Proposed fix for std::cout sync bug in output.cpp

[richmorrison]

When a non-MPI version of the code is compiled and run, print to std::cout is corrupted. Carriage returns appear in the wrong places, and data columns appear as a single line. This behaviour has not been seen when compiled for MPI.

One possible fix presented here replaces the offending stream directs of the form std::cout << std::endl with calls to fmt::print. The problem appears to be mixing the two output methods. Alternative fixes are probably possible.

I've not been able to explain his behaviour, or why compiling with MPI corrects the issue. Presumably fmtlib maintains its own print buffer for performance which is flushed at a different rate to std::cout. A possible complicating factor is that std::endl both introduces a carriage return and flushes the buffer. Perhaps compiling with MPI overrides the default std::cout behaviour.

I've left references to std::cout in other areas of the code. It may be prudent to remove them.

I've been unable to test this code since I can't get the Dockerfile to compile. However, since these are trivial changes I've decided to submit these changes for your review.

[paulromano]

@richmorrison Thanks for your contribution and congrats on your first pull request! We had deliberately changed fmt::print('\n') to std::cout << std::endl at some point to force output to be flushed, but perhaps that was not the right approach. I think what we should do is adopt your change here (using fmt::print) but in each place where we had a endl before, also add std::fflush(stdout). For good measure, we should #include <cstdio> for stdout.

[richmorrison]

@paulromano If fmtlib does maintain a separate buffer, and contains a flush function for the library user, you can possibly discard my proposed changes for a simpler solution.

Best,
Rich

[paulromano]

As far as I'm aware, fmtlib does not maintain a separate buffer -- it simply provides a nicer API on top of normal printf/etc calls. My suggestion to call fflush(stdout) was actually stolen from the fmtlib maintainer 😄

[richmorrison]

Cool. (One afternoon when I get more time I think I'd like to figure out why MPI changes things.)

[richmorrison]

@paulromano I've added your proposed additional changes.

[paulromano]

Thanks again @richmorrison!