Flexi Config Reader is a configuration language that employs a template-reference syntax in order to express repetitive & structured configuration parameters in a more concise manner. Usage of these constructs is optional. At its core, the syntax uses concepts similar to those found in json & toml: nested structures of key/value pairs. One can draw parallels between some of the features found in this syntax and those made available by the jinja templating python library.
The syntax has three core concepts:
struct
- This is a concrete entry of data. A struct
may contain other struct
s as
well as key/value pairs.proto
- The proto
keyword is used to define a template, which can be used to create
a struct
. A proto
may contain other struct
s, reference
s (see below), as well as
key/value pairs. The values in a proto (and all sub-objects) may be represented by a variable.reference
- The reference
keyword is used to turn a proto
into a concrete
struct
. The proto
is reference
d, given a name, and all contained variables are given a value.In summary, a proto
that is reference
d, effectively becomes a struct
.
include
syntax - existing configuration files may be included by other files. There
are no restrictions around a proto
being defined in one file, then reference
d in
another file. This highlights one of the advantages of this syntax: the ability to
define a generic set of templates in one file, which will be used to produce multiple,
repeated concrete structs with different names and parameters in a different file.
The same config files may not be included twiceinclude_relative
syntax - Same as include
, except nested paths will resolve from the
included file's path instead of from the base file's path. See examples/config_example12.cfg
for an example of this keyword. The list of include_relative
statements must come after all include
statements in a config file.[optional]
and [once]
include [once] base.cfg
allows including the same file twice, which may be useful if some base values or
structure is defined in a separate file that may be transitively included from multiple places.
By default, the parser will fail when it encounters a duplicate include, this leads to duplicate key definitions.include [optional] file_may_not_exist.cfg
enables a config to reference files that may be missing.
By default, the parser will fail when it encounters a missing file.include
or include_relative
statements
using the syntax ${ENV_VAR_NAME}
. If the environment variable is not set, the reference
will be replaced with an empty string.proto
defines a templated struct
, one can add additional keys
to the resulting struct
when the proto
is referenced.[override]
- By default, a key can be specified once and only once in the config file. Using
the [override]
keyword allows a value that was previously specified in the config to be
overridden with a new value.While whitespace (except for following the struct
, proto
, reference
, and as
keywords)
is not significant it is convenient in order to better view the structure of the configuration parameters.
struct
keywordThe struct
keyword is used to define a structure of data, e.g.
struct foo {
key1 = 0
key2 = 1.4
key3 = "string"
struct bar {
nested_key = 0
baz = 3.14
}
}
As mentioned above, a struct
may contain key/value pairs as well as other struct
, proto
, or reference
blocks.
When using the struct
syntax, fully-qualified (or dotted) keys are not allowed within the struct. The above example
would result in the following fully-qualified key/value pairs:
foo.key1 = 0
foo.key2 = 1.4
foo.key3 = "string"
foo.bar.nested_key = 0
foo.bar.baz = 3.14
proto
keywordThe proto
keyword is used to define a templated definition of a struct
, e.g.
struct protos {
proto foo {
key1 = 0
key2 = 1.4
key3 = $KEY3
struct bar {
key_a = 0x1A2B
key_b = $KEY_B
}
}
}
The above example defines a proto
called "foo" with two concrete keys and a variable key key3
. There is
also a nested struct (which also contains a contrete key and a variable key key_b
).
This proto
does not add anything to the configuration parameters as is. It must be reference
d in order to
turn it into a concrete struct
of data.
Keys within a proto
(or any struct
nested within a proto
) may be assigned a "variable" which will be resolved later as explained below. A variable begins with a $
and is followed by a variable name matching [A-Z0-9_]+
. The variable name may optionally be enclosed in a set of curly braces (e.g. {...}
).
