SetEnvironmentVariable does not work properly on Unix

[wfurt]

This is related to #4103 and dotnet/coreclr#3140.
In my case I use

Environment.SetEnvironmentVariable("all_proxy",` "http://10.121.122.110:3128");

and I expect HTTP Handler to honor it same way is if set before process start.
In this case, we do no fork child process so fix-up for process start is not applicable.

I understand that setenv() is not thread safe and that we choose not to actually set environment variables but to work on our copy instead.
We may perhaps re-consider and fix it.
There is similar pattern calling openssl where init function is not thread safe. To address that we simply added lock around the call to make it thread safe.
This would be needed only for Linux (PAL?) and I would think the setenv() call is not likely on performance critical path.

I also verified that when i call setnev() via [DllImport] works as expected.
It is workaround but it would be nice if the SetEnvironmentVariable just work.

[jkotas]

If we "fix this", we would introduce a different set of problems: memory leaks and/or potential security issues on systems where setenv is not thread safe. https://github.com/dotnet/coreclr/issues/635#issuecomment-91495713 has my take on this.

I think it is fine to recommend to call setenv() via [DllImport] as workaround if you really need the libc environment changed, and you know what you are doing and understand the consequences.

[jkotas]

http://www.club.cc.cmu.edu/~cmccabe/blog_the_setenv_fiasco.html

[stephentoub]

Also, even if we serialize our own calls to setenv, that could still result in problems with libcurl concurrently calling getenv to read them.

And for ManagedHandler, it'll be using Environment.GetEnvironmentVariable rather than getenv and will "just work".

I don't think there's anything to be done here.

[parched]

Can this note be added to these docs? It would've saved me a fair bit of debugging time.

[danmoseley]

@parched feel free to offer a PR to improve the doc (click edit button on doc page)

I'll close this as apparently that was originally intended