Skip to content

Conversation

@jeffhammond
Copy link
Member

@jeffhammond jeffhammond commented Mar 6, 2025

MPI 5.0 says:

As a result, the functions defined in Section 19.3.4 and Section 19.3.5 are not part of this ABI.

I need to do a short reading to add MPI_F08_Status to the list.

Users no longer need an MPI_F08_Status typedef because the status ABI is no longer opaque and users can do the status conversion manually once they know the Fortran integer size, which they can determine already. I am going to read this change in the next few minutes.

Resolves #46

@jeffhammond jeffhammond self-assigned this Mar 6, 2025
@jeffhammond
Copy link
Member Author

@dalcinl
Copy link
Contributor

dalcinl commented Mar 6, 2025

Jeff, getting rid of old cruft is great, but moves like this one will break API, and thus source compatibility with existing code.

@dalcinl
Copy link
Contributor

dalcinl commented Mar 6, 2025

I'm pretty sure MPI will not be the only third-party code that will be affected badly from the removal of f2c/c2f APIs.

@dalcinl
Copy link
Contributor

dalcinl commented Mar 6, 2025

As MPI did not have an ABI, folks have abused the c2f/f2c machinery to have some sort of lame ABI to pass handles around. I already commented elsewhere the case of Paraview Catalyst.

@jeffhammond
Copy link
Member Author

These are not part of the MPI ABI. MPI_Fint is fundamentally impossible to support in an ABI.

Implementations can add them downstream.

@dalcinl dalcinl marked this pull request as draft March 6, 2025 13:02
@dalcinl
Copy link
Contributor

dalcinl commented Mar 6, 2025

Superseded by #39.

@dalcinl dalcinl closed this Mar 6, 2025
@dalcinl dalcinl deleted the remove-all-MPI_Fint branch March 6, 2025 17:55
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

MPI_T_CB_REQUIRE_* changes

3 participants