Variables may also be used within strings (e.g. "foo.$BAR.baz"
), within a key-value reference, in a list (e.g. [0, ${ONE}, 2]
) or in an expression (e.g. {{ 0.5 * ($MIN + ${MAX}) }}
)
reference
keywordThe reference
keyword is used to turn a proto
into a struct
. Following on from the example above:
reference protos.foo as fuzz {
$KEY3 = "apple"
$KEY_B = 3.14159
+extra_key = 0x7FF
}
The above example introduces the following concepts:
proto
, the fully qualified name of the proto must be used (i.e. protos.foo
instead of foo
).$KEY3
and $KEY_B
that were introduced in protos.foo
and protos.foo.bar
respectively are given values when the proto
is reference
d. These variables can be used in the following situations as explained above
key = "refer.to.$KEY3.thing"
)key = $(reference.$THIS.var)
)+extra_key
is an appended key. The proto
"protos.foo" does not contain this key, but the resulting struct fuzz
will contain this additional key.NOTE: There is a special variable $PARENT_NAME
which can be used to assign the value of the containing parent key to a variable defined in a proto (e.g. we could write $KEY3 = $PARENT_NAME
which would assign the value "fuzz"
to $KEY3
).
The above reference
could be defined as follows using the struct
syntax:
struct fuzz {
key1 = 0
key2 = 1.4
key3 = "apple"
extra_key = 0x7FF
struct bar {
key_a = 0x1A2B
key_b = 3.14159
}
}
NOTE: A proto
with no variables may be referenced using empty curly braces, e.g. The following is valid syntax:
proto foo {
bar = 1
baz = "test"
}
reference foo as copy {}
This construct is useful when you want to fully define a portion of the config and then reference (i.e. copy) it in other places within the config file.
As mentioned above, existing values may be referenced by their key in order to define a different key/value pair. E.g.
struct foo {
key1 = 0
key2 = 1.4
}
struct bar {
key1 = 1
key2 = $(foo.key2)
}
In this case, the key bar.key2
is given the value of foo.key2
, or in this case 1.4
. This construct can be
used within the struct
, proto
or reference
constructs. The syntax is $(path.to.key)
. As mentioned in the
reference
section, a key-value reference may include variables internally (e.g. $(path.to.$KEY)
.
The following value types are supported:
true
or false
)$VAR
), key-value references and/or expressions, but when resolved, the list elements must be homogeneous.The config syntax also supports the ability to express mathematical expressions using numbers and key-value references. When using key-value references within a mathematical expression, the value must evaluate to a numeric value (chained references are allowed as long as the end of the chain is numeric). Mathematical expressions are enclosed within a set of opening and closing curly braces (i.e. {{
and }}
). The keyword pi
may also be used to represent the value of pi more precisely.
The following operators are supported:
+
- plus (both binary and unary)-
- minus (both binary and unary)*
- multiply/
- divide^
or **
- powerpi
- the value of pi(
and )
- Parentheses grouping operationsThe following example:
struct foo {
key1 = 1.25
key2 = -2
val = {{ $(foo.key1) * $(foo.key2) + 2*pi }}
}
struct bar {
key1 = 1e-2
key2 = {{ $(foo.val) * $(bar.key1) ^ 0.5 + 10 }}
}
will evaluate to:
struct foo {
key1 = 1.25
key2 = -2
val = 3.78318
}
struct bar {
key1 = 1e-2
key2 = 10.378318
}
[override]
keywordAs mentioned above, a leaf key can be specified once and only once in the config file (e.g. foo.bar
and baz.bar
are unique keys). Using the [override]
keyword allows the value of a key
that was previously specified in the config to be overridden with a new value. In order to use
[override]
the key/value pair must exist elsewhere in the config (using [override]
on a key not previously defined will result in an error). In addition, the type of the override must match the previously defined value.
An example of usage can be found in example_config1.cfg and is shown below:
struct foo {
bar = 0
baz = 2
buzz = "string"
}
a = $(baz)
b = {{ $(baz) }}
c = $(a)
...
struct foo {
baz [override] = -7
buzz [override] = 13 # <-- This results in an error due to the type change
}
With the above example, the fully resolved config will be (ignoring the error highlighted above):
struct foo {
bar = 0
baz = -7
buzz = "string"
}
a = -7
b = -7
c = -7
Note that the [override]
keyword must come after the key and before the =
, but spacing does
not matter.
The hash (#
) symbol is used to add comments within a config file (think python syntax). Any characters after the #
will be considered part of the comment until the end-of-line. This allows for trailing comments. While there is no multi-line comment character, multi-line comments can be achieved by placing a #
at the beginning of each line.
