Python syntax highlighted Markdown doctest
Command line program and Python library to test Python syntax highlighted code examples in Markdown.
Creates a pytest Python module that tests Python examples in README and other Markdown files.
Reads these from Markdown fenced code blocks:
Python interactive sessions described by doctest.
Python source code and expected terminal output.
The test cases get run later by running pytest.
Simple use case is possible with no Markdown edits at all.
More features selected by adding HTML comment directives to the Markdown.
Set test case name.
Add a pytest custom marker.
Add a pytest.mark.skip decorator.
Promote names defined in a test case to module level globals.
Label any fenced code block for later retrieval (API).
Configurable. Discover and process many Markdown files in a single command.
Add inline annotations to comment out sections of code.
Get code coverage by running pytest with coverage.
Select Python source code blocks as setup and teardown code.
Setup applies to code blocks and optionally to session blocks.
An included Python library: Latest Development tools API.
Python function returns test file in a string. (testfile() in main.py)
Two pytest fixtures. (tester.py)
testfile_creator runs testfile(). Use with testfile_tester.
testfile_tester runs a pytest file with pytest’s pytester in its isolated environment.
Runs phmdoctest and can run pytest too. (simulator.py)
Functions to read fenced code blocks from Markdown. (tool.py)
Test Markdown for Python examples. (tool.py)
Prepare directory for generated test files. (tool.py)
Extract testsuite tree and list of failing trees from JUnit XML. (tool.py)
Available as the pytest plugin pytest-phmdoctest.
default branch status
Introduction | Installation | Sample usage | Sample Usage with HTML comment directives | CI usage | –report | Identifying blocks | Directives | skip | label on code and sessions | label on any fenced code block | pytest skip | pytest skipif | setup | teardown | share-names | clear-names | pytest mark decorator | label skip and mark example | setup and teardown example | share-names clear-names example | Configuration | Inline annotations | skipping blocks with –skip | –skip | short form of –skip | –fail-nocode | –setup | –teardown | Setup example | Setup for sessions | Execution context | Send outfile to stdout | Usage | Run as a Python module | Python API | pytest fixtures | Simulate command line | Hints | Directive hints | Related projects
It is advisable to install in a virtual environment.
python -m pip install phmdoctest
Given the Markdown file example1.md shown in raw form here…
# This is Markdown file example1.md ## Interactive Python session (doctest) ```py >>> print("Hello World!") Hello World! ``` ## Source Code and terminal output Code: ```python from enum import Enum class Floats(Enum): APPLES = 1 CIDER = 2 CHERRIES = 3 ADUCK = 4 for floater in Floats: print(floater) ``` sample output: ``` Floats.APPLES Floats.CIDER Floats.CHERRIES Floats.ADUCK ```
phmdoctest doc/example1.md --outfile test_example1.py
creates the python source code file
test_example1.py shown here…
"""pytest file built from doc/example1.md""" from phmdoctest.functions import _phm_compare_exact def session_00001_line_6(): r""" >>> print("Hello World!") Hello World! """ def test_code_14_output_28(capsys): from enum import Enum class Floats(Enum): APPLES = 1 CIDER = 2 CHERRIES = 3 ADUCK = 4 for floater in Floats: print(floater) _phm_expected_str = """\ Floats.APPLES Floats.CIDER Floats.CHERRIES Floats.ADUCK """ _phm_compare_exact(a=_phm_expected_str, b=capsys.readouterr().out)
Then run a pytest command something like this in your terminal to test the Markdown session, code, and expected output blocks.
Or these two commands:
pytest python -m doctest test_example1.py
line_6 in the function name
session_00001_line_6 is the
line number in example1.md of the first line
of the interactive session.
00001 is a sequence number to
order the doctests.
14 in the function name
test_code_14_output_28 is the
line number of the first line
of python code.
28 shows the line number of the expected
One test case function gets generated for each:
Markdown fenced code block interactive session
Python-code/expected-output Markdown fenced code block pair
--report option below shows the blocks discovered and
how they are tested.
