Even if you love Erlang as we do, from time to time you might ask yourself why some functions you normally find in other languages are not part of the erlang's standard library. When you ask yourself that type of question you should remember that an estimated 2 million people are currently working in COBOL and 1.5 million new lines of COBOL code are written every day. After feeling bad for those developers, you should send a pull request to erlang katana with the functions you use on a daily basis.
To sum up this is a grab bag of useful functions (ideally).
If you find any bugs or have a problem while using this library, please open an issue in this repo (or a pull request :)).
And you can check all of our open-source projects at inaka.github.io
#Objective
Check out as well katana-code and katana-test
ktn_date: functions useful for handling dates and time values.ktn_debug: functions useful for debugging.ktn_fsm: a nice wrapper on top ofgen_fsmktn_json: functions useful for processing & creating JSON.ktn_maps: functions useful for handling maps.ktn_numbers: functions useful for processing numeric values.ktn_recipe: A tool to structure code that consists of sequential steps in which decisions are made.ktn_rpc: functions useful for RPC mechanisms.ktn_task: functions useful for managing asyncronous tasks.ktn_user_default: useful functions for your erlang shell.
This module contains functions that are nice to have in your user default
module, and thereby added to your shell. To use them, copy the ones you want
into your ~/user_default.erl module.
The Katana Test Utilities includes two functions useful for testing REST APIs:
test_response/4 and assert_response/4.
assert_response/4 uses test_response/4 to check that a given assertion
holds. It's usage is identical to test_response/4, except that it fails when
the assertion fails.
test_response/3 provides some useful checks regarding request responses. The
call to test itself does not fail, test_response will return ok if the test
passed and {error, Reason} when it does not.
To have it fail on the same test use assert_response/4. test_response/4 is
intended to be called as:
ok = test_response(status, Response, "201"),
{error, {nomatch, "204", _}} = test_response(status, partial, Response, "40?"),The first argument is what part of the response must be tested, and may be one
of: status, headers or body.
The second argument must be a map with at least three keys:
status, the reponse statusheaders, the list of headers in the responsebody, the response body
The status test matches the response status agains some regex pattern. It has a short form for specifying return codes: the third argument must be a string, either representing the exact return code, a string representing a partial return code (using the ? character as a wildcard, e.g. "20?", "4??"), or a regular expression as processed by the re module. You cannot specify patterns with a ? in the second digit position, e.g. "2?1".
If the status test fails, it returns {error, {nomatch, Pattern, Status}} where Pattern is the provided pattern to match agains and Status is the response status.
The headers test compares a list of headers agains those present in a response. The comparison can check whether the set of headers provided matches exactly (with exact) or if it is a subset of the response headers (with partial). The headers list of is a list of {Header, Value} tuples. Note that the elements in both headers lists do not have to be in the same order, nor have the same case.
If the test should fail, the possible error values returned are:
- {missing_headers, Headers, ResHeaders} where Headers is the list of headers missing from the set of headers in the response ResHeaders.
- {nomatch, Headers, ResHeaders} if the test type was exact and the two sets of headers do not contain the same elements.
The body test's third argument may specify whether the body must match a provided regex ({partial, Pattern}) or match a given body exactly ({match, Text}).
Possible error values for the body assert are:
- {regex_compile_fail, Error}} if the regular expression fails to compile.
- {error, {nomatch, Pattern, Body}} if the body does not match the regex.
- {nomatch, ResBody, Body}} if the body does not match the text.
Recipe (noun): A set of conditions and parameters of an industrial process to obtain a given result.
A recipe is a series of steps to obtain a result. This word was chosen because 'procedure' is overloaded in computer sciences, and it is not exactly a finite state machine.
ktn_recipe arose from the need to restructure code implementing large application business logic decision trees. Long functions, deeply nested case expressions, code with several responsibilities, all make for unreadable and unmaintainable code. Every time one needs to make a change or add a feature, one must think about the non-obvious data flow, complex logic, possible side effects. Exceptions pile upon exceptions, until the code is a house of cards. If one is lucky (or responsible), one has comprehensive test suites covering all cases.
A better way to structure the code would be preferable, one that makes the flow obvious and separates responsibilities into appropriate code segments. One way is a finite state machine. OTP even has a behaviour for exactly this purpose. However, gen_fsm does not exactly fit our needs: To begin with, gen_fsms run in their own process, which may not be what is needed. Second, logic flow is not immediately obvious because the fsm's state depends on the result of each state function. Finally, a gen_fsm transitions only on receiving input, whereas we are looking for something that runs like normal code, "on its own". So, our fsm will be defined by something like a transition table, a single place you can look at and know how it executes. This specification will "drive" the recipe.
The most common case envisioned is a sequential series of steps, each of which makes a decision affecting later outcomes, so our design should be optimized for a single, linear execution flow, allowing for writing the minimum amount of code for this particular case. Keep the common case fast.
The simplest use case: create a module using the ktn_recipe behaviour.
It must export transitions/0, process_result/1, process_error/1 and a
series of unary functions called step functions.
transitions/0 must return a list of atoms naming the step functions, in the
order in which they must be called. Each step function takes a State
variable as input and if the step was succesful emits {ok, NewState} with
the updated state, or {error, NewState} if it was unsuccessful.
After running all the steps, process_result will take the resulting state
as input and should emit whatever you need.
If one of the state functions returns {error, NewState}, then
process_error/1 will be called taking the resulting state as input and
should handle all expected error conditions.
To run the recipe, call ktn_recipe:run(CALLBACK_MODULE, INITIAL_STATE).
