curl/tests/runtests.md

---

c: Copyright (C) Daniel Stenberg, , et al. SPDX-License-Identifier: curl Title: runtests.pl Section: 1 Source: runtests See-also:

• runtests.pl Added-in: 7.5 ---

NAME

runtests.pl - run one or more test cases

SYNOPSIS

runtests.pl [options] [tests]

DESCRIPTION

runtests.pl runs one, several or all the existing test cases in curl's test suite. It is often called from the root Makefile of the curl package with 'make test'.

TESTS

Specify which test(s) to run by specifying test numbers or keywords.

If no test number or keyword is given, all existing tests that the script can find are considered for running. You can specify single test cases to run by specifying test numbers space-separated, like 1 3 5 7 11, and you can specify a range of tests like 45 to 67.

Specify tests to not run with a leading exclamation point, like !66, which runs all available tests except number 66.

Prefix a test number with a tilde (~) to still run it, but ignore the results.

It is also possible to specify tests based on a keyword describing the test(s) to run, like FTPS. The keywords are strings used in the individual tests.

You can also specify keywords with a leading exclamation point and the keyword or phrase, like "!HTTP NTLM auth" to run all tests except those using this keyword. Remember that the exclamation marks and spaces need to be quoted somehow when entered at many command shells.

Prefix a keyword with a tilde (~) to still run it, but ignore the results.

OUTPUT

When running without -s (short output), for instance when running runtests.pl directly rather than via make, each test emits a pair of lines like this:

Test 0045...[simple HTTP Location: without protocol in initial URL]
--pd---e-v- OK (45  out of 1427, remaining: 16:08, took 6.188s, duration: 00:31)

the first line contains the test number and a description. On the second line, the characters at the beginning are flags indicating which aspects of curl's behavior were checked by the test:

s stdout
r stderr
p protocol
d data
u upload
P proxy
o output
e exit code
m memory
v valgrind
E the test was run event-based

The remainder of the second line contains the test result, current test sequence, total number of tests to be run and an estimated amount of time to complete the test run.

OPTIONS

-a

Continue running the rest of the test cases even if one test fails. By default, the test script stops as soon as an error is detected.

-ac \<curl\>

Provide a path to a curl binary to talk to APIs (currently only CI test APIs).

-am

Display test results in automake style output (PASS/FAIL: [number] [name]).

-bundle

Run tests via bundled test binaries. Bundled test binaries contain all tests, and the test name passed as the first argument selects which test run.

-c\<curl\>

Provide a path to a custom curl binary to run the tests with. Default is the curl executable in the build tree.

-d

Enable protocol debug: have the servers display protocol output. If used in conjunction with parallel testing, it is difficult to associate the logs with the specific test being run.

-E \<exclude_file\>

Load the exclude_file with additional reasons why certain tests should be skipped. Useful when testing with external HTTP proxies in which case some of the tests are not appropriate.

The file contains colon-delimited lines. The first field contains the type of exclusion, the second field contains a pattern and the final field contains the reason why matching tests should be skipped. The exclusion types are keyword, test, and tool.

-e or --test-event

Run the test event-based (if possible). This makes runtests invoke curl with --test-event option. This option only works if both curl and libcurl were built debug-enabled.

-f

Force the test to run even if mentioned in DISABLED.

-g

Run the given test(s) with gdb. This is best used on a single test case and curl built --disable-shared. This then fires up gdb with command line set to run the specified test case. Simply (set a break-point and) type 'run' to start.

-gl

Run the given test(s) with lldb. This is best used on a single test case and curl built --disable-shared. This then fires up lldb with command line set to run the specified test case. Simply (set a break-point and) type 'run' to start.

-gw

Run the given test(s) with gdb as a windowed application.

-h, --help

Displays a help text about this program's command line options.

-j[num]

Spawn the given number of processes to run tests in. This defaults to 0 to run tests serially within a single process. Using a number greater than one allows multiple tests to run in parallel, speeding up a test run. The optimum number is dependent on the system and set of tests to run, but 7 times the number of CPU cores is a good figure to start with, or 1.3 times if Valgrind is in use, or 5 times for torture tests. Enabling parallel tests is not recommended in conjunction with the -g option.

