Minor macroexpand issue
Frederic Koehler
On clojure 1.2, macroexpanding with a function name which is a class,
causes this ugly error:
> (macroexpand '(Object))
java.lang.Exception: Expecting var, but Object is mapped to class
java.lang.Object (repl-1:2)
when presumably it should just give '(Object).
I have no clue for any actual use cases of naming your functions after
classes, but:
(let [Object (fn [] 3)] (Object))
is technically valid clojure code, so macroexpand shouldn't just die...
Konrad Hinsen
You are right that
(let [Object (fn [] 3)] (Object))
is valid Clojure code, and it works as expected. However,
(Object)
is not a valid expression. It fails with exactly the same error
message as
(macroexpand '(Object))
so I'd say this is consistent behaviour.
Konrad.
Frederic Koehler
not defined in the context, so I'd say it's a little unexpected.
More annoying is the result this has on macroexpand-all, which breaks
because of this weird behaviour:
(use 'clojure.walk)
(walk/macroexpand-all '(let [Object (fn [])] (Object)))
java.lang.RuntimeException: java.lang.Exception: Expecting var, but
Object is mapped to class java.lang.Object (NO_SOURCE_FILE:0)
Of course, this example is somewhat (not super-) obvious, but in more
complex cases it could be fairly annoying.
> --
> You received this message because you are subscribed to the Google
> Groups "Clojure" group.
> To post to this group, send email to clo...@googlegroups.com
> Note that posts from new members are moderated - please be patient with your first post.
> To unsubscribe from this group, send email to
> clojure+u...@googlegroups.com
> For more options, visit this group at
> http://groups.google.com/group/clojure?hl=en
Armando Blancas
doesn't have to be valid code. Since '(Object) is not a macro (Object
is a Symbol there, not a class) it should just return the form just
like (identity) does:
user=> (identity '(Object))
(Object)
But macroexpand-1 does some extra work for the interop sugar. I
suspect that the error comes when it tries to work out a static
member.
user=> (macroexpand-1 '(.length "s"))
(. "s" length)
user=> (macroexpand-1 '(Boolean/TRUE))
(. Boolean TRUE)
user=> (macroexpand-1 '(String.))
(new String)
>
> - Show quoted text -
Konrad Hinsen
> If you were to say, (macroexpand '(foo)), it returns (foo), even if
> it's
> not defined in the context, so I'd say it's a little unexpected.
True, but then, the function macroexpand is mostly a development tool,
for which I can accept such minor issues. Macroexpansion itself is
part of the evaluation procedure of forms, so it makes no sense to
define what macroexpansion should return for a form that is not legal
code anyway. Neither (Object) nor (foo) are legal code.
> More annoying is the result this has on macroexpand-all, which breaks
> because of this weird behaviour:
>
> (use 'clojure.walk)
> (walk/macroexpand-all '(let [Object (fn [])] (Object)))
>
> java.lang.RuntimeException: java.lang.Exception: Expecting var, but
> Object is mapped to class java.lang.Object (NO_SOURCE_FILE:0)
Right, but then this should be blamed on macroexpand-all, which
implements a crude approximation to how macroexpansion is done by the
compiler.
There's also clojure.contrib.macro-utils/mexpand-all, which does a
much better job, but still fails for this particular example. In fact,
neither one tracks the lexical environment.
Konrad.
Konrad Hinsen
> Right, but then this should be blamed on macroexpand-all, which
> implements a crude approximation to how macroexpansion is done by
> the compiler.
>
> There's also clojure.contrib.macro-utils/mexpand-all, which does a
> much better job, but still fails for this particular example. In
> fact, neither one tracks the lexical environment.
clojure.contrib.macro-utils/mexpand-all now does track the lexical
environment. In fact, it already did before, but it used that
information only for symbol macro expansion.
Konrad.