Undefined symbol COERCE-NAME in footer.lisp

[Christian Schafmeister]

See summary. I found that COERCE-NAME is called when the default package is the ASDF/FOOTER package but it's defined in the ASDF/FIND-SYSTEM package and not imported into ASDF/FOOTER

[Faré (fahree)]

Wow. This bug and crept in the 3.1 development in #+ecl support only, that requires , and no one noticed until now! Sorry. ECL isn't as well maintained anymore :-(

Neighbor bug also in #+ecl: when I renamed to register-pre-built-system to register-preloaded-system, this part of the code was never updated.

Robert: are you on it? "Just" add back find-system as dependency to the footer, and rename register-foo... Or I can do it, if you prefer.

[Faré (fahree)]

Does this fix the issue? Does ASDF now run on Clasp? Can it pass the tests?

make t l=clasp
(after suitably editing the test tools.)

[Robert P. Goldman (rpgoldman)]

Faré wrote:
> Wow. This bug and crept in the 3.1 development in #+ecl support only,
> that requires , and no one noticed until now! Sorry. ECL isn't as well
> maintained anymore :-(
>
> Neighbor bug also in #+ecl: when I renamed to register-pre-built-system
> to register-preloaded-system, this part of the code was never updated.
>
> Robert: are you on it? "Just" add back find-system as dependency to the
> footer, and rename register-foo... Or I can do it, if you prefer.
>

Sure, I'll get it fixed.

Question: how is this to be tested? I keep running all the tests in our
suite: why don't we see this? Is it because it only comes into play when
ASDF is incorporated into the ECL distribution?

[Robert P. Goldman (rpgoldman)]

Also, what is clasp? The only clasp I know of is a solver for answer set programming....

[Faré (fahree)]

clasp is a new CL implementation based on ECL, with substantial parts translated from C to C++, and using LLVM as a backend.

I'm not sure how to test the bug and its fix, but it looks like it would happen if you try to require a contrib e.g. (require 'defsystem) after you loaded asdf. Now, the require mechanism might be hiding the failure in a handler-foo.

[Robert P. Goldman (rpgoldman)]

I'm not finding CLASP through google. "clasp common lisp implementation" finds other CL libraries with that name, but no ECL fork.

Christian, how did you trigger this bug? I think it's fixed, but I don't know how to confirm its existence and check that it's gone away.

[Robert P. Goldman (rpgoldman)]

I'll push the fix when additional information has been supplied:

How was the bug triggered? And how can I replicate it?

[Christian Schafmeister (chris-schaf)]

I COMPILE-FILEd asdf.lisp and loaded the fasl into Clasp and then REQUIREd something else.

Clasp isn’t available yet. I’m waiting for IP lawyers to give the the green light.

Best,

.Chris.

On Aug 29, 2014, at 2:14 PM, Robert P. Goldman <email address hidden> wrote:

> I'll push the fix when additional information has been supplied:
>
> How was the bug triggered? And how can I replicate it?
>
> ** Changed in: asdf
> Status: Confirmed => Incomplete
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1362875
>
> Title:
> Undefined symbol COERCE-NAME in footer.lisp
>
> Status in Another System Definition Facility:
> Incomplete
>
> Bug description:
> See summary. I found that COERCE-NAME is called when the default
> package is the ASDF/FOOTER package but it's defined in the ASDF/FIND-
> SYSTEM package and not imported into ASDF/FOOTER
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/asdf/+bug/1362875/+subscriptions

[Robert P. Goldman (rpgoldman)]

Fare suggests that this bug should be visible in ECL, but I can't replicate it. Or rather, I can't make ECL require anything. But then, I'm not an ECL user, and the ECL documentation does not list any module that I should be able to require. All my attempts to require anything in ECL simply crash, both before and after Fare's patch, with cryptic and garbled error messages.

[Robert P. Goldman (rpgoldman)]

This does not seem to be the same bug -- or is it?

