Using strong-named assemblies can cause difficulties with the GNU LGPL
license, which allows for one to recombine or relink their application
with modified versions of the code. While one solution is to share the
private key so that people can sign the assemblies themselves, this
would break the trust that is expected from signed assemblies. For now,
the easiest fix is to simply not sign the assemblies. Note that by
doing so, we prevent the code from being referenced from other signed
assemblies.
This also fixes an issue introduced in revision ba94d36a80 ("Added
auto-update for WindowsService and Service."), where the WindowsService
project (signed) referenced the AutoUpdater project (not signed).
We also removed instances of <SignAssembly>false</SignAssembly> to be
consistent with newly created .csproj files that do not contain the
SignAssembly element.
This was motivated by the discussion in issue #2814.
By changing to IEnumerable, it is possible to iterate only a portion of the list, which is useful when not all entries are needed (e.g., when testing a connection).
All existing backends have been updated, and any which were able to be changed to yield return results in a straightforward way now do.
Many backends had a try/catch in the List() method. Due to the fact that yield returns can't be placed within a try/catch block, these have been refactored to either scope the try/catch to the parts that (should) be the only places throwing exceptions, so that exceptions are still caught and handled.
Note that lazy evaluation may cause some changes in behavior - exceptions that were previously thrown at the point of invokation of List() may now be thrown while it is being enumerated.
I believe this will not be problematic though, as the only well-known exception seems to be FolderMissingException, which should be thrown by Test(), but TestList() attempts to enumerate the list to force this exception.
Any places that require the legacy behavior can get it by simply converting the lazy enumerable to a List()
InvariantCulture is useful when comparing / sorting human language strings in a culturely correct way. It handles things like accented letters in a way that makes sense to humans (e.g., 'a' should be sorted next to 'á', rather than after 'z').
Ordinal looks just at the raw code points of the characters. As such, it is recommended for use in cases when comparing system strings (file paths, command line parameters, config settings, etc.). Since it doesn't need to use the culture specific sorting rules, this method can often be faster.
For more information, see https://stackoverflow.com/questions/492799/difference-between-invariantculture-and-ordinal-string-comparison (and other related questions)
Kenneth Skovhede