Sample Usage with HTML comment directives
Given the Markdown file shown in raw form here…
<!--phmdoctest-mark.skip--> <!--phmdoctest-label test_example--> ```python print("Hello World!") ``` ``` incorrect expected output ```
phmdoctest tests/one_mark_skip.md --outfile test_one_mark_skip.py
creates the python source code file shown here…
"""pytest file built from tests/one_mark_skip.md""" import pytest from phmdoctest.functions import _phm_compare_exact @pytest.mark.skip() def test_example(capsys): print("Hello World!") _phm_expected_str = """\ incorrect expected output """ _phm_compare_exact(a=_phm_expected_str, b=capsys.readouterr().out)
Run the –outfile with pytest…
$ pytest -vv test_one_mark_skip.py test_one_mark_skip.py::test_example SKIPPED
The HTML comments in the Markdown are phmdoctest directives.
The mark.skip directive adds the @pytest.mark.skip() line.
The label directive names the test case function.
List of Directives
Directives are optional.
Markdown edits are optional.
Test Python examples in README.md in Continuous Integration scripts. In this snippet for Linux the pytest test suite is in the tests folder.
mkdir tests/tmp phmdoctest README.md --report --outfile tests/tmp/test_readme.py pytest --doctest-modules -vv tests
This console shows testing Python examples in project.md. Look for the tmp tests at the bottom. Windows Usage on Appveyor.
See this excerpt from ci.yml Actions usage example. It runs on Windows, Linux, and macOS. Please find the phmdoctest command at the bottom.
No changes to README.md are needed here, look in the last job log.
To see the GFM fenced code blocks in the MARKDOWN_FILE use the
--report option like this:
phmdoctest doc/example2.md --report
which lists the fenced code blocks it found in
the file example2.md.
test role column shows how each fenced code block gets tested.
doc/example2.md fenced blocks ------------------------------------------------ block line test TEXT or directive type number role quoted and one per line ------------------------------------------------ python 9 code 14 output python 20 code 26 output 31 -- python 37 code python 44 code 51 output yaml 59 -- text 67 -- py 75 session python 87 code 94 output py 102 session ------------------------------------------------ 7 test cases. 1 code blocks with no output block.
Python code, expected output, and Python interactive sessions get extracted.
Only GFM fenced code blocks are considered.
A block is a session block if the info_string starts with
and the first line of the block starts with the
To be treated as Python code the opening fence should start with one of these:
```python ```python3 ```py3
plus the block contents can’t start with
The examples use the info_strings
python for code and
py for sessions
since they render with coloring on GitHub, readthedocs, GitHub Pages,
and Python package index.
project.md has more examples of code and session blocks.
It is ok if the info string is laden with additional text, it will be ignored. The entire info string will be shown in the block type column of the report.
An output block is a fenced code block that immediately follows a Python block and starts with an opening fence like this which has an empty info string.
A Python code block has no output if it is followed by any of:
Python code block
Python session block
a fenced code block with a non-empty info string
Test code gets generated for it, but there will be no assertion statement.
Directives are HTML comments containing test generation commands.
They are edited into the Markdown file immediately before a fenced
code block. It is OK if other HTML comments are present.
<!--phmdoctest-skip--> directive in the
raw Markdown below.
With the skip directive no test code will be
generated from the fenced code block.
<!--phmdoctest-skip--> <!--Another HTML comment--> ```python print("Hello World!") ``` Expected Output ``` Hello World! ```
List of Directives
Directive HTML comment | Use on blocks ---------------------------------- | --------------------- <!--phmdoctest-skip--> | code, session, output <!--phmdoctest-label IDENTIFIER--> | code, session <!--phmdoctest-label TEXT--> | any <!--phmdoctest-mark.skip--> | code <!--phmdoctest-mark.skipif<3.N--> | code <!--phmdoctest-setup--> | code <!--phmdoctest-teardown--> | code <!--phmdoctest-share-names--> | code <!--phmdoctest-clear-names--> | code <!--phmdoctest-mark.ATTRIBUTE--> | code
The skip directive or
--skip TEXT command line option
prevents code generation for the code or session block.
The skip directive can be placed on an expected output block.
