Fix gconstpointer fixup

[fkrull]

Ran into this: I have an _equals and _hash that take gconstpointer and I had to change the *mut to *const in the generated code to make it compile.

This change feels like it should have unintended side effects, but also if you're dealing with gconstpointer, *const seems always correct.

[EPashkin]

@fkrull This change broke glib::DateTime because Shared types implement only ToGlibPtr<*mut _>,
can you show bad object definition's "glib_wrapper!" and generated hash ?

[fkrull]

@EPashkin Here's the entire offending file. This is generated by this modified branch, but the only difference is the *const instead of *mut. I can dig out the error message, but I think it was just the opposite of what you described: that the type doesn't implement ToGlibPtr<*mut _>.

(The compiler helpfully suggested the *const implementation to me so I may have jumped to conclusions there.)

[EPashkin]

Yes, Boxed types don't implement ToGlibPtr<*mut _>, so only boxed requires "*const" in fixup_special_functions.
Bad that detection logic for it is not compact.
You can set flag like is_boxed in https://github.com/gtk-rs/gir/blob/master/src/analysis/record.rs#L89-L91 and pass it all way down.

[fkrull]

@EPashkin Ok, I've added a check on that flag which should fix glib.

[EPashkin]

Can you fix pointer for RecordType::AutoBoxed too? This is second branch in https://github.com/gtk-rs/gir/blob/master/src/analysis/record.rs#L89-L91

[fkrull]

@EPashkin Thanks for the hints. I hope that should be it now? glib generates as before and my tricky type got fixed up even without having to specify use_boxed_functions explicitly.

[EPashkin]

Please, rename variable to "is_boxed".

[EPashkin]

@fkrull Good to hear that config not needed in your case.

[EPashkin]

Thanks, I waiting CI just in case, then will do squash+merge