Control: tags -1 +confirmed Hi,
I'm looking for advice here on what would be the best move. On Thu, Jul 9, 2020 at 1:36 PM Lucas Nussbaum <lu...@debian.org> wrote: > Source: icu > Version: 67.1-2 > Severity: serious > Justification: FTBFS on amd64 > During a rebuild of all packages in sid, your package failed to build > on amd64. [...] > > dh_auto_test: error: cd source && make -j4 check VERBOSE=1 returned exit > > code 2 This is confirmed and happens due to a regression in make-dfsg while fixing #963335 [1]. File descriptors are no longer passed to forked makes when building recursively or building parallel. Those can't output to stdout and the output of the first make (entering the top level source directory) is not flushed. Normally it's not a problem, but an ICU self-test from 'make check' executes 'make -f source/config/Makefile.inc SELFCHECK=1 selfcheck' directly. With make-dfsg 4.3-3 the previous makes output already flushed and the mentioned test returns 'passed' as supposed to. But with make-dfsg 4.3-4 the output is delayed and only flushed with this test which then will have the output like this: "make[3]: Entering directory 'icu-67.1/source' passed make[3]: Leaving directory 'icu-67.1/source'". That is, it gets polluted by the output of the main make. I can workaround this by setting no parallel to the tests, which in turn will run slower on multicore processors but will pass. Then Manoj can revert the mentioned bad fix of make-dfsg while upstream will work on a better one. Another way is to report this misbehaviour to GNU to fix. Which way to go first? Regards, Laszlo/GCS [1] https://bugs.debian.org/963335