There it prevents checking expected against actual output.
label on code and sessions
When used on a Python code block or session the label directive changes the name of the generated test function. Example. Two generated tests, the first without a label, shown in pytest -v terminal output:
test_readme.py::test_code_93 FAILED test_readme.py::test_beta_feature FAILED
label on any fenced code block
On any fenced code block, the label directive identifies the block
for later retrieval by the class
FCBChooser is used separately from phmdoctest in
a different pytest file. This allows the test developer to write
additional test cases for fenced code blocks that are not handled by
phmdoctest. The directive value can be any string.
# This is file doc/my_markdown_file.md <!--phmdoctest-label my-fenced-code-block--> ``` The label directive can be placed on any fenced code block. ```
Here is Python code to fetch it:
import phmdoctest.tool chooser = phmdoctest.tool.FCBChooser("doc/my_markdown_file.md") contents = chooser.contents(label="my-fenced-code-block") print(contents)
The label directive can be placed on any fenced code block.
<!--phmdoctest-mark.skip--> directive generates a test
case with a
<!--phmdoctest-mark.skipif<3.N--> directive generates
a test case with the pytest decorator
@pytest.mark.skipif(sys.version_info < (3, N), reason="requires >=py3.N").
N is a Python minor version number.
A single Python code block can assign names visible to other code blocks by adding a setup directive or using the –setup command line option.
Names assigned by the setup block get copied to the test module’s global namespace after the setup block runs.
Here is an example setup block from setup.md:
import math mylist = [1, 2, 3] a, b = 10, 11 def doubler(x): return x * 2
Using setup modifies the execution context of the
Python code blocks in the Markdown file.
doubler are visible
to the other Python code blocks. The objects can be modified.
After the test case generated for the Python code block with the clear-names directive runs, all names that were created by one or more preceding share-names directives get deleted. The names that were shared are no longer visible. This directive also deletes the names assigned by setup. Example.
pytest mark decorator
<!--phmdoctest-mark.ATTRIBUTE--> directive adds
a @pytest.mark.ATTRIBUTE decorator to the
generated test function. ATTRIBUTE is a valid Python attribute
identifier. This defines a marker to pytest that is used to
select and deselect tests. See the pytest documentation section
“Working with custom markers”.
The file mark_example.md contains
example usage of the user defined marker “slow”. It generates
label skip and mark example
phmdoctest doc/directive1.md --outfile test_directive1.py
setup and teardown example
phmdoctest doc/directive2.md --outfile test_directive2.py
Supply a .ini, .cfg, or .toml configuration file in place of the Markdown file. Configuration features:
Choose Markdown files for test file generation. (glob wildcards).
Exclude Markdown files from test file generation. (glob wildcards).
Name the output directory.
Removes stale test files from output directory.
[tool.phmdoctest] section in the configuration file.
How to configure.
Inline annotations comment out sections of code. They can be added to the end of lines in Python code blocks. They should be in a comment.
phmdoctest:omitcomments out a section of code. The line it is on, plus following lines at greater indent get commented out.
phmdoctest:passcomments out one line of code and prepends the pass statement.
Here is a snippet showing how to place
phmdoctest:pass in the code.
The second block shows the code that is generated. Note there is no
phmdoctest:pass. It is not required.
import time def takes_too_long(): time.sleep(100) # delay for awhile. phmdoctest:pass takes_too_long()
import time def takes_too_long(): pass # time.sleep(100) # delay for awhile. phmdoctest:pass takes_too_long()
phmdoctest:omit on single or multi-line statements. Note the two
commented out time.sleep(99). They follow and are indented more
if condition:line with
import time # phmdoctest:omit condition = True if condition: # phmdoctest:omit time.sleep(99) time.sleep(99)
# import time # phmdoctest:omit condition = True # if condition: # phmdoctest:omit # time.sleep(99) # time.sleep(99)
Inline annotation processing counts the number of commented
out sections and adds the count as the suffix
_N to the name of the pytest function in the
generated test file.
Inline annotations are similar, but less powerful
than the Python standard library doctest directive
Improper use of
phmdoctest:omit can cause Python syntax errors.
