened after the pre-processing
  • combined with --ast it is a useful test tool (its primary function)

–c_include_namespace

  • for the C codegen runtimes, it determines the header namespace (as in #include “namespace/file.h”) that goes into the output C file
  • if this option is used, it is prefixed to the first argment to –cg to form the include path in the C file
  • if absent there is no “namespace/” prefix

–c_include_path

  • for the C codegen runtimes, it determines the full header path (as in #include “your_arg”) that goes into the output C file
  • if this option is used, the first argment to –cg controls only the output path and does not appear in include path at all
  • this form overrides –c_include_namespace if both are specified

–objc_c_include_path

  • for ObjC codegen runtimes that need to refer to the generated C code, this represents the header of the C generated code that will be used during inclusion from the ObjC file

Result Types (–rt *)

These are the various outputs the compiler can produce.

–rt c

  • requires two output files (foo.h and foo.c)
  • this is the standard C compilation of the sql file
–cqlrt foo.h
  • emits #include "foo.h" into the C output instead of #include "cqlrt.h"
–generate_type_getters
  • changes C output for CQL result sets so that the field readers used shared functions to get fields of a certain type
  • this style of codegen makes result-sets more interoperable with each other if they have similar shape so it can be useful
–generate_exports
  • adds an additional output file
  • example: `–in foo.sql –generate_exports –rt c –cg foo.h foo.c foo_exports.sql
  • the output file foo_exports.sql includes procedure declarations for the contents of foo.sql
  • basically automatically generates the CQL header file you need to access the procedures in the input from some other file
  • if it were C it would be like auto-generating foo.h from foo.c

–rt lua

  • requires one output file (foo.lua)
  • this is the standard Lua compilation of the sql file

–rt objc

  • objective C wrappers for result sets produced by the stored procedures in the input
  • these depend on the output of a standard codegen run so this is additive
  • requires one output file (foo.h)

–rt schema

  • produces the canonical schema for the given input files
  • stored procedures etc. are removed
  • whitespace etc. is removed
  • suitable for use to create the next or first “previous” schema for schema validation
  • requires one output file

–rt schema_upgrade

  • produces a CQL schema upgrade script which can then be compiled with CQL itself
  • see the chapter on schema upgrade/migration: Chapter 10
  • requires one output file (foo.sql)
–include_regions a b c
  • the indicated regions will be declared
  • used with --rt schema_upgrade or --rt schema
  • in the upgrade case excluded regions will not be themselves upgraded, but may be referred, to by things that are being upgraded
–exclude_regions x y z
  • the indicated regions will still be declared but the upgrade code will be suppressed, the presumption being that a different script already upgrades x y z
  • used with --rt schema_upgrade
–min_schema_version n
  • the schema upgrade script will not include upgrade steps for schema older than the version specified
–schema_exclusive
  • the schema upgrade script assumes it owns all the schema in the database, it aggressively removes other things

–rt json_schema

  • produces JSON output suitable for consumption by downstream codegen
  • the JSON includes a definition of the various entities in the input
  • see the section on JSON output for details

–rt query_plan

  • produces CQL output which can be re-compiled by CQL as normal input
  • the output consists of a set of procedures that will emit all query plans for the DML that was in the input
  • see Chapter 15

–rt stats

  • produces a simple .csv file with node count information for AST nodes per procedure in the input
  • requires one output file (foo.csv)