09-271
To: J3
From: Robert Corbett
Subject: message about floating-point exceptions
Date: 2009 July 27
References: 09-007r2

1.  Introduction
--  ------------

Users hate it when programs issue warnings about floating-point
exceptions on termination.

2.  Discussion
--  ----------

Sun f77 used to issue a retrospective message about
floating-point exceptions on termination.  The message was a
constant source of complaints from users.  Sun f77 did not
issue a restrospective message if the only floating-point
exception that was signaling was the inexact exception.
Telling a user that his program signaled an inexact
exception is tantamount to telling telling him that his
Fortran program has done some floating-point arithmetic.

A former member of this committee was among the most vocal
critics of the retrospective message.

I know that the IEEE floating-point standard requires such a
message to be issued.  Users do not care.  The Fortran
standard does not conform to the IEEE standard in other ways,
such as not changing the values of floating-point constants
when the rounding mode changes.

3.  Edits
--  -----

[187:13-16 8.4p2]
Delete the second sentence of paragraph 2.  In the third
sentence, replace "same unit." with "unit identified by the
named constant ERROR_UNIT (13.8.2.8).

[187:16+ Note 8.27]
Delete the phrase "and signaling exceptions".