2 Copyright (C) Daniel Stenberg, <daniel@haxx.se>, et al.
4 SPDX-License-Identifier: curl
11 See the "Requires to run" section for prerequisites.
13 In the root of the curl repository:
15 ./configure && make && make test
17 To run a specific set of tests (e.g. 303 and 410):
19 make test TFLAGS="303 410"
21 To run the tests faster, pass the -j (parallelism) flag:
23 make test TFLAGS="-j10"
25 "make test" builds the test suite support code and invokes the 'runtests.pl'
26 perl script to run all the tests. The value of `TFLAGS` is passed
27 directly to 'runtests.pl'.
29 When you run tests via make, the flags `-a` and `-s` are passed, meaning
30 to continue running tests even after one fails, and to emit short output.
32 If you'd like to not use those flags, you can run 'runtests.pl' directly.
33 You must `chdir` into the tests directory, then you can run it like so:
37 You must have run `make test` at least once first to build the support code.
39 To see what flags are available for runtests.pl, and what output it emits, run:
41 man ./tests/runtests.1
43 After a test fails, examine the tests/log directory for stdout, stderr, and
44 output from the servers used in the test.
48 - perl (and a unix-style shell)
49 - python (and a unix-style shell, for SMB and TELNET tests)
50 - python-impacket (for SMB tests)
51 - diff (when a test fails, a diff is shown)
52 - stunnel (for HTTPS and FTPS tests)
53 - OpenSSH or SunSSH (for SCP and SFTP tests)
54 - nghttpx (for HTTP/2 and HTTP/3 tests)
55 - nroff (for --manual tests)
56 - An available `en_US.UTF-8` locale
58 ### Installation of python-impacket
60 The Python-based test servers support both recent Python 2 and 3.
61 You can figure out your default Python interpreter with python -V
63 Please install python-impacket in the correct Python environment.
64 You can use pip or your OS' package manager to install 'impacket'.
66 On Debian/Ubuntu the package names are:
68 - Python 2: 'python-impacket'
69 - Python 3: 'python3-impacket'
71 On FreeBSD the package names are:
73 - Python 2: 'py27-impacket'
74 - Python 3: 'py37-impacket'
76 On any system where pip is available:
78 - Python 2: 'pip2 install impacket'
79 - Python 3: 'pip3 install impacket'
81 You may also need to manually install the Python package 'six'
82 as that may be a missing requirement for impacket on Python 3.
84 ### Port numbers used by test servers
86 All test servers run on "random" port numbers. All tests should be written
87 to use suitable variables instead of fixed port numbers so that test cases
88 continue to work independent on what port numbers the test servers actually
91 See [`FILEFORMAT`](FILEFORMAT.md) for the port number variables.
95 The test suite runs stand-alone servers on random ports to which it makes
96 requests. For SSL tests, it runs stunnel to handle encryption to the regular
97 servers. For SSH, it runs a standard OpenSSH server.
99 The listen port numbers for the test servers are picked randomly to allow
100 users to run multiple test cases concurrently and to not collide with other
101 existing services that might listen to ports on the machine.
103 The HTTP server supports listening on a Unix domain socket, the default
104 location is 'http.sock'.
106 For HTTP/2 and HTTP/3 testing an installed `nghttpx` is used. HTTP/3
107 tests check if nghttpx supports the protocol. To override the nghttpx
108 used, set the environment variable `NGHTTPX`. The default can also be
109 changed by specifying `--with-test-nghttpx=<path>` as argument to `configure`.
111 ### Shell startup scripts
113 Tests which use the ssh test server, SCP/SFTP tests, might be badly
114 influenced by the output of system wide or user specific shell startup
115 scripts, .bashrc, .profile, /etc/csh.cshrc, .login, /etc/bashrc, etc. which
116 output text messages or escape sequences on user login. When these shell
117 startup messages or escape sequences are output they might corrupt the
118 expected stream of data which flows to the sftp-server or from the ssh
119 client which can result in bad test behavior or even prevent the test server
122 If the test suite ssh or sftp server fails to start up and logs the message
123 'Received message too long' then you are certainly suffering the unwanted
124 output of a shell startup script. Locate, cleanup or adjust the shell
129 The test script will check that all allocated memory is freed properly IF
130 curl has been built with the `CURLDEBUG` define set. The script will
131 automatically detect if that is the case, and it will use the
132 `memanalyze.pl` script to analyze the memory debugging output.
