Remove CFG_USERINPUTFD and use stdin replacement in affected tests.

[antonsviridenko]

Closes #1609 and closes #1857

Affected tests now use stdin temporarily replaced with a pipe read end.

[codecov]

Codecov Report

Patch coverage: 86.53% and project coverage change: +0.64 🎉

Comparison is base (32149be) 83.81% compared to head (56c28ad) 84.45%.

❗ Current head 56c28ad differs from pull request most recent head d997e6c. Consider uploading reports for the commit d997e6c to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2091      +/-   ##
==========================================
+ Coverage   83.81%   84.45%   +0.64%     
==========================================
  Files         161      143      -18     
  Lines       32395    29147    -3248     
==========================================
- Hits        27151    24617    -2534     
+ Misses       5244     4530     -714     

Impacted Files	Coverage Δ
src/lib/crypto/backend_version.cpp	100.00% <ø> (+26.08%)	⬆️
src/librekey/g23_sexp.hpp	100.00% <ø> (ø)
src/librepgp/stream-packet.h	100.00% <ø> (ø)
src/tests/support.h	100.00% <ø> (ø)
src/tests/utils-rnpcfg.cpp	100.00% <ø> (ø)
src/librepgp/stream-packet.cpp	80.93% <50.00%> (-0.53%)	⬇️
src/tests/generatekey.cpp	89.73% <53.84%> (+1.05%)	⬆️
src/lib/rnp.cpp	79.88% <60.00%> (-0.18%)	⬇️
src/rnpkeys/tui.cpp	88.51% <85.71%> (+1.15%)	⬆️
src/lib/crypto/hash_common.cpp	98.11% <100.00%> (-0.14%)	⬇️
... and 8 more

... and 67 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[antonsviridenko]

 258/259 Test #248: rnp_tests.test_fuzz_verify ....................................................   Passed    5.99 sec
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
ci/lib/install_functions.inc.sh: line 676:  2438 Aborted                 (core dumped) "$@"
Error: Process completed with exit code 134.

[antonsviridenko]

strange, all of the failing centos-and-fedora instances fail with the same std::bad_alloc error in a place not possibly related to the introduced changes

[ni4]

@antonsviridenko It could be somehow related to the redirected stdin (and be thrown from GTest side), as other PRs work well.

[antonsviridenko]

All failed instances use OpenSSL as a backend...

[antonsviridenko]

Any ideas how to debug this?

ci/lib/install_functions.inc.sh: line 676:  2438 Aborted                 (core dumped) "$@"

Looks like it's something related to Python environment in Centos/Fedora.

[ni4]

@antonsviridenko Did you try to revert you commits, related to stdio redirection, and see whether that helps?

[antonsviridenko]

Did you try to revert you commits, related to stdio redirection, and see whether that helps?

I assume that yes, reverting commits will remove CI failures.

I think it could be related to duplicated STDIN not having CLOEXEC flags... And affected platforms run cli_tests through some python wrapper.

[antonsviridenko]

Is there any possibility to recreate the same local environment as in failing centos-and-fedora machines?

[ni4]

@antonsviridenko Wow, this one get lost somehow. Locally you may just use corresponding Docker container, install prerequisites, copy rnp sources there, build and run tests. Another option is https://github.com/nektos/act but I didn't try it yet and don't know in which way it works and is it possible to properly debug things.