-k

Keep output and log files in log/ after a test run, even if no error was detected. Useful for debugging.

-L \<file\>

Load and execute the specified file which should contain perl code. This option allows one to change runtests.pl behavior by overwriting functions and variables and is useful when testing external proxies using curl's regression test suite.

-l

Lists all test case names.

-n

Disable the check for and use of valgrind.

--no-debuginfod

Delete the DEBUGINFOD_URLS variable if that is defined. Makes valgrind, gdb etc not able to use this functionality.

-o \<variablename=value\>

Overwrite the specified internal variable with value. Useful to change variables that did not get a dedicated flag to change them. Check the source to see which variables are available.

-P \<proxy\>

Use the specified HTTP proxy when executing tests, even if the tests themselves do not specify a proxy. This option allows one to test external proxies using curl's regression test suite.

-p

Prints out all files in the log directory to stdout when a test case fails. Practical when used in the automated and distributed tests since then the people checking the failures and the reasons for them might not have physical access to the machine and logs.

-R

Run the tests in a scrambled, or randomized, order instead of sequentially.

The random seed initially set for this is fixed per month and can be set with --seed.

-r

Display run time statistics. (Requires the Perl Time::HiRes module)

-rf

Display full run time statistics. (Requires the Perl Time::HiRes module)

-rm

Force removal of files by killing locking processes. (Windows only, requires the Sysinternals handle[64].exe to be on PATH)

--repeat=[num]

This repeats the given set of test numbers this many times. If no test numbers are given, it repeats ALL tests this many times. It adds the new repeated sequence at the end of the initially given one.

If -R option is also used, the scrambling is done after the repeats have extended the test sequence.

-s

Shorter output. Speaks less than default.

--seed=[num]

When using --shallow or -R that randomize certain aspects of the behavior, this option can set the initial seed. If not set, the random seed is set based on the currently set local year and month and the first line of the "curl -V" output.

--shallow=[num]

Used together with -t. This limits the number of tests to fail in torture mode to no more than num per test case. If this reduces the amount, the script randomly discards entries to fail until the amount is num.

The random seed initially set for this is fixed per month and can be set with --seed.

-t[num]

Selects a torture test for the given tests. This makes runtests.pl first run the tests once and count the number of memory allocations made. It then reruns the test that number of times, each time forcing one of the allocations to fail until all allocations have been tested. By setting num you can force the allocation with that number to be set to fail at once instead of looping through everyone, which is handy when debugging and then often in combination with -g.

--test-duphandle

Passes the --test-duphandle option to curl when invoked. This command line option only exists in debug builds and runs curl normally, but duplicates the easy handle before the transfer and use the duplicate instead of the original handle. This verifies that the duplicate works exactly as good as the original handle.

Because of how the curl tool uses a share object to store and keep some data, not everything is however perfectly copied in the duplicate. In particular HSTS data is not. A specific test case can be set to avoid using --test-duphandle by disabling it on a per test basis.

-u

Error instead of warning on server unexpectedly alive.

-v

Enable verbose output. Speaks more than by default. If used in conjunction with parallel testing, it is difficult to associate the logs with the specific test being run.

-vc \<curl\>

Provide a path to a custom curl binary to run when verifying that the servers running are indeed our test servers. Default is the curl executable in the build tree.

RUNNING TESTS

Many tests have conditions that must be met before the test case can run fine. They could depend on built-in features in libcurl or features present in the operating system or even in third-party libraries that curl may or may not use.

The test script checks most of these by itself to determine when it is safe to attempt to run each test. Those which cannot be run due to failed requirements are simply skipped and listed at the completion of all test cases. In some unusual configurations, the test script cannot make the correct determination for all tests. In these cases, the problematic tests can be skipped using the "!keyword" skip feature documented earlier.

WRITING TESTS

The simplest way to write test cases is to start with a similar existing test, save it with a new number and then adjust it to fit. There is an attempt to document the test case file format in tests/FILEFORMAT.md.