The examples above are snippets that illustrate how to use inline annotations. Here is an example that produces a pytest file from Markdown. The command below takes inline_example.md and generates test_inline_example.py.
phmdoctest doc/inline_example.md --outfile test_inline_example.py
skipping blocks with skip option
If you don’t want to generate test cases for Python
blocks precede the block with a skip directive or
--skip TEXT option. More than one skip directive
--skip TEXTis allowed.
The following describes using
The code in each Python block gets searched
for the substring
TEXT. Zero, one or more blocks will contain
the substring. These blocks will not generate test cases in the
The Python code in the fenced code block gets searched.
The info string is not searched.
Output blocks are not searched.
Both Python code and session blocks get searched.
Case is significant.
The report shows which Python blocks get skipped in the test role column, and the Python blocks that matched each –skip TEXT in the skips section.
This option makes it very easy to inadvertently exclude
Python blocks from the test cases. In the event no test cases get
generated, the option
--fail-nocode described below is useful.
--skip TEXT strings work a little differently.
They select one of the first, second, or last of the Python blocks.
Only Python blocks get counted.
--skip FIRSTskips the first Python block.
--skip SECONDskips the second Python block.
--skip LASTskips the final Python block.
This command using
phmdoctest doc/example2.md --skip "Python 3.7" --skip LAST --report --outfile test_example2.py
Produces the report
doc/example2.md fenced blocks ----------------------------------------------------- block line test TEXT or directive type number role quoted and one per line ----------------------------------------------------- python 9 code 14 output python 20 skip-code "Python 3.7" 26 skip-output 31 -- python 37 code python 44 code 51 output yaml 59 -- text 67 -- py 75 session python 87 code 94 output py 102 skip-session "LAST" ----------------------------------------------------- 5 test cases. 1 skipped code blocks. 1 skipped interactive session blocks. 1 code blocks with no output block. skip pattern matches (blank means no match) ------------------------------------------------ skip pattern matching code block line number(s) ------------------------------------------------ Python 3.7 20 LAST 102 ------------------------------------------------
creates the output file test_example2.py
short form of skip option
This is the same command as above using the short
-s form of the
in two places.
It produces the same report and outfile.
phmdoctest doc/example2.md -s "Python 3.7" -sLAST --report --outfile test_example2.py
--fail-nocode option produces a pytest file that will always
fail when no Python code or session blocks get found.
Evem if no Python code or session blocks exist in the
Markdown file a pytest file gets generated.
This also happens when
--skip eliminates all the
Python code blocks.
The generated pytest file will have the function
If the option
--fail-nocode is passed the
def test_nothing_fails() which raises an
A single Python code block can assign names visible to
other code blocks by giving the
--setup TEXT option.
Please see the setup directive above.
The rules for
TEXT are the same as for
--skip TEXT plus…
Only one block can match
The block cannot match a block that is skipped.
The block cannot be a session block even though session blocks get searched for
It is ok if the block has an output block. It will be ignored.
A single Python code block can supply code run by the pytest
teardown_module() fixture. Use the
--teardown TEXT option.
Please see the teardown directive above.
The rules for
TEXT are the same as for
--setup above except
TEXT won’t match a setup block.
For the Markdown file setup.md run this command to see how the blocks get tested.
phmdoctest doc/setup.md --setup FIRST --teardown LAST --report
doc/setup.md fenced blocks ------------------------------------------------- block line test TEXT or directive type number role quoted and one per line ------------------------------------------------- python 9 setup "FIRST" python 20 code 27 output python 37 code 42 output python 47 code 51 output python 58 teardown "LAST" ------------------------------------------------- 3 test cases.
phmdoctest doc/setup.md --setup FIRST --teardown LAST --outfile test_setup.py
creates the test file test_setup.py
Setup for sessions
The pytest option
--doctest-modules is required to
run doctest on sessions. pytest runs doctests in
a separate context.
For more on this see Execution context below.
To allow sessions to see the variables assigned by the
code block, add the option
Here is an example with setup code and sessions setup_doctest.md. The first part of this file is a copy of setup.md.
This command uses the short form of setup and teardown. -u for setup and -d for teardown.
phmdoctest doc/setup_doctest.md -u FIRST -d LAST --setup-doctest --outfile test_setup_doctest.py
It creates the test file test_setup_doctest.py
When run without
pytest and doctest determine the order of test case execution.
phmdoctest assumes test code and session execution is in file order.
