conversions CustomConversionsTest skip fp_sat_xxx

[wenhu1024]

related #2427

[CLAassistant]

All committers have signed the CLA.

[wenhu1024]

Hi @shajder, since you’ve contributed to these files before #1719 , would you mind reviewing this change? Appreciate any comments or suggestions you may have!

[shajder]

LGTM

[bashbaug]

I think this is fine, but when I run the conversions test suite I don't see any difference in the number of tests that are run.

I'm only running the "wimpy" tests, and only on scalars so the tests run faster, but I don't think this should matter? For completeness my command line is:

./test_conformance/conversions/test_conversions -w -1

If the same tests are run before and after this change then what does it do exactly? Thanks!

[wenhu1024]

If run ./test_conversions fp_sat_xxx , the test result is always pass.

for example:

./test_conversions float_sat_int


conversions...
conversions passed
PASSED sub-test.
PASSED test.

[svenvh]

If the same tests are run before and after this change then what does it do exactly?

The current code structure has the running logic sort of duplicated: one version (ConversionsTest) for running all conversions known to the CTS and another version (CustomConversionsTest) for running specific tests specified on the command line. This patch only affects the filtering for the latter. You can observe the effect of this patch by running e.g. test_conversions float_sat_char before and after the patch.

I'm not sure if silently ignoring the invalid conversion is desirable here, though. Wouldn't it be better to give an error if the user specifies an invalid conversion?