cpp/python: export isl_id

[ftynse]

Export isl_id allocation and name management. Disallow unnamed identifiers by
construction and hide user pointers.

Identifiers can have an optional name and an optional user pointer. However,
identifiers without a name and a user pointer are forbidden. The user pointer
is a void * pointer to a C data structure, managed and potentially deallocated
within C code. It is not fully compatible with C++ objects featuring a virtual
destructor and not sufficiently type-safe. The notion of pointer does not exist
in Python. Therefore, it makes sense to hide the presence of a user pointer
from the bindings. This is achieved by providing the "isl_id_from_name"
(constructor) function that only takes a name of the identifier. If necessary,
user pointers may be accessed through C API, but doing so should be
discouraged. As an alternative, a user of C++ or Python bindings may maintain a
mapping between identifier objects and arbitrary data.

The equality of pointer objects is preserved. However, hiding user pointers may
lead to a situation where two identifier objects in C++ or Python are not equal
although their visible attributes (names) are equal. This can only happen if at
least one of the pointers was obtained from C code.

Since identifiers constructed in C are allowed to not have a name, i.e. have a
NULL name, which cannot be expressed as a C++ std::string, provide an
additional function "isl_id_has_name" to check whether the name is not NULL.

Signed-off-by: Oleksandr Zinenko git@ozinenko.com

[ftynse]

Here is a minimal Python-friendly version of isl_id. This does not require modifying the generator, but does require modifying the public C API.

I remember @nicolasvasilache wanting to avoid exposing user pointers, you are served :)

[ftynse]

We could use operator support from #10 and introduce isl_id_eq function to generate a comparison operator and make isl::id objects directly comparable without modifying the generator.

[tobiasgrosser]

Very cool! This patch implements the essentials of isl::id. It is certainly sufficient for the python bindings, but should also cover the common C++ use cases.

I would pull the extensions of the C interface into a separate commit and motivate there why they are useful.

[ftynse]

But they don't seem to be useful on their own, their only motivation is to have a consistent bindings interface...

…

On 05/09/17 14:26, Tobias Grosser wrote: Very cool! This patch implements the essentials of isl::id. It is certainly sufficient for the python bindings, but should also cover the common C++ use cases. I would pull the extensions of the C interface into a separate commit and motivate there why they are useful. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#21 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABcTa6vrsjlfzfkyDCx-SnbNOmOTEWKIks5sfT3igaJpZM4PM1Z1>.

[tobiasgrosser]

Sure. Still, we commonly commit extensions to the C interface and changes to the exported functions separately. You can just mention that they will be used in the subsequent commit.

Anyway, very cool stuff. Super simple, but as a result also easy to add, I assume.

[tobiasgrosser]

Looks great to me!