|
| 1 | +.. _background: |
| 2 | + |
| 3 | +Background |
| 4 | +========== |
| 5 | + |
| 6 | +This document describes what design decisions were made during the development |
| 7 | +of this library. It also describes how the library works in detail. |
| 8 | + |
| 9 | +This document assumes that you can use the library and have at least limited |
| 10 | +knowledge about the source code. If you feel that it is not true for you, you |
| 11 | +may want to read the :ref:`tutorial` and :ref:`howto`. |
| 12 | + |
| 13 | + |
| 14 | +API design |
| 15 | +---------- |
| 16 | + |
| 17 | +The API should be simple for use to simple cases, but also provide great |
| 18 | +flexibility for the advanced cases. When increasing flexibility of the API it |
| 19 | +should not change the simple API unless it is absolutely required. |
| 20 | + |
| 21 | +API compatibility is paramount. API breaking is only allowed when it is on par |
| 22 | +with the the the gain of the new functionality. |
| 23 | + |
| 24 | +Adding new parameters to functions which have default value is not considered a |
| 25 | +breaking API change. |
| 26 | + |
| 27 | + |
| 28 | +Simple API |
| 29 | +~~~~~~~~~~ |
| 30 | + |
| 31 | +API should be kept as simple as possible. It means that describing an expected |
| 32 | +request and its response should be trivial for the user. For this reason, the |
| 33 | +API is flat: it contains a handful of functions which have many parameters |
| 34 | +accepting built-in python types (such as bytes, string, int, etc) in contrast |
| 35 | +to more classes and functions with less arguments. |
| 36 | + |
| 37 | +This API allows to define an expected request and the response which will be |
| 38 | +sent back to the client in a single line. This is one of the key features so |
| 39 | +using the library is not complicated. |
| 40 | + |
| 41 | +Example: |
| 42 | + |
| 43 | +.. code-block:: python |
| 44 | +
|
| 45 | + def test_query_params(httpserver): |
| 46 | + httpserver.expect_request("/foo", query_string={"user": "user1"}).respond_with_data("OK") |
| 47 | +
|
| 48 | +It is simple in the most simple cases, but once the expectation is more |
| 49 | +specific, the line can grow significantly, so here the user is expected to put |
| 50 | +the literals into variables: |
| 51 | + |
| 52 | +.. code-block:: python |
| 53 | +
|
| 54 | + def test_query_params(httpserver): |
| 55 | + httpserver.expect_request("/foo", query_string=expected_query).respond_with_data("OK") |
| 56 | +
|
| 57 | +
|
| 58 | +If the user wants something more complex, classes are available for this which |
| 59 | +can be instantiated and then specified for the parameters normally accepting |
| 60 | +only built-in types. |
| 61 | + |
| 62 | +The easy case should be made easy, with the possibility of making advanced |
| 63 | +things in a bit more complex way. |
| 64 | + |
| 65 | +Flexible API |
| 66 | +~~~~~~~~~~~~ |
| 67 | + |
| 68 | +The API should be also made flexible as possible but it should not break the |
| 69 | +simple API and not make the simple API complicated. A good example for this is |
| 70 | +the `respond_with_handler` method, which accepts a callable object (eg. a |
| 71 | +function) which receives the request object and returns the response object. |
| 72 | + |
| 73 | +The user can implement the required logic there. |
| 74 | + |
| 75 | +Adding this flexibility however did not cause any change in the simple API, the |
| 76 | +simple cases can be still used as before. |
| 77 | + |
| 78 | + |
| 79 | +Higher-level API |
| 80 | +~~~~~~~~~~~~~~~~ |
| 81 | + |
| 82 | +In the early days of this library, it wanted to support the low-level http |
| 83 | +protocol elements: request status, headers, etc to provide full coverage for the |
| 84 | +protocol itself. This was made in order to made the most advanced customizations |
| 85 | +possible. |
| 86 | + |
| 87 | +Then the project received a few PRs adding `HeaderValueMatcher` and support for |
| 88 | +authorization which relied on the low-level API to add a higher-level API |
| 89 | +without breaking it. In the opposite case, adding a low-level API to a |
| 90 | +high-level would not be possible. |
| 91 | + |
| 92 | +Transparency |
| 93 | +~~~~~~~~~~~~ |
| 94 | + |
| 95 | +The API provided by *pytest-httpserver* is transparent. That means that the |
| 96 | +objects (most importantly the `Request` and `Response` objects) defined by |
| 97 | +*werkzeug* are visible by the user of *pytest-httpserver*, there is no wrapping |
| 98 | +made. This is done by the sake of simplicity. |
| 99 | + |
| 100 | +As *werkzeug* provides a stable API, there's no need to change this in the |
| 101 | +future, however this also limits the library to stick with *werkzeug* in the |
| 102 | +long term. Replacing *werkzeug* to something else would break the API due to |
| 103 | +this transparency. |
| 104 | + |
| 105 | +Requirements |
| 106 | +------------ |
| 107 | + |
| 108 | +This section describes how to work with pytest-httpserver's requirements. |
| 109 | +These are the packages used by the library. |
| 110 | + |
| 111 | +Number of requirements |
| 112 | +~~~~~~~~~~~~~~~~~~~~~~ |
| 113 | + |
| 114 | +It is required to keep the requirements at minimum. When adding a new library to |
| 115 | +the package requirements, research in the following topics should be done: |
| 116 | + |
| 117 | +* code quality |
| 118 | +* activity of the development and maintenance |
| 119 | +* number of open issues, and their content |
| 120 | +* how many people using that library |
| 121 | +* python interpreter versions supported |
| 122 | +* amount of API breaking changes |
| 123 | +* license |
| 124 | + |
| 125 | +Sometimes, it is better to have the own implementation instead of having a tiny |
