using GNU Fortran 95 (GCC) 4.1.0 20051126 (prerelease) with '-g -pedantic
-std=f95', I get a bad / no diagnostic for the following invalid code:
INTEGER, PARAMETER :: K(2)=1
SELECT CASE(I)
CASE(MAXLC(K,1))
END SELECT
END

Comment #1 from kargl at gcc dot gnu dot org 2005-11-27 03:52
What broken here? Where are the details?

Comment #2 from eedelman at gcc dot gnu dot org 2005-11-30 14:18
(In reply to comment #1)
What broken here? Where are the details?

I wondered that as well for a while. The problem, IIUC is that the
case-selector
must be an initialization expression. I'm no language lawyer, but if I
understand the F2003 (draft) standard, MAXLC(k,1), where 'k' is a PARAMETER,
ought to be a perfectly valid initialization expression. The standard, section
7.1.7, says on initialization expressions:

"It is an expression in which each operation is intrinsic, and each primary is

[]

(5) A reference to a transformational standard intrinsic function other than
NULL, where each argument is an initialization expression,"

MAXLC, is, AFAIK, a transformational standard intrinsic, and 'k' and 1 are,
AFAIK, initialization expressions.

It turns out, however, that the F95 standard has a slightly different
definition
of initialization expressions (in section 7.1.6.1); it seems that MAXLC is not
allowed here. So it seems the code is still invalid F95, and we should give a
warning/error message if strict F95 standard checking is requested. I must,
however, say, that I find it a bit wierd that MAXLC isn't allowed in this
context, so I'm not sure what to think here I leave this as unconfirmed
for now.

Comment #3 from eedelman at gcc dot gnu dot org 2005-11-30 18:29
(In reply to comment #2)
context, so I'm not sure what to think here I leave this as unconfirmed
for now.

Ifort 8.1 reports the following error:

In a CASE statement, the case-value must be a constant expression. [MAXLC]
CASE(MAXLC(K,1))
^

Which agrees with my interpretation of the F95 standard. And now when I
actually try it in gfortran (:-)), I get an ICE:

erik:~$ gfortran huj.f90
huj.f90: In function 'MAIN__':
huj.f90:1: internal compiler error: in gfc_conv_constant_to_tree, at
fortran/trans-const.c:276

Confirmed as 'ice-on-invalid'.

Comment #5 from fxcoudert at gcc dot gnu dot org 2006-05-12 21:18
Well, the testcase is valid F2003 but not valid F95. We have to get it working
(for F2003 mode), which probably means adding a simplification function for
MAXLC. And the same is true for all the intrinsics allowed by F2003 in an
initialization expression.

But we also have to get it rejected in F95 mode, which means keeping an
explicit list of the intrinsics that are allowed in initialization expressions
in F95, under the rules of 7.1.6.1 (all subcases a to e of case #7). I see two
options here: either we add a property for all intrinsics, define it in the
list and use that (lots of changes and thinking) or we add a new function
valid_intrinsic_in_init_expr(). I'd prefer the second option, and will be
thinking of something along these lines.

Comment #6 from fxcoudert at gcc dot gnu dot org 2006-06-19 18:21
Paul Thomas proposed a patch that fixes the F95 problem. We still need to write
simplification routines to enable such code (which is valid F2003) to compile
with gfortran. I don't have time for that right now.