For more advanced uses, continue reading.
The main functions are ktn_recipe:run/2-4. These will run a recipe and
return the result.
Recipes may be specified in two ways, which we call "implicit" and
"explicit", for lack of better words. In implicit mode, we pass run/2 a
callback module implementing the ktn_recipe behavior, which implements all
necessary functions. In explicit mode, we explicitly give run/4 the
transition table, a function to process the state resulting from running all
recipe steps to completion, a function to process erroneous states, and the
initial state of the recipe.
Step functions have the following type:
-type state() :: term().
-type output() :: term().
-spec Step(state()) -> {ok, state()}
| {output(), state()}
| {error, state()}
| {halt, state()}.The recipe state may be any value you need, a list, a proplist, a record, a map, etc. ktn_recipe does not use maps itself, although the test suite does.
The initial state passed to run/2-4 will be passed to the first step as-is,
and the subsequent resulting states will be fed to each step.
The step functions should, if possible, have no side effects. If so, and the recipe ends in an error state, it can be aborted without worrying about rolling back the side effects.
For example, if the recipe involves making changes in a database, these
should be made in the process_result/1 function, once it is certain that
all steps completed successfully.
In implicit mode, the module must export process_error/1 and
process_result/1; in explicit mode you may pass whichever function you
desire.
The purpose of the 'result processing function' is to transform the
resulting state into a usable value.
The purpose of the 'error processing function' is to extract the error value
from the state and do something according. If you are unfortunate enough
to have unavoidable side effects in your step functions, you may undo them
here.
Transition tables are lists. As stated, the simplest transition table is a list of atoms naming the step functions, but this is not the only way recipe states may be specified. A transition table is a list of transitions. What is permissible as a transition depends on whether we are running in implicit or explicit mode.
In explicit mode, a transition may be either be an external function, i.e.
fun module:function/arity, or a ternary tuple {F1, I, F2} in which the F1
and F2 elements are explicit functions and the I is the transition input.
This form reprensents a transition from step F1 to step F2, if F1 outputs
{Input, State} instead of {ok, State}.
In implicit mode, in addition to the se two forms of specifying step functions, an atom may be used. The module in which the function resides is assumed to be the provided callback module, hence it is 'implicit'.
Also as stated, in any mode, the return value of step functions must be one of:
{ok, State}{Output, State}{halt, State}{error, State}
If {error, State} is returned, run/2-4 will call the error processing
function. If {halt, State} is called, run/2-4 will call the result
processing function.
If {ok, State} is returned, run will call the next function in the
transition table. That is, it will search the transition table until it
finds the current's functions position, if necessary skip over entries
corresponding to the current function, and call the first function with a
different name it encounters.
If {Output, State} is returned, run will search the transition table for an
entry matching {Step, Output, NextStep}, and select NextStep as the function
to call. In this way, non-linear recipes that jump ahead or loop back may
be specified. It is recommended that the transition table describe a DAG,
unless the program logic really requires loops.
Before running a recipe, run/2-4 will 'normalize' the transition table to
remove implicit assumptions and so that it is composed only of ternary
tuples explicitly listing every transition. As a debugging aid, this
function normalize/1 is exported by the ktn_recipe module.
Suppose you have a web server with an endpoint /conversations, accepting the
DELETE method to delete conversations between users. To delete a
conversation, one must fetch the converation entity from the database, fetch
the contact (the other user in the conversation) specified by the user id
in the conversation's contact_id attribute, check the contact's status (if
it is blocked, etc) and also check whether the client is using version 1 of
our REST API or version 2, since one really deletes the conversation from
the database and the other merely clears messages from the conversation.
We could implement this as a recipe with four steps:
transitions() ->
[ get_conversation
, get_contact
, get_status
, check_api_version
].get_conversation fetches it from the database. If it has already been
deleted, it could return {error, NewState} and store the error in the state.
Likewise for get_contact, and get_status.
get_api_version would extract the api version from the request header or
url, and store the result in the recipe state. If all steps complete
successfully, process_result will effectively make the call to delete the
conversation from the database.
Using ktn_recipes, you could structure your application as:
+----------+ +-------------+ +---------+ +-----------+
| | | | | | | |
| Endpoint +---> Recipes for +---> Entity +---> Databases |
| handlers | | endpoint | | modules | | |
| | | actions | | | | |
| | | | | | | |
+----------+ +-------------+ +---------+ +-----------+
The handlers would have the sole responsibility of handling the request, i. e. parsing URL, header and body parameters, verfying them and invoking the correct recipe. Recipes would implement the business logic. Entity modules would abstract management of your systems entities, including issues such as caching, and eventually persisting the changes to the storage medium or obtaining the required information.
The function ktn_recipe:verify/1 takes either a recipe module or an explicit
transition table and will run several checks to verify that it will run
correctly. You may use this function to test your transition tables.
Note that verify/1 is implemented as a ktn_recipe, so you can use it as an
example.
If a step throws an exception, it will not be caught. That means you may see some of ktn_recipe's internal functions in the stacktrace. In general, the most common errors will be:
-
badly formed transition tables
-
not exporting step functions
-
bad return values from step functions
-
bad state values set by functions
-
and 2) should be detected by running verify on your recipe. 3) and 4) will be detected at run time and an appropriate error will be returned by
run/2-4. Of course, your tests should detect these cases. You do have tests, right?
Finally, there is one more function, ktn_recipe:pretty_print/1, which takes
either a callback module or a transition table, and prints the normalized
transition table.