A list of gfortran junior jobs

[FX Coudert <fxcoudert at gmail dot com>]

Hi all,

As was asked some time ago on the list, we currently don't have a  
list of "junior jobs", i.e. tasks that require only limited  
understanding of the guts of the compiler. Since GCC development is  
now back in stage 1, which is the right time window for adding new  
features, I'd like to list here a few features that are probably easy- 
to-medium difficulty, and that I personnaly would be glad to have  
into the next gfortran release series.

The criterion of choice is "the kind of things that would have seemed  
doable when I submitted my first few patches". Feel free to assign  
them to yourself in bugzilla and tell the list which one you'll be  
trying: that's no commitment, you can always unassign yourself if you  
can't make it work. Ask the list for things you don't understand,  
come and see us on the GCC IRC channel, in short: join the club!

Now, the list itself.

  -- PR28484, "SYSTEM_CLOCK with real-type COUNT_RATE": intrinsics  
are probably the easier part of the Fortran front-end; what is needed  
here is to relax the checking (in check.c, function  
gfc_check_system_clock) and change the resolution to call two  
different library functions depending on the type of the COUNT_RATE  
argument (in iresolve.c, function gfc_resolve_system_clock). Copy the  
existing implementation of SYSTEM_CLOCK in libgfortran to make a  
version for real COUNT_RATE argument.

  -- PR29240, "allow an additional arg to SIGNAL": once again,  
intrinsics are usually easy to deal with. There are already numerous  
intrinsic with optional args that you can peek at to see how it's  
handled, both on front-end and library side.

  -- PR29452, "compile-time and run-time checks for I/O specifiers":  
the problem and solutions are very well described by Tobias Burnus in  
comment #2.

  -- PR28378, "Intrinsic extensions should be deselectable via  
command line": adding a new option is fairly easy; this one needs a  
bit of C programming, probably creating a linked-list of names  
specified by the user to check them later on.

  -- PR24285, "format(1000(a,$)) is not accepted": probably not too  
difficult, you simply have to change an error into a warning; but it  
first requires checking what other compilers accept and do.

Happy hacking!

FX