Re: problem with boost test matrix for serialization related to changes in file system?

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2020-06-07 21:16:54

On 2020-06-08 00:00, Robert Ramey via Boost wrote:
> I made a few checkins to the serialization library which I had 
> previously tested on my working system.  But Now I'm getting some very 
> weird errors in the Boost Test matrix:
> 
> https://www.boost.org/development/tests/develop/developer/serialization.html 
> 
> 
> Anyone who has some useful insight would be free to chime in.
> 
> According to P Dimov, Filesystem is guilty and i should request help 
> from andrey semashev - so here it is.

Yes, it looks like on that system (which is Linux) Windows-only 
libraries bcrypt and advapi32 are being linked with Boost.Filesystem. 
I'm not really sure why that happens, as I'm checking that the target 
system is Windows when I'm adding those libraries:

https://github.com/boostorg/filesystem/blob/a3e517365a41f616b895f410c6ae430fcd7cf89a/build/Jamfile.v2#L21

I would appreciate if someone knowledgeable about Boost.Build could take 
a look to tell what I'm doing wrong.

The two weird things about this problem are:

1. Travis CI on Linux passed.
2. The error log says that both bcrypt and advapi32 are being linked, 
and the code in the Jamfile is selecting only one of those.

I'm going to try marking bcrypt and advapi32 libs explicit, just in case 
it helps.