Consider the following testcase:
! { dg-do run }
! Adapted from fmt_t_6.f testcase for PR 34782
character a(6)
data a / 'a', 'b', 'c', 'd', 'e', 'f' /
write(*,'(T20,A3, T1,A4, T5,A2, T7,A2, T9,A4, T17,A2)')
1 'a', 'b', 'c', 'd', 'e', 'f'
print *, 'should be'
write (*, '(A)') " b c d e f a"
end
gcc version 4.4.0 20080427
% ./fmtstdout.gf44
a b c d e f
should be
b c d e f a
% ./fmtstdout.gf44 > foo
% cat foo
b c d e f a
should be
b c d e f a
I stumbled upon this when working on part I of pr 25561. My patch regtests
cleanly except for fmt_t_6.f. However with the more strict read/write/seek
interface the kind of dirty trick that allowed T edit descriptor to work for
seekable streams won't work anymore as we don't have direct access to the
buffer. T and TL descriptors are special in that they're the only one that can
force us to go backwards in the formatted record. Or are there others?
The only way I can think of to fix this would be something like
On output, if we encounter a T or TL edit descriptor
allocate a temporary buffer, fill it with blanks (' ')
create a temporary internal unit, using the above temp buffer as the 'file'
do all the formatting to this internal unit
when the record is completed
copy the contents from the temp internal unit to the real stream
deallocate temp unit and buffer
On input there are similar concerns, but we already use
st_parameter_dt.u.p.line_buffer there.
Or then one could of course use line_buffer in the write functions just like
they are used for reading.
--
Summary: T edit descriptor doesn't work for non-seekable streams
Product: gcc
Version: 4.4.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libfortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jb at gcc dot gnu dot org
GCC host triplet: i686-pc-linux-gnu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36142
