Le 06/04/2013 10:42, Pierre Labastie a écrit : >> Now, I have a new test failure in pmap: >> FAIL: pmap extra extended output (footer) >> :-( >> Needs more testing, but the behavior seems to be different when running >> in jhalfs and by hand... I think I have got the picture. jhalfs is not at fault, except that the tests suppose some timing conditions, which are not always met.
First, it is possible to use the dejagnu debugging tools: I changed the make check in the jhalfs script to: make RUNTESTFLAGS='-a -de -v -v' DEJATOOL=pmap check. This has the effect that a file dbg.log is created in testsuite. That file contains debugging info on what is matched and how, when using the expect command. The principle of spawn <command>...expect <regexp> commands is that the output of the command run by 'spawn' is monitored against regexp. As soon as a match is found, the following expect command is executed, so that different parts of the output can be monitored in order. In the failed test, we have (simplifying a bit): spawn pmap -X $pid expect "$pid:\\s+\\S+\[^\\r\]+\\s+\\s+\(?:\[A-Z\]\[a-z\]+ +\)+" expect ".+" expect "\[= \]+" Now the first regexp is matched correctly by the first two lines of the output, but then, the next one is matched by any non empty string following. Since the default matching is greedy, it could happen that the whole output matches, and that nothing is left for the last regexp. But that is not always true: expect matches in real time whatever comes from the spawned command. So it may also happen that the second regexp is matched against a partial chunk of output, and that the next chunk of output matches the last regexp. For some reason, the pmap -X test seems to send its whole output at once, so there is nothing left for the last regexp and the test fails. the pmap -XX test (which has the same expect's) sends its output in several chunks, so that something is left for the last regexp and the test passes. Actually, the last regexp might not be what upstream intended, since it may be matched by any sequence of spaces with no '='. So, at the end, it is a problem of badly crafted regexp and race conditions (and I am happy to say that jhalfs has nothing to do with it)... To have the tests pass, we could try the sed: sed -i 's|"\.+"|"\[^=\]+"|' testsuite/pmap.test/pmap.exp But we could as well disable the tests and wait for upstream to fix them... Pierre -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
