Hi Simon,
> 1) The self-test requires network connectivity.
>
> For now, I think we can ignore 1), but it might make sense to consider
> a non-networked self test in the future.
Yes. The test currently fails after 7 minutes when there is no internet
connection:
PASS: test-arpa_inet
Finding www.gnu.org service http...
res -3: Temporary failure in name resolution
Finding www.ibm.com service https...
res -3: Temporary failure in name resolution
Finding microsoft.com service http...
res -3: Temporary failure in name resolution
Finding google.org service ldap...
res -3: Temporary failure in name resolution
FAIL: test-getaddrinfo
PASS: test-netinet_in
PASS: test-stdbool
PASS: test-stdio
PASS: test-string
PASS: test-sys_socket
===================
1 of 7 tests failed
===================
make[4]: *** [check-TESTS] Fehler 1
real 7m29.216s
user 0m0.811s
sys 0m0.136s
> For the second, I kind of prefer having self-tests be noisy, even when
> they succeed.
I think it's ok for them to be noisy if they find something suspicious,
or when the test has to be skipped. (The "Skipping test because ..."
messages already showed me a few portability problems that I wouldn't
have seen otherwise.)
> why do you consider noisy self tests a problem?
Someday, I would like to use "gnulib-tool --with-tests" in the gettext builds.
So that people notice and inform me when some internals that gettext relies
upon is unreliable on some platform. There will be ca. 100 tests or so, and
if many of them output more than 1 or 2 lines when they succeed,
- the output will be hard to read and understand (yes, I do have problems
reading the coreutils "make check" output),
- the communication with the user who reports test failures will be harder.
Bruno
