Configuration detection can fail in ssl-opt.sh

[gilles-peskine-arm]

Depending on the configuration, ssl-opt.sh may skip some tests that it should run or run some tests that will fail. There are in fact errors if you run test-ref-configs.pl, which turn out to be benign in this case.

Here's an excerpt from test-ref-configs.pl:

******************************************
* Testing configuration: config-thread.h
******************************************
…
running ssl-opt.sh -f ECJPAKE.*nolog
ECJPAKE: working, DTLS, nolog .......................................... PASS
tests/ssl-opt.sh: 176: [: Illegal number: 
tests/ssl-opt.sh: 176: [: Illegal number: 
tests/ssl-opt.sh: 183: [: Illegal number: 
tests/ssl-opt.sh: 176: [: Illegal number: 
tests/ssl-opt.sh: 183: [: Illegal number: 
------------------------------------------------------------------------
PASSED (1 / 1 tests (0 skipped))

The problem turns out to be the detection of the value of MBEDTLS_SSL_DTLS_MAX_BUFFERING. If include/mbedtls/config.h does not contain a line with #define MBEDTLS_SSL_DTLS_MAX_BUFFERING … or //#define MBEDTLS_SSL_DTLS_MAX_BUFFERING …, which happens to be the case for configs/config-thread.h, then the function get_config_value_or_default produces empty output and this leads to the error messages above from requires_config_value_at_least and requires_config_value_at_most.

It is legitimate for config.h not to define MBEDTLS_SSL_DTLS_MAX_BUFFERING, since ssl.h will supply a fallback value. I think the whole mechanism to detect the value of MBEDTLS_SSL_DTLS_MAX_BUFFERING should be changed. We could make a program in programs/tests or programs/util that just prints out the value: this would be a lot more robust than parsing header files in sh.

[hanno-becker]

I think adding an example program printing numerous configuration values, as @gilles-peskine-arm suggested, should be straightforward and very helpful for testing.

[simonbutcher]

Isn't this just a duplicate of #2029? Am I missing something?

[gilles-peskine-arm]

No, the issue I'm pointing out here isn't specific to an OS. I think this and #2029 should be fixed together, #2029 could also be fixed by sticking to portable use of sed, and this wouldn't fix #2030.

[ciarmcom]

ARM Internal Ref: IOTSSL-2555