Re: [Qemu-devel] [PATCH] qemu-iotests: add a "how to" to ./README

[Kevin Wolf]

Am 21.07.2017 um 11:34 hat Stefan Hajnoczi geschrieben:
> There is not much getting started documentation for qemu-iotests.  This
> patch explains how to create a new test and covers the overall testing
> approach.
> 
> Cc: Ishani Chugh <address@hidden>
> Signed-off-by: Stefan Hajnoczi <address@hidden>
> ---
>  tests/qemu-iotests/README | 83 
> +++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 83 insertions(+)
> 
> diff --git a/tests/qemu-iotests/README b/tests/qemu-iotests/README
> index 6079b40..8259b9f 100644
> --- a/tests/qemu-iotests/README
> +++ b/tests/qemu-iotests/README
> @@ -14,8 +14,91 @@ Just run ./check to run all tests for the raw image 
> format, or ./check
>  -qcow2 to test the qcow2 image format.  The output of ./check -h explains
>  additional options to test further image formats or I/O methods.
>  
> +* Testing approach
> +
> +Each test is an executable file (usually a bash script) that is run by the
> +./check test harness.  Standard out and standard error are captured to an
> +output file.  If the output file differs from the "golden master" output file
> +for the test then it fails.
> +
> +Tests are simply a sequence of commands that produce output and the test 
> itself
> +does not judge whether it passed or failed.  If you find yourself writing
> +checks to determine success or failure then you should rethink the test and
> +rely on output diffing instead.
> +
> +** Filtering volatile output
> +
> +When output contains absolute file paths, timestamps, process IDs, hostnames,
> +or other volatile strings, the diff against golden master output will fail.
> +Such output must be filtered to replace volatile strings with fixed
> +placeholders.
> +
> +For example, the path to the temporary working directory changes between test
> +runs so it must be filtered:
> +
> +  sed -e "s#$TEST_DIR/#TEST_DIR/#g"
> +
> +Commonly needed filters are available in ./common.filter.
> +
> +** Python tests
> +
> +Most tests are implemented in bash but it is difficult to interact with the 
> QMP
> +monitor.  A Python module called 'iotests' is available for tests that 
> require
> +JSON and interacting with QEMU.
> +
> +* How to create a test
> +
> +1. Choose an unused test number
> +
> +Tests are identified by a unique number.  Look for the highest test case 
> number
> +by looking at the test files.  Then search the address@hidden mailing
> +list to check if anyone has already sent patches using the next available
> +number.  You may need to increment the number a few times to reach an unused
> +number.
> +
> +2. Create the test file
> +
> +Copy an existing test (one that most closely resembles what you wish to test)
> +to the new test number:
> +
> +  cp 001 <test-number>
> +
> +3. Assign groups to the test
> +
> +Add your test to the ./group file.  This file is the index of tests and 
> assigns
> +them to functional groups like "rw" for read-write tests.  Most tests belong 
> to
> +the "rw" and "auto" groups.  "auto" means the test runs when ./check is 
> invoked
> +without a -g argument.
> +
> +Consider adding your test to the "quick" group if it executes quickly (<1s).
> +This group is run by "make check-block" and is often included as part of 
> build
> +tests in continuous integration systems.
> +
> +4. Write the test
> +
> +Edit the test script.  Look at existing tests for examples.
> +
> +5. Generate the golden master file
> +
> +Run your test with "./check <test-number>".  You may need to pass additional
> +options to use an image format or protocol.

./check refuses to even run a test if the reference output is missing.
So in practice you need a 'touch <test-number>.out' first.

> +The test will fail because there is no golden master yet.  Inspect the output
> +that your test generated with "cat <test-number>.out.bad".
> +
> +Verify that the output is as expected and contains no volatile strings like
> +timestamps.  You may need to add filters to your test to remove volatile
> +strings.
> +
> +Once you are happy with the test output it can be used as the golden master
> +with "mv <test-number>.out.bad <test-number>.out".  Rerun the test to verify
> +that it passes.
> +
> +Congratulations, you've created a new test!

Looks good otherwise.

Kevin