Test case order is not significant.
Code and expected output run within a function body of a pytest test case.
If pytest is invoked with
Sessions are run in a separate doctest execution context.
Otherwise, sessions do not run.
Names assigned by setup code are visible to code blocks.
Code blocks can modify the objects created by the setup code.
Code block test case order is significant.
Session order is not significant.
If pytest is run with
pytest runs two separate contexts: one for sessions, one for code blocks.
setup and teardown code gets run twice, once by each context.
the names assigned by the setup code block are
are notvisible to the sessions.
Same as the setup section plus:
names assigned by the setup code block are visible to the sessions.
Sessions can modify the objects created by the setup code.
Session order is significant.
Sessions and code blocks are still running in separate contexts isolated from each other.
A session can’t affect a code block, and a code block can’t affect a session.
Names assigned by the setup code block are globally visible to the entire test suite via the pytest doctest_namespace fixture. See hint near the end Hints.
pytest live logging demo
The live logging demos reveals pytest execution contexts. pytest Live Logs show the execution order of setup_module(), test cases, sessions, and teardown_module(). There are 2 demo invocations in the workflow action called pytest Live Log Demo. GitHub login required.
Send outfile to stdout
To redirect the above outfile to the standard output stream use one of these two commands.
Be sure to leave out
--report when sending –outfile to standard output.
phmdoctest doc/example2.md -s "Python 3.7" -sLAST --outfile -
phmdoctest doc/example2.md -s "Python 3.7" -sLAST --outfile=-
Usage: phmdoctest [OPTIONS] MARKDOWN_FILE MARKDOWN_FILE may also be .toml, .cfg, or .ini configuration file. Options: --outfile TEXT Write generated test case file to path TEXT. "-" writes to stdout. -s, --skip TEXT Any Python code or interactive session block that contains the substring TEXT is not tested. More than one --skip TEXT is ok. Double quote if TEXT contains spaces. For example --skip="python 3.7" will skip every Python block that contains the substring "python 3.7". If TEXT is one of the 3 capitalized strings FIRST SECOND LAST the first, second, or last Python code or session block in the Markdown file is skipped. --report Show how the Markdown fenced code blocks are used. --fail-nocode This option sets behavior when the Markdown file has no Python fenced code blocks or interactive session blocks or if all such blocks are skipped. When this option is present the generated pytest file has a test function called test_nothing_fails() that will raise an assertion. If this option is not present the generated pytest file has test_nothing_passes() which will never fail. -u, --setup TEXT The Python code block that contains the substring TEXT is run at test module setup time. Variables assigned at the outer level are visible as globals to the other Python code blocks. TEXT should match exactly one code block. If TEXT is one of the 3 capitalized strings FIRST SECOND LAST the first, second, or last Python code or session block in the Markdown file is matched. A block will not match --setup if it matches --skip, or if it is a session block. Use --setup-doctest below to grant Python sessions access to the globals. -d, --teardown TEXT The Python code block that contains the substring TEXT is run at test module teardown time. TEXT should match exactly one code block. If TEXT is one of the 3 capitalized strings FIRST SECOND LAST the first, second, or last Python code or session block in the Markdown file is matched. A block will not match --teardown if it matches either --skip or --setup, or if it is a session block. --setup-doctest Make globals created by the --setup Python code block or setup directive visible to session blocks and only when they are tested with the pytest --doctest-modules option. Please note that pytest runs doctests in a separate context that only runs doctests. This option is ignored if there is no --setup option. --version Show the version and exit. --help Show this message and exit.
Run as a Python module
To run phmdoctest from the command line:
python -m phmdoctest doc/example2.md --report
Call main.testfile() to generate a pytest file in memory. Please see the Python API here. The example generates a pytest file from doc/setup.md and compares the result to doc/test_setup.py.
from pathlib import Path import phmdoctest.main generated_testfile = phmdoctest.main.testfile( "doc/setup.md", setup="FIRST", teardown="LAST", ) expected = Path("doc/test_setup.py").read_text(encoding="utf-8") assert expected == generated_testfile
Use fixture testfile_creator to generate a test file in memory.
