Skip to content

Latest commit

 

History

History
667 lines (443 loc) · 18.3 KB

README.rst

File metadata and controls

667 lines (443 loc) · 18.3 KB
https://api.travis-ci.org/derrley/yamlicious.png

Yamlicious is a (work-in-progress) lightweight configuration library built on top of YAML. It's for folks who love to write their configuration files in YAML, but who find themselves writing additional boilerplate each time they want to use YAML to encode a configuration file.

In addition to a parser, yamlicious provides code that:

  • Validates the correctness of a file and prints an understandable failure message when the file doesn't quite match
  • Understands your environemnt, and allows you to configure different properties for different environment dimensions without unnecessary DRY-violation
  • Reads configuration from multiple files
  • Provides a cascading set of configuration sources, including environment variables
  • Lazily loads certain parts of a large configuration. (read: when your "configuration" really starts to feel more like "data.")
  • Is generically extensible to new behavior, just in case you need to add something we haven't thought of

Yamlicious does all this by being like any other YAML parser, but by treating a handful of specific keys in a special way. It can be configured to recognize and evaluate any subset of these feature keys, allowing a developer to bake in as much (or as little!) of the crazy capabilities as seems reasonable for the situation. (The craziest stuff is turned off by default.)

Skip to Feature Key Definitions if you want the formality without the English verbiage.

A configuration file specified in Yamlicious is just YAML. In fact, nothing stops you from parsing a Yamlicious file with any YAML library. This can be useful for searching, cleaning, or editing (with syntax highlighting) the file.

Yamlicious gives your configuration file automatic access to the environment.

some:
  key:
    user_$(USER): not cool
    user_kyle: cool

Don't worry. The Yamlicious API gives you (the developer) the ability to both limit the environment variables your configuration file is allowed to access and provide overriding values for anything in the environment. You're neither forced to use environment variables nor doomed to a free for all.

Yamlicious supports list values in the environment, and splits environment variables on the comma character. If a Yamlicious file substitutes a list variable into a string, that string renders into multiple strings (one for each value in the list variable).

If a string references a variable that's not in the environment, yamlicious does not modify the string. It will leave the variable reference alone, $() characters included. This makes string substitution have a neat property -- it can be applied to the same document iteratively as the environment grows.

List substitution behavior varies subtly by situation.

some_list: $(MY_LIST)

renders to the following, if MY_LIST=one,two,three is in the environment.

some_list:
  - one
  - two
  - three
some_list:
  - first
  - $(LIST)

becomes

some_list:
  - first
  - one
  - two
  - three

Key strings are special, because you almost certainly don't intend to make a list into the key of a dictionary. Instead, you likely mean to define a key in the dictionary for each item in the list. Yamlicious provides a special variable in the environment, _KEY, to help you out in this situation.

$(LIST): $(_KEY) is in the list!

becomes

one: one is in the list
two: two is in the list
three: three is in the list

_KEY is set to the first-level key in the document, regardless of whether the key was derived from string substitution. To get the second-level key, use __KEY, and so forth:

$(LIST):
  second_level_key: $(_KEY) on top and $(__KEY) on bottom

becomes

one:
  second_level_key: one on top and second_level_key on bottom
two:
  second_level_key: two on top and second_level_key on bottom
three:
  second_level_key: three on top and second_level_key on bottom

This is interpreted as a dot product. Yamlicious will substitute every combination of variables between the two lists.

If BOYS=joey,johnny,bobby and GIRLS=sally,mary then:

"$(BOYS) likes $(GIRLS)"

becomes:

- joey likes sally
- joey likes mary
- johnny likes sally
- johnny likes mary
- bobby likes sally
- bobby likes mary

Note -- the rest of the "positional" list substitution rules (defined in the immediately previous sections) apply to dot product substitutions.

Sometimes, it makes sense to define sub-configuration somewhere outside the main configuration file. (e.g., secrets go somewhere special.) Yamlicious gives you the insert key to accomplish this.

