What does it have to do with the compiler?

Posted Jun 21, 2019 9:46 UTC (Fri) by NAR (subscriber, #1313)
Parent article: C, Fortran, and single-character strings

I don't quite get what this bug has to do with the compiler. I mean it's like a function that has a "flags" arguments which previously only was using the lowest 2 bits, but now it starts to use the 3rd bit and suddenly all calls are broken that previously set the 3rd bit... Also it doesn't seem that hard to fix the callers with a sed command.

The other thing I don't get: it's C calling FORTRAN, so I'd guess there are C header files generated from the FORTRAN code that contain the prototype of the public functions and those should have the length parameter, shouldn't they?

What does it have to do with the compiler?

Posted Jun 21, 2019 10:32 UTC (Fri) by bjartur (guest, #67801) [Link]

There is no ABI for passing string arguments to Fortran procedures. Since the note broken code was written there was invented an API that works as you imagine. The code does not use it, because the code is older than the API, and it wouldn't fix the lack of an ABI. Speed can be gained by recompiling or swapping out BLAS/LAPACK, and hitherto this was not thought to require recompiling R or CBLAS/LAPACKE.

What does it have to do with the compiler?

Posted Jun 21, 2019 13:42 UTC (Fri) by mathstuf (subscriber, #69389) [Link]

> so I'd guess there are C header files generated from the FORTRAN code

Maybe for BLAS/LAPACK, but this isn't how it works in general since most Fortran code used directly from C happens within the same project. There's usually some configure-time logic to determine the mangling strategy of the in-use Fortran compiler. This then guides a macro selection around the core symbol names in a header which has the `extern` declarations of those functions with the right mangling. I suspect that the length arguments are visible in those C declarations, but not in the actual Fortran code.

What does it have to do with the compiler?

Posted Jun 21, 2019 19:10 UTC (Fri) by joib (subscriber, #8541) [Link]

> it's C calling FORTRAN, so I'd guess there are C header files generated from the FORTRAN code that contain the prototype of the public functions and those should have the length parameter, shouldn't they?

They have C prototypes for the Fortran functions, but they are presumably hand-written and they have omitted the length parameter. Which is the reason behind this entire mess.