# Comments can start at the beginning of a line
# Leading spaces are also okay
struct foo { # A trailing comment works like this
key1 = 0
# comments can also be embedded in a struct/proto/reference
key2 = "a" # We might leave a comment here to explain this variable
}
# A multi-line comment might look something
# like this, which is basically just a few
# normal comments in a row.
See config_example3.cfg
and config_example5.cfg
for additional examples of comment usage.
When interfacing with other systems and tools that don't support FlexiConfig natively it's often helpful to have a JSON representation of the fully materialized FlexiConfig tree. To export a config to a JSON string, you can use one of the following options:
#include <flexi_cfg/reader.h>
#include <flexi_cfg/visitor-json.h>
// Parse the config
auto cfg = flexi_cfg::Parser::parse(std::filesystem::path("config.cfg"));
// Human readable indented JSON format
auto visitor = flexi_cfg::visitor::PrettyJsonVisitor();
cfg.visit(visitor);
std::string json = visitor;
// Compact JSON format
auto visitor = flexi_cfg::visitor::JsonVisitor();
cfg.visit(visitor);
std::string json = visitor;
import flexi_cfg
# Parse the config
cfg = flexi_cfg.parse("config.cfg")
# Human readable indented JSON format
json = cfg.json(pretty=True)
# Compact JSON format
json = cfg.json()
The C++ API provides a visitor pattern through flexi_cfg::Reader.visit(visitor)
to traverse the fully materialized FlexiConfig tree. This API offers a way to examine the configuration structure and data without requiring prior knowledge of specific keys
or types
in your code. One example of its utility is the [JSON Output](#JSON Output) functionality; additional formats can be similarly supported with ease.
Implement a visitor class
that matches one or more of the following <concepts> and pass it to the flexi_cfg::Reader
which will then invoke the respective callback functions as it encounters the various FlexiConfig elements the visitor is interested in:
The C++ implementation uses the taocpp::pegtl
library to define the grammar.
PEGTL
uses a templatized syntax to define the grammar (which can be found here). This is used
for parsing the raw config file, along with a set of actions which define how to act on the parse
output. Once the raw config files are parsed, there is a second pass that does the following:
proto
s into struct
s.PEGTL
also provides some additional functionality to analyze the defined grammar and to generate a parse-tree from a supplied configuration file.
The following dependencies are required in order to compile the code:
PEGTL
- The core library used for implementing the PEG-based parsermagic_enum
- A header only library providing static reflection for enums{fmt}
- A formatting library that provides an alternative to c stdio and C++ iostreamsrange-v3
- A range library for C++14/17/20googletest
- The Google unit testing frameworkAll of these dependencies are automatically collected/installed via CMake FetchContent
. Currently, there is no mechanism for using pre-installed versions.
This project is built using CMake. While there are a variety of ways to use cmake, these simple steps should lead to a successful build:
From the root of the source tree:
mkdir build
cd build
cmake ..
make
Or alternatively, using ninja
:
mkdir build
cd build
cmake -G Ninja ..
ninja
All C++-based tests can be found in the the tests
directory. Any new tests should be added here as well.
As mentioned above, the Google testing framework is used for testing. Once
the code is built, the tests can be run by executing ctest
from the build directory. Refer to the
ctest
documentation for further details. The tests can
also be run individually by executing the individual gtest binaries from the tests
directory within your build
directory. See the googletest documentation for options.
In addition to the tests, there are a number of simple applications that provide example code for the library usage.
config_build
- This application can be used to parse a config file and build the resulting config tree. Usage: ./src/config_reader ../example/config_example5.cfg
.config_reader_example
- This reads the config_example5.cfg
configuration file and attempts to read a variety of variables from it. This uses a verbose mode, which generates a lot of debug printouts, tracing the parsing and construction of the config data.Python bindings for this library are provided via pybind11. These bindings are not built by default, but can be enabled by setting
CFG_PYTHON_BINDINGS=ON
via cmake. The installation directory of these bindings can also be specified by setting
CFG_PYTHON_INSTALL_DIR
. If this variable is not set, the default system directory will be used. e.g. From the root of the repo:
mkdir build
cd build
cmake -DCFG_PYTHON_BINDINGS=ON -DCFG_PYTHON_INSTALL_DIR=/my/custom/path ..
ninja
ninja install