Allow building as a subdir

[robert-shade]

Allow building as a subdirectory (like with cmake FetchContent)

Fixes #5688

Signed-off-by: Robert Shade [email protected]

Status

READ

Requires Backporting

NO

Migrations

If there is any API change, what's the incentive and logic for it.

NO

[daverodgman]

Tested by making a project that refers to Mbed TLS in this way - as a non-CMake expert this PR seems to address the problem (and doesn't introduce new cmake version requirements).

[daverodgman]

Please could you add a Changelog

[tom-cosgrove-arm]

Hi, every commit needs a "Signed-off-by" line, including the ChangeLog one, so would you mind updating that commit?

[daverodgman]

It looks like this is failing on the FreeBSD CI (but only for internal CI, not on OpenCI). The log shows:

`******************************************************************

• test_cmake_out_of_source: test: cmake 'out-of-source' build

• Sat May 21 17:14:57 UTC 2022

---

Running tests...

Test project /tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/mbedtls_out_of_source_build

  Start  1: aes.cbc-suite

1/95 Test #1: aes.cbc-suite .............................. Passed 0.00 sec

  Start  2: aes.cfb-suite

2/95 Test #2: aes.cfb-suite .............................. Passed 0.00 sec

  Start  3: aes.ecb-suite

3/95 Test #3: aes.ecb-suite .............................. Passed 0.00 sec

  Start  4: aes.ofb-suite

............ [redacted a load of passing tests for brevity] ...........

93/95 Test #93: version-suite .............................. Passed 0.00 sec

  Start 94: x509parse-suite

94/95 Test #94: x509parse-suite ............................ Passed 0.44 sec

  Start 95: x509write-suite

95/95 Test #95: x509write-suite ............................ Passed 0.30 sec

100% tests passed, 0 tests failed out of 95

Total Test time (real) = 24.06 sec

^^^^test_cmake_out_of_source: test: cmake 'out-of-source' build: ./tests/ssl-opt.sh -f 'Default' > ssl-opt.out 2> ssl-opt.err -> 1^^^^

Default ................................................................ PASS

make[1]: Entering directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/library'

make[1]: Leaving directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/library'

make[1]: Entering directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/programs'

make[2]: Entering directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/programs/fuzz'

make[2]: Leaving directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/programs/fuzz'

make[1]: Leaving directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/programs'

make[1]: Entering directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/tests'

make[1]: Leaving directory '/tmp/workspace/mbed-tls-pr-head_PR-5837-head/src/tests'

---

• Done, cleaning up `

[gilles-peskine-arm]

The relevant part of the log is

^^^^test_cmake_out_of_source: test: cmake 'out-of-source' build: ./tests/ssl-opt.sh -f 'Default' > ssl-opt.out 2> ssl-opt.err -> 1^^^^
Default ................................................................ PASS

which corresponds to

    ./tests/ssl-opt.sh -f 'Default' >ssl-opt.out 2>ssl-opt.err
    grep PASS ssl-opt.out
    cat ssl-opt.err >&2

This is a sanity check for running SSL tests after an out-of-source build and the expected output is

Default ................................................................ PASS
Default, DTLS .......................................................... PASS
PASSED (2 / 2 tests (0 skipped))

The only expected reason why “Default” would pass but not “Default, DTLS” is a problem with udp_proxy (which we systematically use for DTLS tests in ssl-opt.sh). But there's no plausible reason why there would be a problem with the proxy, especially one that's specific to out-of-tree builds with a specific set of tools.

And pr-merge is passing.

So this looks like a mysterious one-off.

[gilles-peskine-arm]

Please indicate clearly near the beginning that this is about CMake. For readers that don't have CMake in mind, the first sentence looks quite mysterious.

[gilles-peskine-arm]

Done in #5905.

[daverodgman]

So this looks like a mysterious one-off.

I've re-triggered CI to see if it works this time

[daverodgman]

Seems to be a one-off, CI is happy now.

[daverodgman]

Since we don't have an update on the changelog wording for two weeks, I'm going to merge as-is. We can always tidy up the changelog later - I will make a note on the 3.2 release task.

[pfeatherstone]

Great stuff! Shame we now need Jinja2 to build mbedtls.