Attempt to solve fread stack imbalance on Windows when rereading.

[mattdowle]

Closes #2481

[codecov-io]

Codecov Report

Merging #2488 into master will decrease coverage by 0.12%.
The diff coverage is 78.87%.

@@            Coverage Diff             @@
##           master    #2488      +/-   ##
==========================================
- Coverage   91.55%   91.43%   -0.13%     
==========================================
  Files          63       63              
  Lines       12046    12059      +13     
==========================================
- Hits        11029    11026       -3     
- Misses       1017     1033      +16

Impacted Files	Coverage Δ
src/freadR.c	89.38% <43.33%> (-3.57%)	⬇️
src/fread.c	95.96% <88.39%> (-0.35%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8bf7334...907a031. Read the comment docs.

[st-pasha]

I believe this continue should remain

[mattdowle]

Reason for removal was :
error: invalid branch to/from OpenMP structured block
but maybe the continue could be just afterwards. Will try ...

[st-pasha]

Not sure if this would help or not, but I found the following explanation:

Within a given development domain, [calling conventions] tend not to be a problem. Each language generally has a convention that is implicit for all method calls. C/C++ uses the same convention for invocation of C/C++ calls, Python for other Python calls, etc. When crossing domains, it can become a problem if one domain doesn't use the same as another. Perhaps most common in windows, a function exported with "C" style declarations (cdecl) may cause an unbalanced stack (or worse) when called as though it had a stdcall convention, which is the method recognized by WINAPI (windows system) calls.

So the root of the problem could be that some Windows system call somehow gets invoked with wrong calling convention, which after many calls may lead to stack corruption. Which call it is I'm not sure, but could be snprintf or clock_gettime or something similar.

[st-pasha]

you can add here

bool reportProgress = false;
#pragma omp master
reportProgress = args.showProgress;

then everywhere where you check (me==0 && args.showProgress) you can just replace with reportProgress