From bug # 870366 http://bugzilla.mozilla.org/show_bug.cgi?id=870366 backend_file.write() seems to not be flushing the final written line to disk. writing an extra newline is the difference between this unit test passing or failing: elif isinstance(obj, JSPreferenceFiles): backend_file.write('PREF_JS_EXPORTS = %s\n' % obj.js_preference_files) backend_file.write('\n') # intentional: workaround until buffering problem fixed. With extra newline to force flushing: ===================================== TEST-PASS | /media/_Q/mozilla/bugs/870366/python/mozbuild/mozbuild/test/backend/test_recursivemake.py | test_xpcshell_master_manifest CALLING: write PREF_JS_EXPORTS = [aa/aa.js] CALLING: write PREF_JS_EXPORTS = [bb/bb.js] CALLING: write PREF_JS_EXPORTS = [cc/cc.js] CALLING: write PREF_JS_EXPORTS = [dd/dd.js] CALLING: write PREF_JS_EXPORTS = [valid_val/pref.js] ================================================== backend.mk ['PREF_JS_EXPORTS = aa/aa.js', 'PREF_JS_EXPORTS = bb/bb.js', 'PREF_JS_EXPORTS = cc/cc.js', 'PREF_JS_EXPORTS = dd/dd.js', 'PREF_JS_EXPORTS = valid_val/pref.js'] ================================================== # Without backend_file.write('\n') ================================== [870366] cat ~/bad TEST-UNEXPECTED-FAIL | /media/_Q/mozilla/bugs/870366/python/mozbuild/mozbuild/test/backend/test_recursivemake.py | line 243, test_js_preference_files: Lists differ: ['PREF_JS_EXPORTS = aa/aa.js',... != ['PREF_JS_EXPORTS = aa/aa.js',... Second list contains 1 additional elements. First extra element 4: PREF_JS_EXPORTS = valid_val/pref.js ['PREF_JS_EXPORTS = aa/aa.js', 'PREF_JS_EXPORTS = bb/bb.js', 'PREF_JS_EXPORTS = cc/cc.js', - 'PREF_JS_EXPORTS = dd/dd.js'] ? ^ + 'PREF_JS_EXPORTS = dd/dd.js', ? ^ + 'PREF_JS_EXPORTS = valid_val/pref.js'] TEST-PASS | /media/_Q/mozilla/bugs/870366/python/mozbuild/mozbuild/test/backend/test_recursivemake.py | test_makefile_conversion [snip] TEST-PASS | /media/_Q/mozilla/bugs/870366/python/mozbuild/mozbuild/test/backend/test_recursivemake.py | test_xpcshell_master_manifest CALLING: write PREF_JS_EXPORTS = [aa/aa.js] CALLING: write PREF_JS_EXPORTS = [bb/bb.js] CALLING: write PREF_JS_EXPORTS = [cc/cc.js] CALLING: write PREF_JS_EXPORTS = [dd/dd.js] CALLING: write PREF_JS_EXPORTS = [valid_val/pref.js] ================================================== backend.mk ['PREF_JS_EXPORTS = aa/aa.js', 'PREF_JS_EXPORTS = bb/bb.js', 'PREF_JS_EXPORTS = cc/cc.js', 'PREF_JS_EXPORTS = dd/dd.js'] ================================================== gmake: *** [run-/media/_Q/mozilla/bugs/870366/python/mozbuild/mozbuild/test/backend/test_recursivemake.py] Error 1

I suspect this is just the test code not reading the file properly. The files should all be closed by the time the test code opens and reads them. I highly doubt there is a buffering issue.

I strongly feel this was the result of user error. If we had a buffering problem writing backend.mk files, we'd see all kinds of random failures. We haven't.