[PATCH, libgfortran] Fix PR21875 (FM111.f) BN and BZ format specifiers

[Jerry DeLisle]

Jerry DeLisle wrote:
>
> I have isolated two separate problems.  The first is discovered 
> coincidentally bacause of the manual entry of numbers from the console 
> in the reduced case given in bugzilla that you referred to.
> 
> 1) When reading from stdin, the flags.blank was set to BLANK_ZERO.  This 
> causes trailing blanks on input fields to be converted to zero's 
> resulting in exponents out of range.  This is in io/unit.c.  The default 
> should be set to BLANK_UNSPECIFIED.
> 
> Once that is fixed, then 2 happens
> 
> 2) Exponents with trailing blanks are getting garbaged up for some 
> reason.  I am in the middle of fixing that.
> 
The attached patches fix the problems in libgfortran and revise the test 
case to include a test of the snippet similar to the SPECfp problem with 
trailing blanks in exponents.

Regression tested (excluding real_const_1.f90, real_const_2.f90, and 
elemental.f90, which are failing from something else).

2005-07-19  Jerry DeLisle  <jvdelisle@verizon.net>

* io/read.c (read_f): Fix reading of trailing blanks in   	exponents 
related to BZ and BN format specifier logic.
* io/unit.c (init_units): Change default setting of u->flags.blank
to BLANK_UNSPECIFIED for stdin, stdout, and stderr.

OK to commit to mainline and 4.0?

Jerry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bz-bn.diff
Type: text/x-patch
Size: 3843 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20050720/5045e159/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bz_bn_test.diff
Type: text/x-patch
Size: 2005 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20050720/5045e159/attachment-0001.bin>