Revamp constructors - Githubissues

mulimoen commented 3 years ago

This moves and renames the constructors of CsMatBase. This makes construction of matrices for non-Copy types easier, and unifies constructors.

Modified:

new: Now implemented on CsMatBase instead of CsMatI. Does not sort indices. Panics on failure
new_csc: Counterpart to above
try_new: Returns a tuple with all owned data + error on failure
try_new_csc: Counterpart to above

New:

new_unchecked: Unsafe constructor for CsMatBase, does not do any data checks.

Removed:

new_view: new can now be used instead
new_view_raw: This function is really unsafe, we are opening ourselves to a use-after-free by making a lifetime from nothing. I'd rather push this responsibility away from this crate.

Maybe we could remove new_vecview_raw? This is not in use by any of our crates, and suffers from the same problems as new_view_raw.

Fixes #273

mulimoen commented 3 years ago

On second thought I think it would be ok to have both new and try_new. I will however change try_new so it returns the owned data.

mulimoen commented 3 years ago

@vbarrielle What is your though on new_vecview_raw? This can easily be replaced by new_unchecked as was done with sprs-ldl

vbarrielle commented 3 years ago

@vbarrielle What is your though on new_vecview_raw? This can easily be replaced by new_unchecked as was done with sprs-ldl

I think the more we can use new_unchecked or new_trusted to replace these calls, the better. If we can have a very coherent API it'll be easier to code against sprs.

sparsemat / sprs

Revamp constructors #274