[patch, Fortran] PR61933 Inquire on internal unit

[Janne Blomqvist]

On Wed, Jan 14, 2015 at 4:10 AM, Jerry DeLisle <jvdelisle@charter.net> wrote:
> Ping
>
>
> On 01/09/2015 07:10 PM, Jerry DeLisle wrote:
>>
>> Hello all,
>>
>> I had to do a bit of research on this.  I hope I have this correct.  For
>> background information, an "internal file" is a character variable that is
>> connected to an internal unit.  Likewise, an internal unit refers to an
>> internal
>> file which is a character variable.
>>
>> The subject refers to an "internal unit".  This is actually touching on
>> the
>> fringes of User Defined Derived Type I/O. In at least the draft version of
>> F2008
>> Standard I have at the moment, I find the following:
>>
>> 1) "The unit argument passed to a defined input/output procedure will be
>> negative when the parent input/output statement specified an internal
>> unit, or
>> specified an external unit that is a NEWUNIT value. When an internal unit
>> is
>> used with the INQUIRE statement, an error condition will occur, and any
>> variable
>> specified in an IOSTAT= specifier will be assigned the value
>> IOSTAT_INQUIRE_INTERNAL_UNIT from the intrinsic module ISO FORTRAN ENV"
>>
>> Also regarding IOSTAT= we have:
>>
>> 2) "Execution of an input/output statement containing the IOSTAT=
>> specifier
>> causes the scalar-int-variable in the IOSTAT= specifier to become defined
>> with:
>> the processor-dependent positive integer value of the constant
>> IOSTAT_INQUIRE_INTERNAL_UNIT from the intrinsic module
>> ISO_FORTRAN_ENV(13.8.2)
>> if a unit number in an INQUIRE statement identifies an internal file"
>>
>> 3) The standard lists as one of the processor dependencies - "the negative
>> value
>> of the unit argument to a defined input/output procedure if the parent
>> data
>> transfer statement accesses an internal file"
>>
>> To summarize all this, a negative unit number is used to communicate to
>> defined
>> type input/output procedures that the object of their I/O actions is an
>> internal
>> file, aka a character variable rather than an external file.  To avoid
>> ambiguity, the negative value should not be one created by NEWUNIT.
>> INQUIRE on
>> an internal file is not allowed and a specific error code is defined in
>> the
>> standard to be IOSTAT_INQUIRE_INTERNAL_UNIT.  I suspect the standards
>> committee
>> recognized the need to communicate to DTIO procedures and defined this
>> error
>> code to enable user code to check for this condition and pass it back up
>> to the
>> parent?
>>
>> When I first implemented NEWUNIT, I reserved the values -1 thru -9 for
>> future
>> use. Our earlier email discussions on this PR centered around the case of
>> UNIT=-1.  This patch generates an error for -1 only.  We can always change
>> this
>> later if the need to use -2 thru -9 comes about.
>>
>> Assuming we will implement the UDDTIO, we will then use -1 for the unit
>> number
>> to convey when an internal unit has been specified in the parent
>> input/output
>> procedure. As far as I can tell, the value used is processor defined and
>> it
>> appears that the cray COMPILER uses -1. Nothing special about that choice.
>>
>> I have eliminated the previous hack* I did in trans-io.c to handle the
>> -fdefault-integer-8 problem way back in PR33055, reject UNIT=-1 at compile
>> time
>> in io.c, and simplified the definition of unit existence to whether or not
>> the
>> unit is connected in inquire.c.
>>
>> Finally, IOSTAT_INQUIRE_INTERNAL_UNIT was already defined for us in the
>> intrinsic module ISO_FORTRAN_ENV but was never added into translate_error
>> in
>> runtime/error.c, so I have now done so and modified the test case to check
>> it.
>>
>> Regression tested on x86-64-linux. New test cases added.
>>
>> OK for trunk?
>>
>> Regards,
>>
>> Jerry
>>
>> * hey, what can I say, my work is cheap as in free beer.
>>
>> 2015-01-09  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
>>
>>      PR libgfortran/61933
>>      * io/inquire.c (inquire_via_unit): Set existing to true if a
>>      gfc_unit stucture was found for the given unit number.
>>      * runtime/error.c (translate_error): Add case for
>>      LIBERROR_INQUIRE_INTERNAL_UNIT.
>>
>> 2015-01-09  Jerry DeLisle  <jvdelisle@gcc.gnu.org>
>>
>>      PR fortran/61933
>>      * io.c (gfc_match_inquire): Generate error if unit number in
>>      inquire statement is a constant -1.  All other values allowed.
>>      * trans-io.c (gfc_trans_inquire): Delete dummy iostat variable.

I had a vague recollection that the F2008 standard said somewhere that
the unit -1 is reserved for some kind of error situations. Thus it
would be better to use, say, -2 for the unit reserved for internal
derived type I/O. But, for the life of me I cannot find this in the
standard now, so I guess I remembered incorrectly. So the patch is Ok
as is. Thanks.

-- 
Janne Blomqvist