Note: insert_ is a functional feature key. (Defined more in the Functional Feature Keys section.) These are keys that participate in something like a function call -- the entire map that contains a functional key evaluates to functional behavior applied to the key's value. (No project is complete without a smidge of functional programming.) You can only use one of these keys in a map at a time because yamlicious replaces the key-containing map with a document -- the result of the function applied to the key's value. Multiple keys is an abiguous definition.

some_place:
  placed_here:
    _insert: other/file.yaml

In this case, the rendered YAML output of other/file.yaml is placed under the placed_here key.

some_place:
  placed_here:
    contents of:
      - that other file
      - which can be arbitrary YAML

You can use variable substitution with the insert feature to get conditional configuration.

user_settings:
  _insert: $(USER)/conf.yaml

Yamlicious allows you to merge an external file into a bit of config.

merged_settings:
  _merge:
    - some_list: ['thing']
      some_thing: 'thing'

    - _insert: some_other_place.yaml

When you ask Yamlicious to do this, it will use a strategy I call safe deep merge with list append. Yamlicious merges dictionaries recursively by combining their key-value pairs. It merges lists by list addition. It refuses, however, to merge anything else. (Anything else would be shoot-self-in-foot territory, and I'd rather not encourage it.)

if some_other_place.yaml looked like this:

some_list: ['second_thing']
some_other_thing: 'thing'

The above configuration would render as follows:

merged_settings:
  some_list: ['thing', 'second_thing']
  some_thing: 'thing'
  some_other_thing: 'thing'

If you're looking to implement the common default override pattern, specify The Default Document as part of the Yamlicious API. That feature is specifically built to help you not have to allow arbitrary overrides when including files. If you absolutely must allow overrides, use the merge_override keyword, but note that it is turned off by default.

Loading several files and merging them is a common pattern, and it would be nice if folks didn't have to be verbose if that's the behavior they're looking for. This is what the insert_merge key is for.

merged_stuff:
  _insert_merge:
    - first/place.yaml
    - second/place.yaml
    - third/place.yaml

This key will load each file in order and merge that file into the previous file.

If you'd like to merge an entire document with your own, use the include feature key.

Note: include is a document feature key. (Defined more in the Document Feature Keys section.) Unlike functional feature keys, which apply behavior to any map embedded anywhere in the document hierarchy, document feature keys apply behavior to the entire document, and therefore must exist at the top of the YAML document.

You can also use the env document key to place new variables into the environment.

_env:
  COOLEST_PERSON: kyle

These variables can be used either in the same document (although the utility of that is not immediately obvious, other than for mitigating DRY violation) or, more importantly, in documents that include it. Yamlicious supports this by taking special care to re-run string substitution each time it changes a document's environment. (Remember, string substitution is idempotent.)

This behavior is somewhat dangerous if the included document defines a variable that's already defined in the including document. If the including document uses string subtitution to define included document paths, those substitutions can happen using only the initial version of the environment (before it is mutated by the act of inclusion). If the included document then changes any key that's used in the process of inclusion, things get hard to reason about.

Rather than allow such craziness, Yamlicious bans it. That is, it does not allow multiple documents included in the same parent document to define differing versions of the same environment variable. It does allow actual environment variables to coexist with (and override) those defined in env. Not allowing this would be brittle and would remove a very common use case, where setting an environment variable changes some sort of important behavior.

Yamlicious merge-overrides the configuration document it renders with a default document that it is configured to use.

This is the only place that, by default, uses the merge-override (rather than safe merge) behavior. For that reason, it's best to use the default document feature to specify override behavior. If you're wanting override behavior that can't be done by using the default document, chances are you're doing something that's either too complex or wrong. If you insist, there's always merge_override.

To specify a condition in-line, you can use the functional conditional feature keys (case and cond), each inspired by Lisp. This adds a bit too much Turing completeness to the project for the taste of most, so these are disabled by default.

case_configuration:
  '_case':
    - '$(USER)'
    - {'kyle': 'is awesome'}
    - {'_otherwise': 'is not awesome'}
cond_configuration:
  '_cond':
    - {"$(ENV) in ['test', 'prod']": 'go!'}
    - {true: 'undefined'}

Note the use of the python expression. This is mostly for convenience and terseness. Nobody wants to write a boolean expression in YAML, and I don't particularly want to implement it, either, so Yamlicious eval() s every single string that it finds below either functional conditional key.

List substitution works in both kinds of functional conditional. For example, if GOOD_USERS=kyle,anthony, then the following expression

access_configuration:
  '_case':
    - {'$(GOOD_USERS)': 'go!'}
    - {'_otherwise':  'stay. :('}

evaluates to

access_configuration:
  '_case':
    - {'kyle': 'go!',
       'anthony': 'go!'}
    - ['_otherwise',  'stay. :(']

Yamlicious is careful to "do the right thing" here. While there is no defined order in how it matches either the key 'anthony' or 'kyle', it will try to match both before falling back to the otherwise key.

