Le 20 juil. 2012 à 10:53, Hans Aberg a écrit : >>> On 19 Jul 2012, at 16:37, Akim Demaille wrote: >>> >>>> Bison 2.6 is born, the parents are well, and happy. >>> >>> I ran this on Mac OS X 10.7.4 where using the system compiler clang >>> (llvm-gcc whic is pointe to by /usr/bin/gcc is now deprecated) supplied by >>> Xcode 4.3.3, and I go the strange warning below by the following command >>> sequence: >>> >>> Build without the source tree using >>> $ tar -xf bison-2.6.tar.xz >>> $ mkdir build-2.6 >>> $ cd build-2.6/ >>> $ export CC=/usr/bin/clang CXX=/usr/bin/clang++ >>> $ ../bison-2.6/configure >>> $ make >>> >>> Then start a new shell where CC and CXX are not defined, in which 'make >>> check' produces the first warnings below. >>> >>> After that, in the original shell CC are defined, 'make check' produces the >>> second warnings and ending with '0 tests were successful. >> >> I cannot reproduce it following your script. BTW, I believe the >> mkdir is useless, and ./configure would suffice: you seem to build >> in place. > > It is required when building GCC, and the developers get quite upset when > people still try to use ./configure. So I wanted to try it out, even though > it is not a requirement for Bison.
I am all in favor on making a VPATH build. My point is different: you *seem* to be making a VPATH build, but you are not: $ tar -xf bison-2.6.tar.xz creates bison-2.6 $ mkdir build-2.6 probably issues an error such as mkdir: bison-2.6: File exists $ cd build-2.6/ enters the directory where the sources are. If you want to make a VPATH build, you must use some other name than bison-2.6 for your build tree. >> What's the result of "wc tests/testsuite"? I have >> >> 164175 551986 5095566 tests/testsuite > > In this test, I got: > $ wc tests/testsuite.log > 517 4004 34775 tests/testsuite.log I really mean testsuite, not testsuite.log. But, as you report below, the issue is just that two make checks were stepping on each other's toes. > Returning to the stuff above: on Mac OS X, in Terminal, after 'make' and > 'make pdf', I created a new Tab before doing 'make check', which creates a > new shell from scratch, though set to the same directory. But then the CC and > CXX variables are undefined in this new shell. I then saw the first set of > warnings I posted. Then I started another 'make check' in the original > directory, with CC and CXX, that did indeed run concurrently with the first > one, but my assumption was that because of the delay, it should not interfere. Never run two make check concurrently. Maybe Autotest should provide some explicit warnings about this, indeed.
