Enable nullable on some modules

[slozier]

Add nullable annotations on some modules.

Other notable changes:

• Excludes PerformModuleReload from the analyzer so we don't need to pragma disable it.
• Gets rid of ø from params and ParamDictionary arguments. There are still a couple \u00F8 left over in other arguments. Not sure if they're necessary or not...
• Remove NotNone from ParamDictionary arguments since it's never null.
• Add NotNone to all params arguments. Not having this can result in a null instead of the empty array. There seem to be some cases where this isn't necessary but I haven't quite figure out the logic so I'm just putting it everywhere (with the help of an analyzer).
• Other small bug fixes.

[BCSharp]

Also excludes PerformModuleReload from the analyzer so we don't need to pragma disable it.

Great! I was just about starting to do the same…

[BCSharp]

• Gets rid of ø from params and ParamDictionary arguments. There are still a couple \u00F8 left over in other arguments. Not sure if they're necessary or not...

Strange that there are some \u00F8 in some other arguments. We would probably have to look at those instances case by case (if somebody has the patience).

I would suggest to further standardize the parameter names to args and kwargs, since these are Python names by convention and they do show up in generated signatures by help(...). I see in some places kwArgs (.NET camelCase) or otherArgs (???)

• Remove NotNone from ParamDictionary arguments since it's never null.

• Add NotNone to all params arguments. Not having this can result in a null instead of the empty array. There seem to be some cases where this isn't necessary but I haven't quite figure out the logic so I'm just putting it everywhere (with the help of an analyzer).

The odd behaviour is most likely because IronLanguages/dlr#249 is not finished yet — only the prohibition by keyword part is done. The prohibition by position is not, but the goal is that ParamDictionary is prohibited by position as well, while params array is allowed (this is only partly implemented). So the rules you put in the Analyzer are ultimately correct (for IronPython, at least).

The logic for params array is (or is supposed to be) that without [NotNone], when None is passed, then args is null (classic binding by position). With [NotNone], args becomes [null] (an array with 1 element of value null). The reason for supporting binding by position is that it matches C#, and IPY needs to be able to call .NET API like C# can.

I am a bit surprised that ParamDictionary never binds by position (hence is never null), perhaps I implemented more than I remember now. Or perhaps there are still some corner cases when it is possible (like multiple overloads with various number of parameters combined with optionals).

[BCSharp]

I suppose PerformModuleReload is special only if it has SpecialMethodAttribute? Although I haven't seen any non-special ones around.

[slozier]

From what I can tell SpecialMethodAttribute doesn't affect anything. 🤷

[slozier]

Strange that there are some \u00F8 in some other arguments. We would probably have to look at those instances case by case (if somebody has the patience).

It's not too bad, count is down to 35 lines after this PR. Some of them seem "legit" like dict.update where I think we'd need something like PositionalOnly to express things properly. Others are more of a head scratcher, but hopefully they're there for a reason...

I would suggest to further standardize the parameter names to args and kwargs, since these are Python names by convention and they do show up in generated signatures by help(...). I see in some places kwArgs (.NET camelCase) or otherArgs (???)

Sure, I can look at that.

I am a bit surprised that ParamDictionary never binds by position (hence is never null), perhaps I implemented more than I remember now. Or perhaps there are still some corner cases when it is possible (like multiple overloads with various number of parameters combined with optionals).

I did double check this with a simple void delete_me([ParamDictionary] IDictionary<string, object> x) which failed with a TypeError: delete_me() takes no arguments (1 given) if called with delete_me(None). Error message is not great, but at least it's throwing as expected...

I think it's when we have both a ParamDictionary and a params array that the NotNone would be unnecessary. The dictionary prevents binding by position (because it's always before params which has to be last) so when calling with None we end up with the [null] array. Anyway, I'm going to leave the analyzer as-is for now since I think it makes the intent clearer.

[slozier]

I would suggest to further standardize the parameter names to args and kwargs, since these are Python names by convention and they do show up in generated signatures by help(...). I see in some places kwArgs (.NET camelCase) or otherArgs (???)

Sure, I can look at that.

Haha, I just wrote that I would look at it then clicked on merge. 🤦 I've got more analyzer tweaks coming up so will include it in that...