1 06/20/80: Known bugs in the installed >unb>bound_fort_ new_fortran compiler.
2 This compiler is Release 7 of fortran new_fortran.
3
4 265 raising a double precision number to a very small
5 positive power can result in exponent overflow
6 actually in pl1_operators_.
7 264 the optimizer does not convert constants which are
8 assigned directly to variables to the data type of the
9 variable on the left hand side of the assignment
10 related to bug 262.
11 263 the optimizer does not protect itself from errors such
12 as underflow, overflow, and zerodivide while performing
13 constant folding.
14 262 the optimizer does not pad character constants which
15 are directly assigned to variables of an arithmetic
16 type.
17
18
19 261 the optimizer may common expressions that are not in
20 fact common within loops that are entered only in the
21 middle.
22 259 the compiler truncates the source segment and sets the
23 bit count randomly when the source and object segments
24 are the same. This is related to bug 205.
25 257 referencing a multi-dimensional array as a vector in an
26 equivalence statement does not work if the vector
27 subscript is larger than the first dimension of the
28 array. The 1966 ANSI standard explicitly allows
29 multi-dimensional arrays to be referenced as vectors
30 within equivalence statements; the 1977 ANSI standard
31 explicitly prohibits it.
32 246 the compiler may not detect that too many initial
33 values have been supplied in a data statement if the
34 last item in the data statement is a repeated value
35 whose repetition count was larger than needed.
36
37
38 244 the real sin intrinsic function gives a notably less
39 accurate result for arguments in the range from 1/256
40 to about 1/6000 actually in pl1_operators_.
41 243 the real asin intrinsic function may erroneously try to
42 take the square root of a negative number actually in
43 pl1_operators_. The dasin, acos, and dacos
44 intrinsics fail similarly.
45 238 optimizing code generator does not remove a variable
46 from the machine state when its alias is set. This can
47 cause problems if an equivalenced variable is used as
48 the only subscript of a subscripted reference.
49 205 the compiler should not truncate an existing object
50 segment until it has determined that it is going to try
51 to make a new one, i.e. that the code generator will be
52 invoked.
53
54
55 178 the -table option may not produce the desired results
56 if the -optimize option is also used.
57 164 error messages may not include a line number.
58
59