> (require :defsystem)
ASDF could not load defsystem because Error while trying to load definition for system defsystem from pathname SYS:DEFSYSTEM.ASD.NEWEST: don't recognize component type NIL.

Condition of type: LOAD-SYSTEM-DEFINITION-ERROR
Error while trying to load definition for system defsystem from pathname SYS:DEFSYSTEM.ASD.NEWEST: ...

Note the garbling because of what looks like nesting of error messages.

There's no package error here, AFAICT, so this looks like a different bug to me.

[Faré (fahree)]

I investigated a bit, and found that ECL provides a .asd for each of its builtin modules, such that after you load asdf, the first function in ext:*module-provider-functions*, namely ASDF/OPERATE:MODULE-PROVIDE-ASDF, will find the other modules. But if you (pop ext:*module-provider-functions*) this function off and try to e.g. (require :sockets), you'll hit the bug:

Condition of type: UNDEFINED-FUNCTION
The function ASDF/FOOTER::COERCE-NAME is undefined.

Therefore the bug is present on ECL, but hidden by a feature in the common case. It still needs to be fixed.

[Robert P. Goldman (rpgoldman)]

OK, I'll check the patched version against this test case and push it.

Thanks.

[Robert P. Goldman (rpgoldman)]

Actually, I do not confirm this behavior on ECL 13.5.1 on Mac OS X. Per Christian's instructions, I loaded the fresh ASDF and then I tried to do what you said:

> (load "build/asdf.fas")

#P"/Users/rpg/lisp/asdf/build/asdf.fas"
> (pop ext:*module-provider-functions*)

ASDF/OPERATE:MODULE-PROVIDE-ASDF
> (require :sockets)

("SB-BSD-SOCKETS" "SOCKETS")
> ^D

I agree that there's a bug here and that it should be fixed, but I'm disturbed that I cannot make it appear. What version of ECL are you using?

[Faré (fahree)]

I just compiled the latest ECL from git master:

ecl -eval '(require :asdf)' -eval '(progn (uiop:writeln (asdf:asdf-version)) (pop ext:*module-provider-functions*) (require :sockets))'
;;; Loading #P"/home/tunes/local/stow/ecl/lib/ecl-13.5.1/asdf.fas"
"3.1.2"
An error occurred during initialization:
The function ASDF/FOOTER::COERCE-NAME is undefined..

[Faré (fahree)]

I introduced the bug some time between 3.0.3 and 3.1.2:

ecl -eval '(load "build/asdf-3.0.3.lisp")' -eval '(progn (format t "~S~%" (asdf:asdf-version)) (pop ext:*module-provider-functions*) (require :sockets) (uiop:quit))'
;;; Loading "/home/tunes/cl/asdf/build/asdf-3.0.3.lisp"
"3.0.3"

ecl -eval '(load "build/asdf-3.1.2.lisp")' -eval '(progn (format t "~S~%" (asdf:asdf-version)) (pop ext:*module-provider-functions*) (require :sockets) (uiop:quit))'
;;; Loading "/home/tunes/cl/asdf/build/asdf-3.1.2.lisp"
"3.1.2"
An error occurred during initialization:
The function ASDF/FOOTER::COERCE-NAME is undefined..

With my patch, it works:
ecl -eval '(load "build/asdf.lisp")' -eval '(progn (format t "~S~%" (asdf:asdf-version)) (pop ext:*module-provider-functions*) (require :sockets) (uiop:quit))'
;;; Loading "/home/tunes/cl/asdf/build/asdf.lisp"
"3.1.3.5"

...but it doesn't fix a ext:*module-provider-functions* broken by 3.1.2 or 3.1.3:
ecl -eval '(load "build/asdf-3.1.2.lisp")' -eval '(progn (asdf:upgrade-asdf) (format t "~S~%" (asdf:asdf-version)) (pop ext:*module-provider-functions*) (require :sockets) (uiop:quit))'
;;; Loading "/home/tunes/cl/asdf/build/asdf-3.1.2.lisp"
"3.1.3.5"
An error occurred during initialization:
The function #:COERCE-NAME is undefined..