FMAKUNBOUND a traced function makes UNTRACE signal an error

[Jānis Džeriņš]

Removing the global function definition from a traced symbol (function) makes
UNTRACE with no arguments (i.e., the intent being of untracing everything
that's being traced) to signal an UNDEFINED-FUNCTION error.

As it is now (SBCL 1.0.43.74), it is impossible to remove the funbound symbol
from the trace list.

If the same function is redefined and traced again, the same symbol is
inserted into the list of traced functions; calling UNTRACE with no arguments
now will warn about function being not traced. (What is a function that is
not being traced doing in the trace list, anyway?)

A sketch of the test case is attached.

[Jānis Džeriņš (jonis)]

Sorry for the misformatting, and I misunderstood the meaning of "this file is a patch", so, the attached file is a test case, not a patch.

[Attila Lendvai (attila-lendvai)]

i have an advertised fix for that since ages available at the link below and now attached here also.

http://dwim.hu/gitweb/gitweb.cgi?p=sbcl;a=commitdiff;h=3d097a76dd2cfbbd3f4ebfed5a45f9cd07e6d3e7

[Jānis Džeriņš (jonis)]

With the current fix calling UNTRACE with no arguments does not signal an error, but I would still expect the trace list be empty after the call. The spec says: "... If untrace is used with no function-names, all functions currently being traced are untraced." (http://www.lispworks.com/documentation/HyperSpec/Body/m_tracec.htm)

[Nikodemus Siivola (nikodemus)]

Yep.

CL-USER> (defun foo () t)
FOO
CL-USER> (trace foo)
(FOO)
CL-USER> (fmakunbound 'foo)
FOO
CL-USER> (untrace)
T
CL-USER> (trace)
(FOO)

[Nikodemus Siivola (nikodemus)]

In 1.0.44.5.