71778 2011-11-08 00:50:25 -0800 [Qt] Incremental build is flakey on Windows 2011-11-14 04:50:54 -0800 1 1 1 Unclassified WebKit Tools / Tests 528+ (Nightly build) All All RESOLVED FIXED Qt, QtTriaged P1 Blocker --- 71222 1 ossy vestbo ossy oldest_to_newest 497816 0 ossy 2011-11-08 00:50:25 -0800 It seems fails are absolutely unrelated to commits. I think the dependency tree is broken in the Makefile, and we get flakey fails because of we build with make -j4. fail1: (r99437-r99438) make[3]: *** No rule to make target `compiler_cssvalues_make_all', needed by `first'. Stop. make[2]: *** [sub-DerivedSources-pri] Error 2 fail2: (r99455) make[3]: *** No rule to make target `compiler_inspectorValidate_make_all', needed by `first'. Stop. make[2]: *** [sub-DerivedSources-pri] Error 2 fail3: (r99463) make[3]: *** No rule to make target `compiler_cssvalues_make_all', needed by `first'. Stop. make[3]: Entering directory `/ramdisk/qt-windows-32bit-release/build/WebKitBuild/Release/Source/WebCore' make[3]: Leaving directory `/ramdisk/qt-windows-32bit-release/build/WebKitBuild/Release/Source/WebCore' make[2]: *** [sub-DerivedSources-pri] Error 2 fail4: (r99465) make[3]: *** No rule to make target `compiler_keywordlut_make_all', needed by `first'. Stop. make[2]: *** [sub-DerivedSources-pri] Error 2 make[3]: *** No rule to make target `compiler_inspectorValidate_make_all', needed by `first'. Stop. make[2]: *** [sub-DerivedSources-pri] Error 2 fail4: (r99515) make[3]: *** No rule to make target `compiler_keywordlut_make_all', needed by `first'. Stop. make[2]: *** [sub-DerivedSources-pri] Error 2 497857 1 ossy 2011-11-08 01:43:48 -0800 One more fail because of this bug: (r99540) make[3]: *** No rule to make target `compiler_keywordlut_make_all', needed by make[2]: *** [sub-DerivedSources-pri] Error 2 499124 2 vestbo 2011-11-09 10:37:26 -0800 What's likely happening is that make qmake triggers qmake foo.pro -o Makefile.foo, and somehow make does not understand that it needs to wait for that to finish before parallelizing running make -f Makefile.foo Clear case: http://build.webkit.org/builders/Qt%20Windows%2032-bit%20Debug/builds/29880/steps/compile-webkit/logs/stdio In theory the makefile rules are set up so that make should be able to realize this, and it seems to work everywhere else, except on the mingw bots. 500216 3 vestbo 2011-11-10 13:32:25 -0800 I know what the issue is now. The rules that were supposed to make "qmake derived, make derived, qmake traget" sequential was under an unix ifdef, as it depends on gnumake, but this obviously failed for mingw. I'll make a patch tomorrow. 501806 4 vestbo 2011-11-14 04:50:54 -0800 Fixed in r99949