On Oct 21, 2014, at 7:25 PM, Dag-Erling Smørgrav <[email protected]> wrote:
> Warner Losh <[email protected]> writes: >> Author: imp >> Date: Tue Oct 21 20:29:42 2014 >> New Revision: 273417 >> URL: https://svnweb.freebsd.org/changeset/base/273417 >> >> Log: >> You aren't allowed to test WITH_xxx or WITHOUT_xxx here, so remove it. >> Even if you were allowed to test for it, the test makes no sense as it >> always results in adding -DWITH_ATF unless WITH_ATF was already >> defined. But if MK_ATF != no, then we know it was defined. This, in >> turn, caused tools/build/options/makemake always think WITH_ATF is the >> default, which removed control of that from sys.conf.mk. >> >> To get the intent of the deleted comment, another mechanism is >> required, assuming that the intent of that comment is desirable. > > This broke the paralell build. It prevents building lib/atf entirely in > stage 4.3, so the programs in libexec/atf have nothing to link against > unless lib/atf just happens to get built before libexec/atf. That > almost never happens: lib and libexec are built in paralell, and lib/atf > is one of the last items in lib whereas libexec/atf is one of the first > items in libexec. > > (on a side note, libatf doesn't need to be in _prebuild_libs, since no > other library depend on it) Yea, it was AFU. I’ll fix this. I think we need to have a MK_TESTS_SUPPORT that builds the libatf stuff when yes, and omits it when no, since we don’t want the tests building when we’re building the 4.3 stage. Warner > DES > -- > Dag-Erling Smørgrav - [email protected]
signature.asc
Description: Message signed with OpenPGP using GPGMail