134 Also, if you run tests on a machine where valgrind is found, the script will
135 use valgrind to run the test with (unless you use `-n`) to further verify
138 The `runtests.pl` `-t` option enables torture testing mode. It runs each
139 test many times and makes each different memory allocation fail on each
140 successive run. This tests the out of memory error handling code to ensure
141 that memory leaks do not occur even in those situations. It can help to
142 compile curl with `CPPFLAGS=-DMEMDEBUG_LOG_SYNC` when using this option, to
143 ensure that the memory log file is properly written even if curl crashes.
147 If a test case fails, you can conveniently get the script to invoke the
148 debugger (gdb) for you with the server running and the same command line
149 parameters that failed. Just invoke `runtests.pl <test number> -g` and then
150 just type 'run' in the debugger to perform the command through the debugger.
154 All logs are generated in the log/ subdirectory (it is emptied first in the
155 runtests.pl script). They remain in there after a test run.
159 A curl build with `--enable-debug` offers more verbose output in the logs.
160 This applies not only for test cases, but also when running it standalone
161 with `curl -v`. While a curl debug built is
162 ***not suitable for production***, it is often helpful in tracking down
165 Sometimes, one needs detailed logging of operations, but does not want
166 to drown in output. The newly introduced *connection filters* allows one to
167 dynamically increase log verbosity for a particular *filter type*. Example:
169 CURL_DEBUG=ssl curl -v https://curl.se
171 will make the `ssl` connection filter log more details. One may do that for
172 every filter type and also use a combination of names, separated by `,` or
175 CURL_DEBUG=ssl,http/2 curl -v https://curl.se
177 The order of filter type names is not relevant. Names used here are
178 case insensitive. Note that these names are implementation internals and
181 Some, likely stable names are `tcp`, `ssl`, `http/2`. For a current list,
182 one may search the sources for `struct Curl_cftype` definitions and find
183 the names there. Also, some filters are only available with certain build
188 All test cases are put in the `data/` subdirectory. Each test is stored in
189 the file named according to the test number.
191 See [`FILEFORMAT`](FILEFORMAT.md) for a description of the test case file
196 gcc provides a tool that can determine the code coverage figures for the
197 test suite. To use it, configure curl with `CFLAGS='-fprofile-arcs
198 -ftest-coverage -g -O0'`. Make sure you run the normal and torture tests to
199 get more full coverage, i.e. do:
204 The graphical tool `ggcov` can be used to browse the source and create
205 coverage reports on \*nix hosts:
209 The text mode tool `gcov` may also be used, but it doesn't handle object
210 files in more than one directory correctly.
214 The runtests.pl script provides some hooks to allow curl to be tested on a
215 machine where perl can not be run. The test framework in this case runs on
216 a workstation where perl is available, while curl itself is run on a remote
217 system using ssh or some other remote execution method. See the comments at
218 the beginning of runtests.pl for details.
220 ## Test case numbering
222 Test cases used to be numbered by category ranges, but the ranges filled
223 up. Subsets of tests can now be selected by passing keywords to the
224 runtests.pl script via the make `TFLAGS` variable.
226 New tests are added by finding a free number in `tests/data/Makefile.inc`.
230 Here's a quick description on writing test cases. We basically have three
231 kinds of tests: the ones that test the curl tool, the ones that build small
232 applications and test libcurl directly and the unit tests that test
233 individual (possibly internal) functions.
237 Each test has a master file that controls all the test data. What to read,
238 what the protocol exchange should look like, what exit code to expect and
239 what command line arguments to use etc.
241 These files are `tests/data/test[num]` where `[num]` is just a unique
242 identifier described above, and the XML-like file format of them is
243 described in the separate [`FILEFORMAT`](FILEFORMAT.md) document.
247 A test case that runs the curl tool and verifies that it gets the correct
248 data, it sends the correct data, it uses the correct protocol primitives
253 The libcurl tests are identical to the curl ones, except that they use a
254 specific and dedicated custom-built program to run instead of "curl". This
255 tool is built from source code placed in `tests/libtest` and if you want to
256 make a new libcurl test that is where you add your code.
260 Unit tests are placed in `tests/unit`. There's a tests/unit/README
261 describing the specific set of checks and macros that may be used when
262 writing tests that verify behaviors of specific individual functions.
264 The unit tests depend on curl being built with debug enabled.