Re: [RFC] fseek, an alternative implementation (PR22359)

[Jerry DeLisle <jvdelisle at verizon dot net>]

Daniel Franke wrote:
> While working on the docs last week, I learned, that the FSEEK intrinsic is 
> not implemented yet. 
>
> The FSEEK implementation of g77 takes a statement label to jump to, if an 
> error occurs. Since there is no other intrinsic that takes a statement label, 
> considerable effort would be necessary to implement the framework needed. As 
> demand is low (one PR, no dupes, no usage of FSEEK shown by google-code 
> search), it might be worthwile to implement the FSEEK intrinsic for 
> completeness and replace the statement label by a status flag:
>
>   g77:        CALL FSEEK(unit, offset, whence, *100)
>   gfortran:   CALL FSEEK(unit, offset, whence, status)
>               IF (status /= 0) goto 100
>
> This is not 100% compability, but reasonable close.
>
> Please find a preliminary patch that implements FSEEK using a status flag 
> attached. Of course, this implementation (still) has the same problems with 
> casting/assigning of the status flag as the other intrinsics. 

I like the idea of the status flag.  I will repeat here Janne's comment on IRC 
to consider casting the integer offset to GFC_OFFSET and having only one type 
function call in the library.  Otherwise the patch looks reasonable.

Consider for the testsuite:  Some large seeks mingled with some other IO 
statements to make sure we have no odd side effects.  I have seen enough bugs do 
to the library getting lost. You may want to keep extremes out of the test suite 
so we don't hang systems that do not support LARGE_FILE I/O, but do some on your 
local machine. :)

Jerry