#44 Recursive file load loop if asdf file is in the future

[Liam Healy]

If the .asd file has a timestamp in the future, due
e.g. to clock disparity between file server and
desktop, ASDF gets in an infinite recursion of file
loads on this file; see sample at
http://paste.lisp.org/display/25828. Perhaps clock
skew could be detected and reported, as make does?

[Kambiz Darabi]

During the first call to find-system, the asd file is loaded which leads to defsystem setting the car of the hash table entry in *defined-systems* to the current time:

(setf (car s) (get-universal-time)))

later, during the same call, find-system is called again which compares the car of the *defined-systems* entry to the file-write-date:

(< (car in-memory) (file-write-date on-disk))))

This test fails, if the file-write-date is in the future.

One way to fix this problem might be to keep the file-write-date of the system for which find-system is being called in a var and use that file-write-date instead of get-universal-time.

Here is the (probably horrible) patch just to illustrate the issue:

Index: asdf.lisp

RCS file: /cvsroot/cclan/asdf/asdf.lisp,v
retrieving revision 1.128
diff -r1.128 asdf.lisp
369a370,374
> (defvar *found-system* (list nil 0)
> "A two element list which holds the coerce-name and file-write-date
> of the current system during a find-system call to prevent recursive
> calls to find-system")
>
420c425,426
< (let ((*package* package))
---
> (let ((*package* package)
> (*found-system* (list name (file-write-date on-disk))))
429c435,436
< (delete-package package))))
---
> (delete-package package)
> (setf *found-system* (list nil 0)))))
436c443
< (defun register-system (name system)
---
> (defun asdf::register-system (name system)
438,439c445,452
< (setf (gethash (coerce-name name) *defined-systems*)
< (cons (get-universal-time) system)))
---
> (let ((name (coerce-name name)))
> (setf (gethash name *defined-systems*)
> (cons (on-disk-or-universal-time name) system))))
>
> (defun on-disk-or-universal-time (name)
> (if (equal (coerce-name name) (car *found-system*))
> (second *found-system*)
> (get-universal-time)))
1106c1119
< (setf (car s) (get-universal-time)))
---
> (setf (car s) (on-disk-or-universal-time ',name)))

[Kambiz Darabi]

Oops. I wrote

This test fails, if the file-write-date is in the future.

The correct sentence would have been:

This test always succeeds, if the file-write-date is in the future, so (load on-disk) is attempted again.