[lmi] 'authenticity_test' broken years ago?

[Greg Chicares]

Vadim--Can you explain this puzzle? With or without
  https://github.com/vadz/lmi/pull/117
one particular test in 'authenticity_test' misbehaves,
though in a way that (defectively, I suppose) isn't
counted as an error. It occurs here:

  /// Postcondition: the file passes a test with the 'lmi_md5sum' program.

  void PasskeyTest::InitializeMd5sumFile() const
  {
...
      std::string s = "lmi_md5sum --check --status " + 
std::string(md5sum_file());
      system_command(s);

and I would expect that command to succeed, printing something like
  coleridge: OK
but here's what I observe, in all cases running wine+debian-testing;
first, without PR117:

---->--BEGIN-->----
/opt/lmi/src/lmi[0]$make unit_tests unit_test_targets=authenticity_test.exe
make[2]: Nothing to be done for 'build_unit_tests'.
List of unit-test targets that did not build successfully:

List ends.

Running authenticity_test:
  Result of 'md5sum --version':
md5sum (PCP patchlevel 2) (GNU textutils) 1.22
Exit code 1 from command 'md5sum --check --status validated.md5'.

... 76 tests succeeded
----<--END--<----

Similarly, with PR117, the line corresponding to "Exit code 1" above is:

---->--BEGIN-->----
Integrity check failed for 'coleridge'
----<--END--<----

However, if I run the same command manually, after preserving the files
that it uses thus:

 void PasskeyTest::RemoveTestFiles(char const* file, int line) const
 {
     std::vector<std::string> filenames;
     filenames.push_back("expiry");
     filenames.push_back("passkey");
-      filenames.push_back("coleridge");
-      filenames.push_back(md5sum_file());
+//    filenames.push_back("coleridge");
+//    filenames.push_back(md5sum_file());

then MD5 validation succeeds, both for debian's native 'md5sum',
and also for the new custom-built one run under wine:

/opt/lmi/gcc_i686-w64-mingw32/build/ship[0]$md5sum --check validated.md5    
coleridge: OK
/opt/lmi/gcc_i686-w64-mingw32/build/ship[0]$md5sum --check --status 
validated.md5    
/opt/lmi/gcc_i686-w64-mingw32/build/ship[0]$wine lmi_md5sum.exe --check 
validated.md5
coleridge: OK
/opt/lmi/gcc_i686-w64-mingw32/build/ship[0]$wine lmi_md5sum.exe --check 
--status validated.md5

At this point I thought that the problem must lie in
'system_command_non_wx.cpp', which uses ::CreateProcessA(),
perhaps in some subtly incorrect way (though it seems to have
worked well for many years). But then I changed the unit test
to use '--version' without '--check', and...it says the file
is "OK", but reports that testing its integrity fails:

---->--BEGIN-->----
/opt/lmi/src/lmi[0]$make $coefficiency unit_tests 
unit_test_targets=authenticity_test.exe         
make[2]: Nothing to be done for 'build_unit_tests'.
List of unit-test targets that did not build successfully:

Running authenticity_test:

List ends.
  Result of 'lmi_md5sum --version':
lmi_md5sum 0.9

This program is a limited functionality variant of GNU 'md5sum' utility
and is part of the 'Let Me Illustrate' project.
Command is: 'lmi_md5sum --check validated.md5'. [I added that trace statement.]
coleridge: OK
Integrity check failed for 'coleridge'

.... 72 tests succeeded

!!!! no errors detected
/opt/lmi/src/lmi[0]$
----<--END--<----

How can this be? I'm sure I'm missing something obvious.