[lfs-dev] gcc 5.1.0 test suite time out errors.

Armin K. krejzi at email.com
Fri Jun 12 11:31:01 PDT 2015


On 12.6.2015 19:17, Bruce Dubbs wrote:
> Armin K. wrote:
>> On 12.6.2015 17:12, Bruce Dubbs wrote:
>>> Laurence Dawson wrote:
>>>
>>>>                  === g++ Summary ===
>>>>
>>>> # of expected passes            93236
>>>> # of unexpected successes       2
>>>> # of expected failures          339
>>>> # of unsupported tests          3645
>>>> /sources/gcc-build/gcc/testsuite/g++/../../xg++  version 5.1.0 (GCC)
>>>>
>>>>                  === gcc tests ===
>>>>
>>>>
>>>> Running target unix
>>>> FAIL: c-c++-common/goacc/acc_on_device-2.c scan-rtl-dump-times expand
>>>> "\\\\(call [^\\\\n]* acc_on_device" 0
>>>>
>>>>                  === gcc Summary ===
>>>>
>>>> # of expected passes            113926
>>>> # of unexpected failures        1
>>>> # of expected failures          259
>>>> # of unsupported tests          1807
>>>> /sources/gcc-build/gcc/xgcc  version 5.1.0 (GCC)
>>>
>>> If your only failure is that one, then you are doing pretty well.  One
>>> out of about 200,000 is not very much.
>>>
>>> By the way, google is your friend:
>>>
>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66332
>
>> Contains a fix that "fixes a test case" (one more reason for me to find
>> tests useless) to mark it as "expected failure" (and another reason).
>
> If you get a thousand failures, then we have probably done something
> wrong.  An individual failure is probably only useful to the upstream
> developers, but indeed an individual failure is as likely to be in the
> test (or the test harness) as what is being tested.
>
>    -- Bruce
>

Not really. If you get thousands of failures (in case of GCC), it means 
a problem with your hardware, not system (in case of the OP, where 
insufficient hw was assigned to a VM). It can also mean a missing 
dependency, not mounted virtual file systems, etc. You can get all 
passes and still a screwed up toolchain (linking to /tools, that is - 
been there seen that). Sanity check and bootstrap are enough (first one 
for verifying the link to correct libraries and second one for verifying 
if a compiler builds anything).


More information about the lfs-dev mailing list