Multiple threads and verbose output lead to intermingled output

[sfiruch]

When I run clasp 3.3.3 with a single thread, everything works as expected:

clasp -t1 -V3 test.dimacs
...
v 36971 -36972 -36973 -36974 -36975 -36976 -36977 -36978 -36979 -36980 36981
v 36982 -36983 -36984 -36985 -36986 -36987 -36988 -36989 -36990 -36991 -36992
v -36993 -36994 36995 36996 -36997 -36998 -36999 -37000 -37001 -37002 -37003
v -37004 -37005 -37006 -37007 37008 37009 -37010 -37011 -37012 -37013 -37014
v -37015 -37016 -37017 -37018 -37019 37020 37021 -37022 -37023 -37024 -37025
v -37026 -37027 -37028 -37029 -37030 37031 37032 -37033 -37034 -37035 -37036
v -37037 -37038 -37039 -37040 -37041 -37042 37043 37044 -37045 -37046 -37047
...

If I allow clasp to use more than one thread, the following thing happens:

clasp -t2 -V3 test.dimacs
...
v 36971 -36972 -36973 -36974 -36975 -36976 -36977 -36978 -36979 -36980 36981
v 36982 -36983 -36984 -36985 -36986 -36987 -36988 -36989 -36990 -36991 -36992
v -36993 -36994 36995 36996 -36997 -36998 -36999 -37000 -37001 -37002 -37003
v -37004 -37005 -37006c  0:R|  28497/26173  |  157182/1797    |      6797/0.198 |    1803/2000000   |
 -37007 37008 37009 -37010 -37011 -37012 -37013 -37014
v -37015 -37016 -37017 -37018 -37019 37020 37021 -37022 -37023 -37024 -37025
v -37026 -37027 -37028 -37029 -37030 37031 37032 -37033 -37034 -37035 -37036
v -37037 -37038 -37039 -37040 -37041 -37042 37043 37044 -37045 -37046 -37047

Note how a c-line starts right in the middle of a v-line.

It appears that both threads are writing without synchronization. I can reliably reproduce the problem on Win10, x64, both when writing to StdOut and when redirecting output to a file.

[BenKaufmann]

Hi,
this is actually by design. clasp currently only synchronizes output of models (i.e. v lines of different models won't intermingle) and ensures that individual comments lines are printed as a whole. However, during solving it does not do any synchronization between comment lines (which are mostly intended for debugging/informational purposes) and output of models and/or optimization values as this would complicate the design and slow down solving.

One compromise between full synchronization and intermingled output that I could easily implement would be to suppress (and discard) output of comments while a model is printed.

[sfiruch]

That would be great! Making sure that lines with optimization values are printed as a whole would be another way to make the output more useful.