Match error messages for TypeError in File specs

[herwinw]

There are two different code paths: true/false/nil uses the default inspect string in the error message, other objects use the class name.

The conversion into Integer is even more inconsistent, where it's sometimes "from X to integer" and sometimes "of X into Integer".

[herwinw]

Kind of a sequel to truffleruby/truffleruby#4140, it turned out the error handling in Natalie is all over the place too.
I picked File pretty much at random, probably because I used that one too at the TruffleRuby case.

[herwinw]

Okay, so it looks like File.truncate/File#truncate uses off_t, which is a different size in our CI on Linux than it is in MacOs and Windows. Based on that, the code uses either rb_num2long or rb_num2ll. And these functions have different error messages:

https://github.com/ruby/ruby/blob/86dba8cfaeabb3b86df921da24b3243b9ce4ab2a/numeric.c#L3100-L3105

https://github.com/ruby/ruby/blob/86dba8cfaeabb3b86df921da24b3243b9ce4ab2a/numeric.c#L3128-L3133

I would consider this a bug in MRI.

[eregon]

Yeah those inconsistencies are pretty annoying.
I'd suggest to make a PR to fix them in CRuby to just no implicit conversion from nil to integer (into is weird and less used AFAIK).
In the meanwhile specs could accept either and then we can just be consistent in TruffleRuby/Natalie/etc.

[eregon]

I'd suggest to make a PR to fix them in CRuby to just no implicit conversion from nil to integer (into is weird and less used AFAIK).

Though from the diff into seems more common.
I don't really care but I agree it should be consistent.

A quick search gives https://github.com/search?q=repo%3Aruby%2Fruby+%2Fno+implicit+conversion+%2F+path%3A*.c&type=code
convert_type is the generic one so I think we should be consistent with that one, and it uses into:
https://github.com/ruby/ruby/blob/86dba8cfaeabb3b86df921da24b3243b9ce4ab2a/object.c#L3338-L3345

[trinistr]

I'd suggest to make a PR to fix them in CRuby to just no implicit conversion from nil to integer (into is weird and less used AFAIK).

There is also the annoying capitalization issue of "integer" vs "Integer", etc. Considering that generic message uses class's name, it would be nice if the custom messages were also consistent in that regard.

[herwinw]

Upstream case: https://bugs.ruby-lang.org/issues/21864

[eregon]

Could you allow no implicit conversion of X into Integer here?
I think other Rubies should use that for consistency and should be able to pass this spec with that.
Also will make it easier if https://bugs.ruby-lang.org/issues/21864 gets fixed.

[eregon]

Same, could you allow no implicit conversion of nil into Integer here?

[eregon]

Given ruby/ruby#16078 (comment), let's use raise_consistent_error here too, that will result in better and simpler expectations.
I can add it to mspec in parallel so there is no need to wait for that PR + sync.

[eregon]

@herwinw Pushed to ruby/mspec@4ea4bf7, could you use it here?