#3099 Float read is single-precision, though float() is double-precision

[Stavros Macrakis]

In Maxima on SBCL 1.1.8, when there is no explicit exponent in floating-point input, it is read as a single float. (Is this an SBCL bug? I don't see it in Maxima 5.36 on SBCL 1.2.10.)

foo: [1.0, 1.0e0, 1.0d0, float(1)]$
:lisp (mapcar 'type-of (cdr $foo))
=>
(SINGLE-FLOAT DOUBLE-FLOAT DOUBLE-FLOAT DOUBLE-FLOAT)

Maxima 5.37post (built 2016-02-22 19:29:21)
SBCL 1.1.8.0-19cda10 on MacOS 10.11.3

[Stavros Macrakis]

Re-tested on Maxima 5.37post, SBCL 1.2.11, same result.
SBCL 1.2.11 is the most up-to-date binary available for OSX/AMD64 at http://www.sbcl.org/platform-table.html . Only Windows and Linux have more recent binaries.

[Robert Dodier]

It appears that *READ-DEFAULT-FLOAT-FORMAT* needs to be set to DOUBLE-FLOAT for SBCL. (Apparently that's the default for some installations of SBCL, but at some, including yours it appears, have SINGLE-FLOAT.)

See INITIALIZE-RUNTIME-GLOBALS in src/init-cl.lisp. There is already some implementation-specific code there for the float format; I guess we need something for SBCL too.

[Robert Dodier]

To verify -- what does :lisp *READ-DEFAULT-FLOAT-FORMAT* report on your system?

[Stavros Macrakis]

READ-DEFAULT-FLOAT-FORMAT => single-float
setting it to double-float fixes the problem

[Robert Dodier]

Fixed by commit 0443536. Closing this report accordingly.