Pass the test file to fixture testfile_tester to run
the test file in the pytester environment.
Fixture API | Example.
See more uses in tests/test_examples.py, tests/test_details.py, and
The fixtures run pytest much faster than
below since there is no subprocess call.
In the readthedocs documentation see the section Development tools API 1.4.0.
pytest’s pytester is suitable for pytest plugin development.
Simulate command line
To simulate a command line call to phmdoctest from
within a Python script
phmdoctest.simulator offers the
it creates the –outfile in a temporary directory
optionally runs pytest on the outfile
pytest can return a JUnit XML report
useful during development to validate the command line and prevent use of a stale –outfile
Please see the Latest Development tools API section or
the docstring of the function
run_and_pytest() in the file
Pass pytest_options as a list of strings as shown below.
import phmdoctest.simulator command = "phmdoctest doc/example1.md --report --outfile temporary.py" simulator_status = phmdoctest.simulator.run_and_pytest( well_formed_command=command, pytest_options=["--doctest-modules", "-v"] ) assert simulator_status.runner_status.exit_code == 0 assert simulator_status.pytest_exit_code == 0
To read the Markdown file from the standard input stream. Use
Write the test file to a temporary directory so that it is always up to date.
In CI scripts the following shell command will create the temporary directory tmp in the tests folder on Windows, Linux, and macOS.
python -c "from pathlib import Path; d = Path('tests') / 'tmp'; d.mkdir(mode=0o700)"
It is easy to use –output by mistake instead of
If Python code block has no output, put assert statements in the code.
Use pytest option
--doctest-modulesto test the sessions.
Markdown indented code blocks (Spec section 4.4) are ignored.
simulator_status.runner_status.exit_code == 2 is the click command line usage error.
Since phmdoctest generates code, the input file should be from a trusted source.
An empty code block gets given the role
del-code. It is not tested.
Use special TEXT values FIRST, SECOND, LAST for the command line options
--teardownsince they only match one block.
The variable names
_phm_expected_strshould not be used in Markdown Python code blocks since they may be used in generated code.
Setup and teardown code blocks cannot have expected output.
To have pytest collect a code block with the label directive start the value with
--setup-doctestoption, names assigned by the setup code block are globally visible to the entire test suite. This is due to the scope of the pytest doctest_namespace fixture. Try using a separate pytest command to test just the phmdoctest test.
The module phmdoctest.fixture is imported at pytest time to support setup, teardown, share-names, and clear-names features.
The phmdoctest Markdown parser finds fenced code blocks enclosed by html
</details>tags. The tags may require a preceding and trailing blank line to render correctly. See example in tests/test_details.py.
Try redirecting phmdoctest standard output into PYPI Pygments to colorize the generated test file.
python -m phmdoctest project.md --outfile - | pygmentize
If the –outfile is written into a folder that pre-exists in the repository, consider adding the outfile name to .gitignore. If the outfile name later changes, the change will be needed in .gitignore too.
# Reserved for generated test file. tests/test_readme.py
Only put one of setup, teardown, share-names, or clear-names on a code block.
Only one block can be setup. Only one block can be teardown.
The setup or teardown block can’t have an expected output block.
Label directive does not generate a test case name on setup and teardown blocks.
Directives displayed in the
--reportstart with a dash like this:
Code generated by Python blocks with setup and teardown directives runs at the pytest fixture
Code generated by Python blocks with share-names and clear-names directives are collected and run by pytest like any other test case.
A malformed HTML comment ending is bad. Make sure it ends with both dashes like
-->. Running with
--reportwill expose that problem.
The setup, teardown, share-names, and clear-names directives have logging. To see the log messages, run pytest with the option:
There is no limit to number of blank lines after the directive HTML comment but before the fenced code block.
<!--phmdoctest-mark.xfail-->might be useful as an alternative to
<!--phmdoctest-mark.skip-->for failing examples.
<!--phmdoctest-mark.ATTRIBUTE-->will not be effective when used with
<!--phmdoctest-teardown-->because pytest marks can only be applied to tests. They have no effect on fixtures. Setup and teardown use fixtures.