On Sat, Jul 5, 2014 at 11:57 AM, Robert P. Goldman rpgoldman@sift.info wrote:
In the process of making a test for bug1335323, I discovered that the top-level call to look up the system in LOAD-SYSTEM/OPERATE, is FIND-COMPONENT, not FIND-SYSTEM, so that the REINITIALIZE-SOURCE-REGISTRY-AND-RETRY restart is not available.
Well, FIND-COMPONENT calls FIND-SYSTEM, and if the failure happens there, the restart will be available, won't it?
Also, for some reason, I am unable to find the "force1" subsystem of "test-asdf", although "test-asdf" is found correctly:
Works for me, from a fresh sbcl having loaded asdf.
1: (OPERATE LOAD-OP "test-asdf/force1") 2: (FIND-COMPONENT NIL "test-asdf/force1") 3: (FIND-COMPONENT "test-asdf/force1" NIL) 4: (FIND-SYSTEM "test-asdf/force1" NIL) 5: (FIND-SYSTEM "test-asdf" NIL) 5: FIND-SYSTEM returned #<PACKAGE-INFERRED-SYSTEM "test-asdf"> 5: (FIND-SYSTEM "test-asdf" NIL) 5: FIND-SYSTEM returned #<PACKAGE-INFERRED-SYSTEM "test-asdf"> 4: FIND-SYSTEM returned NIL 3: FIND-COMPONENT returned NIL 2: FIND-COMPONENT returned NIL
Any idea why this is happening? It's after an attempt to clear the ASDF cache, and I wonder if that's implicated -- maybe just smashing the *asdf-cache* isn't correct. When I first call FIND-SYSTEM, I do find the system:
I don't know how and when you "smash the cache", but test-asdf/force1 is defined at the end of test-asdf.asd. Therefore, if you interrupt the loading of this file after test-asdf itself was defined, you can reach a point when test-asdf is defined but not test-asdf/force1. Or maybe you somehow wiped part but not all of the *registered-systems* ?
; Registering #<SYSTEM "test-asdf/force1"> Clearing the caches and finding.....
I have pushed my bug1335323 topic branch to cl.net, since I could use some help with it, in case anyone has the time to examine it. (whoops. I can see now that I should be keeping the cache entries for ASDF, POIU, and their components; that's probably a bug, since if you are changing ASDF you're probably better off unwinding completely and starting over, instead of trying to handle an error)
Brief look: * your clear-asdf-cache should probably use clrhash to preserve identity. * it's probably unsafe to throw the entire cache in the middle of the build :-/ maybe somehow the restart should be part of the call-with-asdf-cache, not of find-system, and should have a different name than the current one. Also, once again, I suggest two restarts: one that just restarts and lets you choose what to blow up or not, and one that blows up everything that makes sense. Maybe the two restarts are at a different place. Maybe the restart that blows up the cache is actually in call-with-asdf-cache, which is safer.
—♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org Once upon a time, there was a person whose life was so good there was no story to tell about it. — Foer