| 126 | +library added to the requirements, which may cause compatibility issues. |
| 127 | + |
| 128 | + |
| 129 | +Requirements version restrictions |
| 130 | +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 131 | + |
| 132 | +In general, the package requirements should have no version restrictions. For |
| 133 | +example, the *werkzeug* library has no restrictions, which means that if a new |
| 134 | +version comes out of it, it is assumed that *pytest-httpserver* will be able to |
| 135 | +run with it. |
| 136 | + |
| 137 | +Many people uses this library in an environment having full of other packages |
| 138 | +and limiting version here will limit their versions in their requirements also. |
| 139 | +For example if there's a software using *werkzeug* `1.0.0` and our requirements |
| 140 | +have `<0.9` specified it will make *pytest-httpserver* incompatible with their |
| 141 | +software. |
| 142 | + |
| 143 | + |
| 144 | +Requirements testing |
| 145 | +~~~~~~~~~~~~~~~~~~~~ |
| 146 | + |
| 147 | +Currently it is required to test with only the latest version of the required |
| 148 | +packages. However, if there's an API breaking change which affects |
| 149 | +*pytest-httpserver*, a decision should be made: |
| 150 | + |
| 151 | +* apply version restrictions, possibly making *pytest-httpserver* incompatible |
| 152 | + with some other software |
| 153 | + |
| 154 | +* add workaround to *pytest-httpserver* to support both APIs |
| 155 | + |
| 156 | + |
| 157 | +HTTP server |
| 158 | +----------- |
| 159 | + |
| 160 | +The chosen HTTP server which drives this library is imlemented by the *werkzeug* |
| 161 | +library. The reason behind this decision is that *werkzeug* is used by Flask, a |
| 162 | +very popular web framework and it provides a proven, stable API in the long |
| 163 | +term. |
| 164 | + |
| 165 | +Supported python versions |
| 166 | +------------------------- |
| 167 | + |
| 168 | +Supporting the latest python versions (such as 3.7 and 3.8 at the time of |
| 169 | +writing this), is a must. Supporting the older versions is preferred, following |
| 170 | +the state of the officially supported python versions by PSF. |
| 171 | + |
| 172 | +The library should be tested periodically on the supported versions. |
| 173 | + |
| 174 | +Dropping support for old python versions is possible if supporting would cause |
| 175 | +an issue or require extensive workaround. Currently, 3.4 is still supported by |
| 176 | +the library, however it is deprecated by PSF. As it causes no problems for |
| 177 | +*pytest-httpserver* (there's an additional requirement for this in the setup.py, |
| 178 | +but that's all), the support for this version will be maintained as long as |
| 179 | +possible. Once a new change is added to the library which require great effort |
| 180 | +to maintain compatibility with 3.4, the support for it will be dropped. |
| 181 | + |
| 182 | + |
| 183 | +Testing and coverage |
| 184 | +-------------------- |
| 185 | + |
| 186 | +It is not required to have 100% test coverage but all possible use-cases should |
| 187 | +be covered. Github actions is used to test the library on all the supported |
| 188 | +python versions, and tox.ini is provided if local testing is desired. |
| 189 | + |
| 190 | +When a bug is reported, there should be a test for it, which would re-produce |
| 191 | +the error and it should pass with the fix. |
| 192 | + |
| 193 | +Server starting and stopping |
| 194 | +---------------------------- |
| 195 | + |
| 196 | +The server is started when the first test is run which uses the httpserver |
| 197 | +fixture. It will be running till the end of the session, and new tests will use |
| 198 | +the same instance. A cleanup is done between the tests which restores the clean |
| 199 | +state (no handlers registered, empty log, etc) to avoid cross-talk. |
| 200 | + |
| 201 | +The reason behind this is the time required to stop the server. For some reason, |
| 202 | +*werkzeug* (the http server used) needs about 1 second to stop itself. Adding this |
| 203 | +time to each test is not acceptable in most of the cases. |
| 204 | + |
| 205 | +Note that it is still compatible with *pytest-xdist* (a popular pytest extension |
| 206 | +to run the tests in parallel) as in such case, distinct test sessions will be |
| 207 | +run and those will have their own http server instance. |
| 208 | + |
| 209 | + |
| 210 | +Fixture scope |
| 211 | +------------- |
| 212 | + |
| 213 | +Due to the nature of the http server (it is run only once), it seems to be a |
| 214 | +good recommendation to keep the httpserver fixture session scoped, not function |
| 215 | +scoped. The problem is that the cleanup which needs to be done between the |
| 216 | +tests (as the server is run only once, see above), and that cleanup needs to be |
| 217 | +attached to a function scoped fixture. |
| 218 | + |
| 219 | +HTTP port selection |
| 220 | +------------------- |
| 221 | + |
| 222 | +In early versions of the library, the user had to specify which port the server |
| 223 | +should be bound. This later changed to have an so-called ephemeral port, which |
| 224 | +is a random free port number chosen by the kernel. It is good because it |
| 225 | +guarantees that it will be available and it allows parallel test runnings for |
| 226 | +example. |
| 227 | + |
| 228 | +In some cases it is not desired (eg if the code being tested has wired-in port |
| 229 | +number), in such cases it is still possible to specify the port number. |
| 230 | + |
| 231 | +Also, the host can be specified which allows to bind on "0.0.0.0" so the server |
| 232 | +is accessible from the network in case you want to test a javascript code |
| 233 | +running on a different server in a browser. |
0 commit comments