Be careful to not do something like this unless you really mean it:

access_configuration:
  '_case':
    - {'$(GOOD_USERS)': '$(_KEY)'}
    - {'_otherwise':  'stay. :('}

While it will technically work, Yamlicious offers no definition for what the above expression evaluates to -- the order of iteration for a map/dictionary is an implementation detail.

If you notice an explosion in the number of Yamlicious files that your program includes, and you also notice that only a few of them ever get used, you'll likely want to conditionally load said files only when they're needed. Yamlicious provides two lazy loading keys to help you with this.

The lazy key changes nothing about the semantic meaning of the document it points to. It does change the time when functional key evaluation happens. Yamlicious evaluates embedded functional keys at lookup time, rather than during the depth-first functional key evaluation of the entire document.

In this example

_lazy:
  one:
    _insert: some/other/file.yaml

The insert evaluation happens only when someone tries to look at the one key.

The lazy_lookup key delays functional key evaluation just like lazy, and it also allows you to use string substitution of the special variable $(_KEY) to define how every key in the document is looked up. Rather than defining a document for every key in the map, you define one expression that, after string subtitution, can evaluate to any key.

To get the most power, pair lazy lookup with file inclusion. Here's an example inspired by YAML configuration of SQL tables.

tables:
  _lazy_lookup:
    _insert_merge:
      - generic/schema/$(_KEY).table.yaml
      - $(SYSTEM)/schema/$(_KEY).table.yaml
      - $(INSTITUTION)/schema/$(_KEY).table.yaml

Note that there's nothing that prevents lazy-loaded documents from merging with one another. If you're feeling particularly masochistic, you can define this confusing yet equivalent thing.

tables:
  _merge:
    - _lazy_lookup:
        _insert_merge:
          - generic/schema/$(_KEY).table.yaml
          - $(SYSTEM)/schema/$(_KEY).table.yaml
    - _lazy_lookup:
        _insert_merge:
          - $(INSTITUTION)/schema/$(_KEY).table.yaml

Yamlicious goes through the following phases when processing a document:

  1. String substitution.
  2. Document key evaluation.
  3. String substitution.
  4. Functional key evaluation (depth-first).

Enough with your words. Let's define this stuff explicitly.

These keys must be placed at the top level of a document, and affect the entire document that they're placed inside. They disappear when rendered.

_env:
  <variable>: <value>
  ...

Sets document environment variables to given values.

_include:
  - <file-path>
  - <file-path>
  - ...

Loads and safe-merges several files into the document.

The key must exist by itself in its containing dictionary. The feature key, itself, describes a transformation operation on the given document.

_<feature-key>: <document>
_insert <file-path>

Evaluates to the loaded and processed configuration document found at file_path.

_merge: [ <document>, <document>, ... ]

Uses safe-merge-with-list-append to merge given documents together. Can safely merge dictionaries and lists, but nothing else.

_merge_override: [ <document>, <document>, ... ]

Uses deep-merge to merge given documents together. Can safely merge anything. For scalar values, documents further down the list override documents earlier in the list.

_insert_merge [ <file-path>, <file-path>, ... ]

Loads files and then merges them with safe-merge-with-list-append.

(NOT IMPLEMENTED)

'_case':
  - <key-python-expression>
  - {<match-python-expression>, <outcome-python-expression>}
  ...

Functional case. Evaluates to the first outcome expression whose match expression is python-equal to the key expression.

(NOT IMPLEMENTED)

'_otherwise': <expression>

Evaluates to a case condition that always matches.

(NOT IMPLEMENTED)

'_cond':
  - {<boolean-python-expression>, <outcome-python-expression>}
  ...

Functional cond. Evaluates to the first outcome expression whose boolean expression is true.

(NOT IMPLEMENTED)

'_lazy': <document>

Evaluates to document, but where each of the keys in document is lazy-loaded.

(NOT IMPLEMENTED)

'_lazy_lookup': <value-expression>

Evaluates to a lazy-loaded dictionary, where every key is evaluated at lookup time by evaluating the value-expression, which is allowed to use the _KEY environment variable

TBD

Yamlicious comes with a convenient command, yamlicious, that reads input from stdin and writes to stdout. It uses a default configuration, along with all environment variables, in order to process the yaml document fed to it on standard in:

[10:49:46][kderr@Kyles-MacBook-Pro][~/Repositories/derrley/yamlicious]
$ cat /tmp/test
Hello: "$(PWD) is the current wd"


[10:49:52][kderr@Kyles-MacBook-Pro][~/Repositories/derrley/yamlicious]
$ cat /tmp/test | yamlicious
{Hello: /Users/kderr/Repositories/derrley/yamlicious is the current wd}

TBD

TBD